fix: vision(#623 ): Claude identity isolation for disinto-chat (#707 )

2026-04-12 02:22:07 +00:00
83 changed files with 1157 additions and 6095 deletions
--- a/.dockerignore
+++ b/.dockerignore
@ -1,7 +1,8 @@
-# Secrets — prevent .env files and encrypted secrets from being baked into the image
+# Secrets — prevent .env files from being baked into the image
 .env
 .env.enc
-secrets/
+.env.vault
 .env.vault.enc
 # Version control — .git is huge and not needed in image
 .git
--- a/.env.example
+++ b/.env.example
@ -45,9 +45,7 @@ FORGE_PREDICTOR_TOKEN=                     # [SECRET] predictor-bot API token
 FORGE_PREDICTOR_PASS=                      # [SECRET] predictor-bot password for git HTTP push
 FORGE_ARCHITECT_TOKEN=                     # [SECRET] architect-bot API token
 FORGE_ARCHITECT_PASS=                      # [SECRET] architect-bot password for git HTTP push
-FORGE_FILER_TOKEN=                         # [SECRET] filer-bot API token (issues:write on project repo only)
+FORGE_BOT_USERNAMES=dev-bot,review-bot,planner-bot,gardener-bot,vault-bot,supervisor-bot,predictor-bot,architect-bot
 FORGE_FILER_PASS=                          # [SECRET] filer-bot password for git HTTP push
 FORGE_BOT_USERNAMES=dev-bot,review-bot,planner-bot,gardener-bot,vault-bot,supervisor-bot,predictor-bot,architect-bot,filer-bot
 # ── Backwards compatibility ───────────────────────────────────────────────
 # If CODEBERG_TOKEN is set but FORGE_TOKEN is not, env.sh falls back to
@ -63,10 +61,6 @@ FORGE_BOT_USERNAMES=dev-bot,review-bot,planner-bot,gardener-bot,vault-bot,superv
 WOODPECKER_TOKEN=                          # [SECRET] Woodpecker API token
 WOODPECKER_SERVER=http://localhost:8000     # [CONFIG] Woodpecker server URL
 WOODPECKER_AGENT_SECRET=                   # [SECRET] shared secret for server↔agent auth (auto-generated)
 # Woodpecker privileged-plugin allowlist — comma-separated image names
 # Add plugins/docker (and others) here to allow privileged execution
 WOODPECKER_PLUGINS_PRIVILEGED=plugins/docker
 # WOODPECKER_REPO_ID — now per-project, set in projects/*.toml [ci] section
 # Woodpecker Postgres (for direct DB queries)
@ -79,46 +73,27 @@ WOODPECKER_DB_NAME=woodpecker              # [CONFIG] Postgres database name
 CHAT_OAUTH_CLIENT_ID=                     # [SECRET] Chat OAuth2 client ID (auto-generated by init)
 CHAT_OAUTH_CLIENT_SECRET=                 # [SECRET] Chat OAuth2 client secret (auto-generated by init)
 DISINTO_CHAT_ALLOWED_USERS=               # [CONFIG] CSV of allowed usernames (disinto-admin always allowed)
 FORWARD_AUTH_SECRET=                      # [SECRET] Shared secret for Caddy ↔ chat forward_auth (#709)
 # ── Vault-only secrets (DO NOT put these in .env) ────────────────────────
 # These tokens grant access to external systems (GitHub, ClawHub, deploy targets).
-# They live ONLY in secrets/<NAME>.enc (age-encrypted, one file per key) and are
+# They live ONLY in .env.vault.enc and are injected into the ephemeral runner
-# decrypted into the ephemeral runner container at fire time (#745, #777).
+# container at fire time (#745). lib/env.sh explicitly unsets them so agents
-# lib/env.sh explicitly unsets them so agents can never hold them directly —
+# can never hold them directly — all external actions go through vault dispatch.
 # all external actions go through vault dispatch.
 #
 #   GITHUB_TOKEN          — GitHub API access (publish, deploy, post)
 #   CLAWHUB_TOKEN         — ClawHub registry credentials (publish)
 #   CADDY_SSH_KEY         — SSH key for Caddy log collection
 #   (deploy keys)         — SSH keys for deployment targets
 #
-# To manage secrets: disinto secrets add/show/remove/list
+# To manage vault secrets: disinto secrets edit-vault
 # (vault redesign in progress: PR-based approval, see #73-#77)
 # ── Project-specific secrets ──────────────────────────────────────────────
 # Store all project secrets here so formulas reference env vars, never hardcode.
 BASE_RPC_URL=                              # [SECRET] on-chain RPC endpoint
 # ── Local Qwen dev agent (optional) ──────────────────────────────────────
 # Set ENABLE_LLAMA_AGENT=1 to emit agents-llama in docker-compose.yml.
 # Requires a running llama-server reachable at ANTHROPIC_BASE_URL.
 # See docs/agents-llama.md for details.
 ENABLE_LLAMA_AGENT=0                       # [CONFIG] 1 = enable agents-llama service
 ANTHROPIC_BASE_URL=                        # [CONFIG] e.g. http://host.docker.internal:8081
 # ── Tuning ────────────────────────────────────────────────────────────────
 CLAUDE_TIMEOUT=7200                        # [CONFIG] max seconds per Claude invocation
 # ── Host paths (Nomad-portable) ────────────────────────────────────────────
 # These env vars externalize host-side bind-mount paths from docker-compose.yml.
 # At cutover, Nomad jobspecs reference the same vars — no path translation.
 # Defaults point at current paths so an empty .env override still works.
 CLAUDE_BIN_DIR=/usr/local/bin/claude          # [CONFIG] host path to claude CLI binary (resolved by `disinto init`)
 CLAUDE_CONFIG_FILE=${HOME}/.claude.json       # [CONFIG] host path to claude config JSON file
 CLAUDE_DIR=${HOME}/.claude                    # [CONFIG] host path to .claude directory (reproduce/edge)
 AGENT_SSH_DIR=${HOME}/.ssh                    # [CONFIG] host path to SSH keys directory
 SOPS_AGE_DIR=${HOME}/.config/sops/age         # [CONFIG] host path to SOPS age key directory
 # ── Claude Code shared OAuth state ─────────────────────────────────────────
 # Shared directory used by every factory container so Claude Code's internal
 # proper-lockfile-based OAuth refresh lock works across containers. Both
--- a/.gitignore
+++ b/.gitignore
@ -3,6 +3,7 @@
 # Encrypted secrets — safe to commit (SOPS-encrypted with age)
 !.env.enc
 !.env.vault.enc
 !.sops.yaml
 # Per-box project config (generated by disinto init)
@ -32,9 +33,6 @@ docker/agents/bin/
 # Note: This file is now committed to track volume mount configuration
 # docker-compose.yml
 # Generated Caddyfile — single source of truth is generate_caddyfile in lib/generators.sh
 docker/Caddyfile
 # Python bytecode
 __pycache__/
 *.pyc
--- a/.woodpecker/agent-smoke.sh
+++ b/.woodpecker/agent-smoke.sh
@ -98,38 +98,50 @@ echo "syntax check done"
 echo "=== 2/2  Function resolution ==="
-# Enumerate ALL lib/*.sh files in stable lexicographic order (#742).
+# Required lib files for LIB_FUNS construction. Missing any of these means the
-# Previous approach used a hand-maintained REQUIRED_LIBS list, which silently
+# checkout is incomplete or the test is misconfigured — fail loudly, do NOT
-# became incomplete as new libs were added, producing partial LIB_FUNS that
+# silently produce a partial LIB_FUNS list (that masquerades as "undef" errors
-# caused non-deterministic "undef" failures.
+# in unrelated scripts; see #600).
 REQUIRED_LIBS=(
  lib/agent-sdk.sh lib/env.sh lib/ci-helpers.sh lib/load-project.sh
  lib/secret-scan.sh lib/formula-session.sh lib/mirrors.sh lib/guard.sh
  lib/pr-lifecycle.sh lib/issue-lifecycle.sh lib/worktree.sh
 )
 for f in "${REQUIRED_LIBS[@]}"; do
  if [ ! -f "$f" ]; then
    printf 'FAIL [missing-lib] expected %s but it is not present at smoke time\n' "$f" >&2
    printf '  pwd=%s\n' "$(pwd)" >&2
    printf '  ls lib/=%s\n' "$(ls lib/ 2>&1 | tr '\n' ' ')" >&2
    echo '=== SMOKE TEST FAILED (precondition) ===' >&2
    exit 2
  fi
 done
 # Functions provided by shared lib files (available to all agent scripts via source).
 #
-# Excluded from LIB_FUNS (not sourced inline by agents):
+# Included — these are inline-sourced by agent scripts:
 #   lib/env.sh              — sourced by every agent (log, forge_api, etc.)
 #   lib/agent-sdk.sh        — sourced by SDK agents (agent_run, agent_recover_session)
 #   lib/ci-helpers.sh       — sourced by pollers and review (ci_passed, classify_pipeline_failure, etc.)
 #   lib/load-project.sh     — sourced by env.sh when PROJECT_TOML is set
 #   lib/secret-scan.sh      — standalone CLI tool, run directly (not sourced)
 #   lib/formula-session.sh  — sourced by formula-driven agents (acquire_run_lock, check_memory, etc.)
 #   lib/mirrors.sh          — sourced by merge sites (mirror_push)
 #   lib/guard.sh            — sourced by all polling-loop entry points (check_active)
 #   lib/issue-lifecycle.sh  — sourced by agents for issue claim/release/block/deps
 #   lib/worktree.sh         — sourced by agents for worktree create/recover/cleanup/preserve
 #
 # Excluded — not sourced inline by agents:
 #   lib/tea-helpers.sh      — sourced conditionally by env.sh (tea_file_issue, etc.); checked standalone below
 #   lib/ci-debug.sh         — standalone CLI tool, run directly (not sourced)
 #   lib/parse-deps.sh       — executed via `bash lib/parse-deps.sh` (not sourced)
 #   lib/hooks/*.sh          — Claude Code hook scripts, executed by the harness (not sourced)
-EXCLUDED_LIBS="lib/ci-debug.sh lib/parse-deps.sh"
+#
-
+# If a new lib file is added and sourced by agents, add it to LIB_FUNS below
-# Build the list of lib files in deterministic order (LC_ALL=C sort).
+# and add a check_script call for it in the lib files section further down.
 # Fail loudly if no lib files are found — checkout is broken.
 mapfile -t ALL_LIBS < <(LC_ALL=C find lib -maxdepth 1 -name '*.sh' -print | LC_ALL=C sort)
 if [ "${#ALL_LIBS[@]}" -eq 0 ]; then
  echo 'FAIL [no-libs] no lib/*.sh files found at smoke time' >&2
  printf '  pwd=%s\n' "$(pwd)" >&2
  echo '=== SMOKE TEST FAILED (precondition) ===' >&2
  exit 2
 fi
 # Build LIB_FUNS from all non-excluded lib files.
 # Use set -e inside the subshell so a failed get_fns aborts loudly
 # instead of silently shrinking the function list.
 LIB_FUNS=$(
-  set -e
+  for f in "${REQUIRED_LIBS[@]}"; do get_fns "$f"; done | sort -u
  for f in "${ALL_LIBS[@]}"; do
    # shellcheck disable=SC2086
    skip=0; for ex in $EXCLUDED_LIBS; do [ "$f" = "$ex" ] && skip=1; done
    [ "$skip" -eq 1 ] && continue
    get_fns "$f"
  done | sort -u
 )
 # Known external commands and shell builtins — never flag these
@ -180,14 +192,13 @@ check_script() {
  while IFS= read -r fn; do
    [ -z "$fn" ] && continue
    is_known_cmd "$fn" && continue
-    # Use here-string (<<<) instead of pipe to avoid SIGPIPE race (#742):
+    if ! printf '%s\n' "$all_fns" | grep -qxF "$fn"; then
    # with pipefail, `printf | grep -q` can fail when grep closes the pipe
    # early after finding a match, causing printf to get SIGPIPE (exit 141).
    # This produced non-deterministic false "undef" failures.
    if ! grep -qxF "$fn" <<< "$all_fns"; then
      printf 'FAIL [undef] %s: %s\n' "$script" "$fn"
-      printf '  all_fns count: %d\n' "$(grep -c . <<< "$all_fns")"
+      # Diagnostic dump (#600): if the function is expected to be in a known lib,
-      printf '  LIB_FUNS contains "%s": %s\n' "$fn" "$(grep -cxF "$fn" <<< "$LIB_FUNS")"
+      # print what the actual all_fns set looks like so we can tell whether the
      # function is genuinely missing or whether the resolution loop is broken.
      printf '  all_fns count: %d\n' "$(printf '%s\n' "$all_fns" | wc -l)"
      printf '  LIB_FUNS contains "%s": %s\n' "$fn" "$(printf '%s\n' "$LIB_FUNS" | grep -cxF "$fn")"
      printf '  defining lib (if any): %s\n' "$(grep -l "^[[:space:]]*${fn}[[:space:]]*()" lib/*.sh 2>/dev/null | tr '\n' ' ')"
      FAILED=1
    fi
@ -213,7 +224,6 @@ check_script lib/issue-lifecycle.sh   lib/secret-scan.sh
 # Still checked for function resolution against LIB_FUNS + own definitions.
 check_script lib/ci-debug.sh
 check_script lib/parse-deps.sh
 check_script lib/sprint-filer.sh
 # Agent scripts — list cross-sourced files where function scope flows across files.
 check_script dev/dev-agent.sh
--- a/.woodpecker/detect-duplicates.py
+++ b/.woodpecker/detect-duplicates.py
@ -292,8 +292,6 @@ def main() -> int:
        "21aec56a99d5252b23fb9a38b895e8e8": "Verification helper: check body for Decomposed from pattern",
        "60ea98b3604557d539193b2a6624e232": "Verification helper: append sub-issue number",
        "9f6ae8e7811575b964279d8820494eb0": "Verification helper: for loop done pattern",
        # Standard lib source block shared across formula-driven agent run scripts
        "330e5809a00b95ade1a5fce2d749b94b": "Standard lib source block (env.sh, formula-session.sh, worktree.sh, guard.sh, agent-sdk.sh)",
    }
    if not sh_files:
--- a/.woodpecker/publish-images.yml
+++ b/.woodpecker/publish-images.yml
@ -1,64 +0,0 @@
 # .woodpecker/publish-images.yml — Build and push versioned container images
 # Triggered on tag pushes (e.g. v1.2.3). Builds and pushes:
 #   - ghcr.io/disinto/agents:<tag>
 #   - ghcr.io/disinto/reproduce:<tag>
 #   - ghcr.io/disinto/edge:<tag>
 #
 # Requires GHCR_TOKEN secret configured in Woodpecker with push access
 # to ghcr.io/disinto.
 when:
  event: tag
  ref: refs/tags/v*
 clone:
  git:
    image: alpine/git
    commands:
      - AUTH_URL=$(printf '%s' "$CI_REPO_CLONE_URL" | sed "s|://|://token:$FORGE_TOKEN@|")
      - git clone --depth 1 "$AUTH_URL" .
      - git fetch --depth 1 origin "$CI_COMMIT_REF"
      - git checkout FETCH_HEAD
 steps:
  - name: build-and-push-agents
    image: plugins/docker
    settings:
      repo: ghcr.io/disinto/agents
      registry: ghcr.io
      dockerfile: docker/agents/Dockerfile
      context: .
      tags:
        - ${CI_COMMIT_TAG}
        - latest
      username: disinto
      password:
        from_secret: GHCR_TOKEN
  - name: build-and-push-reproduce
    image: plugins/docker
    settings:
      repo: ghcr.io/disinto/reproduce
      registry: ghcr.io
      dockerfile: docker/reproduce/Dockerfile
      context: .
      tags:
        - ${CI_COMMIT_TAG}
        - latest
      username: disinto
      password:
        from_secret: GHCR_TOKEN
  - name: build-and-push-edge
    image: plugins/docker
    settings:
      repo: ghcr.io/disinto/edge
      registry: ghcr.io
      dockerfile: docker/edge/Dockerfile
      context: docker/edge
      tags:
        - ${CI_COMMIT_TAG}
        - latest
      username: disinto
      password:
        from_secret: GHCR_TOKEN
--- a/.woodpecker/run-secret-scan.sh
+++ b/.woodpecker/run-secret-scan.sh
@ -1,68 +0,0 @@
 #!/usr/bin/env bash
 set -euo pipefail
 # run-secret-scan.sh — CI wrapper for lib/secret-scan.sh
 #
 # Scans files changed in this PR for plaintext secrets.
 # Exits non-zero if any secret is detected.
 # shellcheck source=../lib/secret-scan.sh
 source lib/secret-scan.sh
 # Path patterns considered secret-adjacent
 SECRET_PATH_PATTERNS=(
  '\.env'
  'tools/vault-.*\.sh'
  'nomad/'
  'vault/'
  'action-vault/'
  'lib/hvault\.sh'
  'lib/action-vault\.sh'
 )
 # Build a single regex from patterns
 path_regex=$(printf '%s|' "${SECRET_PATH_PATTERNS[@]}")
 path_regex="${path_regex%|}"
 # Get files changed in this PR vs target branch.
 # Note: shallow clone (depth 50) may lack the merge base for very large PRs,
 # causing git diff to fail — || true means the gate skips rather than blocks.
 changed_files=$(git diff --name-only --diff-filter=ACMR "origin/${CI_COMMIT_TARGET_BRANCH}...HEAD" || true)
 if [ -z "$changed_files" ]; then
  echo "secret-scan: no changed files found, skipping"
  exit 0
 fi
 # Filter to secret-adjacent paths only
 target_files=$(printf '%s\n' "$changed_files" | grep -E "$path_regex" || true)
 if [ -z "$target_files" ]; then
  echo "secret-scan: no secret-adjacent files changed, skipping"
  exit 0
 fi
 echo "secret-scan: scanning $(printf '%s\n' "$target_files" | wc -l) file(s):"
 printf '  %s\n' "$target_files"
 failures=0
 while IFS= read -r file; do
  # Skip deleted files / non-existent
  [ -f "$file" ] || continue
  # Skip binary files
  file -b --mime-encoding "$file" 2>/dev/null | grep -q binary && continue
  content=$(cat "$file")
  if ! scan_for_secrets "$content"; then
    echo "FAIL: secret detected in $file"
    failures=$((failures + 1))
  fi
 done <<< "$target_files"
 if [ "$failures" -gt 0 ]; then
  echo ""
  echo "secret-scan: $failures file(s) contain potential secrets — merge blocked"
  echo "If these are false positives, verify patterns in lib/secret-scan.sh"
  exit 1
 fi
 echo "secret-scan: all files clean"
--- a/.woodpecker/secret-scan.yml
+++ b/.woodpecker/secret-scan.yml
@ -1,32 +0,0 @@
 # .woodpecker/secret-scan.yml — Block PRs that leak plaintext secrets
 #
 # Triggers on pull requests touching secret-adjacent paths.
 # Sources lib/secret-scan.sh and scans each changed file's content.
 # Exits non-zero if any potential secret is detected.
 when:
  - event: pull_request
    path:
      - ".env*"
      - "tools/vault-*.sh"
      - "nomad/**/*"
      - "vault/**/*"
      - "action-vault/**/*"
      - "lib/hvault.sh"
      - "lib/action-vault.sh"
 clone:
  git:
    image: alpine/git
    commands:
      - AUTH_URL=$(printf '%s' "$CI_REPO_CLONE_URL" | sed "s|://|://token:$FORGE_TOKEN@|")
      - git clone --depth 50 "$AUTH_URL" .
      - git fetch --depth 50 origin "$CI_COMMIT_REF" "$CI_COMMIT_TARGET_BRANCH"
      - git checkout FETCH_HEAD
 steps:
  - name: secret-scan
    image: alpine:3
    commands:
      - apk add --no-cache bash git grep file
      - bash .woodpecker/run-secret-scan.sh
--- a/AGENTS.md
+++ b/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Disinto — Agent Instructions
 ## What this repo is
@ -31,19 +31,19 @@ disinto/                 (code repo)
 ├── supervisor/    supervisor-run.sh — formula-driven health monitoring (polling-loop executor)
 │                  preflight.sh — pre-flight data collection for supervisor formula
 ├── architect/     architect-run.sh — strategic decomposition of vision into sprints
-├── action-vault/  vault-env.sh — shared env setup (vault redesign in progress, see #73-#77)
+├── vault/         vault-env.sh — shared env setup (vault redesign in progress, see #73-#77)
 │                  SCHEMA.md — vault item schema documentation
 │                  validate.sh — vault item validator
 │                  examples/ — example vault action TOMLs (promote, publish, release, webhook-call)
-├── lib/           env.sh, agent-sdk.sh, ci-helpers.sh, ci-debug.sh, load-project.sh, parse-deps.sh, guard.sh, mirrors.sh, pr-lifecycle.sh, issue-lifecycle.sh, worktree.sh, formula-session.sh, stack-lock.sh, forge-setup.sh, forge-push.sh, ops-setup.sh, ci-setup.sh, generators.sh, hire-agent.sh, release.sh, build-graph.py, branch-protection.sh, secret-scan.sh, tea-helpers.sh, action-vault.sh, ci-log-reader.py, git-creds.sh, sprint-filer.sh, hvault.sh
+├── lib/           env.sh, agent-sdk.sh, ci-helpers.sh, ci-debug.sh, load-project.sh, parse-deps.sh, guard.sh, mirrors.sh, pr-lifecycle.sh, issue-lifecycle.sh, worktree.sh, formula-session.sh, stack-lock.sh, forge-setup.sh, forge-push.sh, ops-setup.sh, ci-setup.sh, generators.sh, hire-agent.sh, release.sh, build-graph.py,
 │                  branch-protection.sh, secret-scan.sh, tea-helpers.sh, vault.sh, ci-log-reader.py, git-creds.sh
 │                  hooks/ — Claude Code session hooks (on-compact-reinject, on-idle-stop, on-phase-change, on-pretooluse-guard, on-session-end, on-stop-failure)
 ├── projects/      *.toml.example — templates; *.toml — local per-box config (gitignored)
 ├── formulas/      Issue templates (TOML specs for multi-step agent tasks)
-├── docker/        Dockerfiles and entrypoints: reproduce, triage, edge dispatcher, chat (server.py, entrypoint-chat.sh, Dockerfile, ui/)
+├── docker/        Dockerfiles and entrypoints: reproduce, triage, edge dispatcher, chat (server.py, Dockerfile, ui/)
 ├── tools/         Operational tools: edge-control/ (register.sh, install.sh, verify-chat-sandbox.sh)
 ├── docs/          Protocol docs (PHASE-PROTOCOL.md, EVIDENCE-ARCHITECTURE.md)
 ├── site/          disinto.ai website content
-├── tests/         Test files (mock-forgejo.py, smoke-init.sh, lib-hvault.bats)
+├── tests/         Test files (mock-forgejo.py, smoke-init.sh)
 ├── templates/     Issue templates
 ├── bin/           The `disinto` CLI script
 ├── disinto-factory/  Setup documentation and skill
@ -86,7 +86,7 @@ Each agent has a `.profile` repository on Forgejo storing `knowledge/lessons-lea
 - All scripts start with `#!/usr/bin/env bash` and `set -euo pipefail`
 - Source shared environment: `source "$(dirname "$0")/../lib/env.sh"`
 - Log to `$LOGFILE` using the `log()` function from env.sh or defined locally
- Never hardcode secrets — agent secrets come from `.env.enc`, vault secrets from `secrets/<NAME>.enc` (age-encrypted, one file per key)
+- Never hardcode secrets — agent secrets come from `.env.enc`, vault secrets from `.env.vault.enc` (or `.env`/`.env.vault` fallback)
 - Never embed secrets in issue bodies, PR descriptions, or comments — use env var references (e.g. `$BASE_RPC_URL`)
 - ShellCheck must pass (CI runs `shellcheck` on all `.sh` files)
 - Avoid duplicate code — shared helpers go in `lib/`
@ -113,13 +113,10 @@ bash dev/phase-test.sh
 | Supervisor | `supervisor/` | Health monitoring | [supervisor/AGENTS.md](supervisor/AGENTS.md) |
 | Planner | `planner/` | Strategic planning | [planner/AGENTS.md](planner/AGENTS.md) |
 | Predictor | `predictor/` | Infrastructure pattern detection | [predictor/AGENTS.md](predictor/AGENTS.md) |
-| Architect | `architect/` | Strategic decomposition (read-only on project repo) | [architect/AGENTS.md](architect/AGENTS.md) |
+| Architect | `architect/` | Strategic decomposition | [architect/AGENTS.md](architect/AGENTS.md) |
 | Filer | `lib/sprint-filer.sh` | Sub-issue filing from merged sprint PRs | ops repo pipeline (deferred, see #779) |
 | Reproduce | `docker/reproduce/` | Bug reproduction using Playwright MCP | `formulas/reproduce.toml` |
 | Triage | `docker/reproduce/` | Deep root cause analysis | `formulas/triage.toml` |
 | Edge dispatcher | `docker/edge/` | Polls ops repo for vault actions, executes via Claude sessions | `docker/edge/dispatcher.sh` |
 | agents-llama | `docker/agents/` (same image) | Local-Qwen dev agent (`AGENT_ROLES=dev`), gated on `ENABLE_LLAMA_AGENT=1` | [docs/agents-llama.md](docs/agents-llama.md) |
 | agents-llama-all | `docker/agents/` (same image) | Local-Qwen all-roles agent (all 7 roles), profile `agents-llama-all` | [docs/agents-llama.md](docs/agents-llama.md) |
 > **Vault:** Being redesigned as a PR-based approval workflow (issues #73-#77).
 > See [docs/VAULT.md](docs/VAULT.md) for the vault PR workflow details.
@ -138,7 +135,7 @@ Issues flow: `backlog` → `in-progress` → PR → CI → review → merge →
 |---|---|---|
 | `backlog` | Issue is queued for implementation. Dev-poll picks the first ready one. | Planner, gardener, humans |
 | `priority` | Queue tier above plain backlog. Issues with both `priority` and `backlog` are picked before plain `backlog` issues. FIFO within each tier. | Planner, humans |
-| `in-progress` | Dev-agent is actively working on this issue. Only one issue per project is in-progress at a time. Also set on vision issues by filer-bot when sub-issues are filed (#764). | dev-agent.sh (claims issue), filer-bot (vision issues) |
+| `in-progress` | Dev-agent is actively working on this issue. Only one issue per project is in-progress at a time. | dev-agent.sh (claims issue) |
 | `blocked` | Issue is stuck — agent session failed, crashed, timed out, or CI exhausted. Diagnostic comment on the issue has details. Also used for unmet dependencies. | dev-agent.sh, dev-poll.sh (on failure) |
 | `tech-debt` | Pre-existing issue flagged by AI reviewer, not introduced by a PR. | review-pr.sh (auto-created follow-ups) |
 | `underspecified` | Dev-agent refused the issue as too large or vague. | dev-poll.sh (on preflight `too_large`), dev-agent.sh (on mid-run `too_large` refusal) |
@ -180,8 +177,8 @@ Humans write these. Agents read and enforce them.
 | AD-002 | **Concurrency is bounded per LLM backend, not per project.** One concurrent Claude session per OAuth credential pool; one concurrent session per llama-server instance. Containers with disjoint backends may run in parallel. | The single-thread invariant is about *backends*, not pipelines. **(a) Anthropic OAuth credentials race on token refresh** — each container uses a per-session `CLAUDE_CONFIG_DIR`, so Claude Code's native lockfile-based OAuth refresh handles contention automatically without external serialization. (Legacy: set `CLAUDE_EXTERNAL_LOCK=1` to re-enable the old `flock session.lock` wrapper for rollback.) **(b) llama-server has finite VRAM and one KV cache** — parallel inference thrashes the cache and risks OOM. All llama-backed agents serialize on the same lock. **(c) Disjoint backends are free to parallelize.** Today `disinto-agents` (Anthropic OAuth, runs `review,gardener`) runs concurrently with `disinto-agents-llama` (llama, runs `dev`) on the same project — they share neither OAuth state nor llama VRAM. **(d) Per-project work-conflict safety** (no duplicate dev work, no merge conflicts on the same branch) is enforced by `issue_claim` (assignee + `in-progress` label) and per-issue worktrees — that's a separate guard that does NOT depend on this AD. |
 | AD-003 | The runtime creates and destroys, the formula preserves. | Runtime manages worktrees/sessions/temp. Formulas commit knowledge to git before signaling done. |
 | AD-004 | Event-driven > polling > fixed delays. | Never `waitForTimeout` or hardcoded sleep. Use phase files, webhooks, or poll loops with backoff. |
-| AD-005 | Secrets via env var indirection, never in issue bodies. | Issue bodies become code. Agent secrets go in `.env.enc` (SOPS-encrypted), vault secrets in `secrets/<NAME>.enc` (age-encrypted, one file per key). Referenced as `$VAR_NAME`. Runner gets only vault secrets; agents get only agent secrets. |
+| AD-005 | Secrets via env var indirection, never in issue bodies. | Issue bodies become code. Agent secrets go in `.env.enc`, vault secrets in `.env.vault.enc` (SOPS-encrypted when available; plaintext `.env`/`.env.vault` fallback supported). Referenced as `$VAR_NAME`. Runner gets only vault secrets; agents get only agent secrets. |
-| AD-006 | External actions go through vault dispatch, never direct. | Agents build addressables; only the vault exercises them (publishes, deploys, posts). Tokens for external systems (`GITHUB_TOKEN`, `CLAWHUB_TOKEN`, deploy keys) live only in `secrets/<NAME>.enc` and are decrypted into the ephemeral runner container. `lib/env.sh` unsets them so agents never hold them. PRs with direct external actions without vault dispatch get REQUEST_CHANGES. (Vault redesign in progress: PR-based approval on ops repo, see #73-#77) |
+| AD-006 | External actions go through vault dispatch, never direct. | Agents build addressables; only the vault exercises them (publishes, deploys, posts). Tokens for external systems (`GITHUB_TOKEN`, `CLAWHUB_TOKEN`, deploy keys) live only in `.env.vault.enc` and are injected into the ephemeral runner container. `lib/env.sh` unsets them so agents never hold them. PRs with direct external actions without vault dispatch get REQUEST_CHANGES. (Vault redesign in progress: PR-based approval on ops repo, see #73-#77) |
 **Who enforces what:**
 - **Gardener** checks open backlog issues against ADs during grooming; closes violations with a comment referencing the AD number.
@ -189,6 +186,8 @@ Humans write these. Agents read and enforce them.
 - **Dev-agent** reads AGENTS.md before implementing; refuses work that violates ADs.
 - **AD-002 is a runtime invariant; nothing for the gardener to check at issue-groom time.** OAuth concurrency is handled by per-session `CLAUDE_CONFIG_DIR` isolation (with `CLAUDE_EXTERNAL_LOCK` as a rollback flag). Per-issue work is enforced by `issue_claim`. A violation manifests as a 401 or VRAM OOM in agent logs, not as a malformed issue.
 ---
 ## Phase-Signaling Protocol
 When running as a persistent tmux session, Claude must signal the orchestrator
@ -197,4 +196,5 @@ at each phase boundary by writing to a phase file (e.g.
 Key phases: `PHASE:awaiting_ci` → `PHASE:awaiting_review` → `PHASE:done`.
 Also: `PHASE:escalate` (needs human input), `PHASE:failed`.
 See [docs/PHASE-PROTOCOL.md](docs/PHASE-PROTOCOL.md) for the complete spec, orchestrator reaction matrix, sequence diagram, and crash recovery.
--- a/architect/AGENTS.md
+++ b/architect/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Architect — Agent Instructions
 ## What this agent is
@ -10,9 +10,9 @@ converses with humans through PR comments.
 ## Role
 - **Input**: Vision issues from VISION.md, prerequisite tree from ops repo
- **Output**: Sprint proposals as PRs on the ops repo (with embedded `## Sub-issues` blocks)
+- **Output**: Sprint proposals as PRs on the ops repo, sub-issue files
 - **Mechanism**: Bash-driven orchestration in `architect-run.sh`, pitching formula via `formulas/run-architect.toml`
- **Identity**: `architect-bot` on Forgejo (READ-ONLY on project repo, write on ops repo only — #764)
+- **Identity**: `architect-bot` on Forgejo
 ## Responsibilities
@ -24,17 +24,16 @@ converses with humans through PR comments.
   acceptance criteria and dependencies
 4. **Human conversation**: Respond to PR comments, refine sprint proposals based
   on human feedback
-5. **Sub-issue definition**: Define concrete sub-issues in the `## Sub-issues`
+5. **Sub-issue filing**: After design forks are resolved, file concrete sub-issues
-   block of the sprint spec. Filing is handled by `filer-bot` after sprint PR
+   for implementation
   merge (#764)
 ## Formula
 The architect pitching is driven by `formulas/run-architect.toml`. This formula defines
 the steps for:
 - Research: analyzing vision items and prerequisite tree
- Pitch: creating structured sprint PRs with embedded `## Sub-issues` blocks
+- Pitch: creating structured sprint PRs
- Design Q&A: refining the sprint via PR comments after human ACCEPT
+- Sub-issue filing: creating concrete implementation issues
 ## Bash-driven orchestration
@ -58,31 +57,21 @@ APPROVED review → start design questions (model posts Q1:, adds Design forks s
  ↓
 Answers received → continue Q&A (model processes answers, posts follow-ups)
  ↓
-All forks resolved → finalize ## Sub-issues section in sprint spec
+All forks resolved → sub-issue filing (model files implementation issues)
  ↓
 Sprint PR merged → filer-bot files sub-issues on project repo (#764)
  ↓
 REJECT review → close PR + journal (model processes rejection, bash merges PR)
 ```
 ### Vision issue lifecycle
-Vision issues decompose into sprint sub-issues. Sub-issues are defined in the
+Vision issues decompose into sprint sub-issues tracked via "Decomposed from #N" in sub-issue bodies. The architect automatically closes vision issues when all sub-issues are closed:
 `## Sub-issues` block of the sprint spec (between `<!-- filer:begin -->` and
 `<!-- filer:end -->` markers) and filed by `filer-bot` after the sprint PR merges
 on the ops repo (#764).
-Each filer-created sub-issue carries a `<!-- decomposed-from: #<vision>, sprint: <slug>, id: <id> -->`
+1. Before picking new vision issues, the architect checks each open vision issue
-marker in its body for idempotency and traceability.
+2. For each, it queries for sub-issues with "Decomposed from #N" in their body (regardless of state)
 3. If all sub-issues are closed, it posts a summary comment listing completed sub-issues
 4. The vision issue is then closed automatically
-The filer-bot (via `lib/sprint-filer.sh`) handles vision lifecycle:
+This ensures vision issues transition from `open` → `closed` once their work is complete, without manual intervention.
 1. After filing sub-issues, adds `in-progress` label to the vision issue
 2. On each run, checks if all sub-issues for a vision are closed
 3. If all closed, posts a summary comment and closes the vision issue
 The architect no longer writes to the project repo — it is read-only (#764).
 All project-repo writes (issue filing, label management, vision closure) are
 handled by filer-bot with its narrowly-scoped `FORGE_FILER_TOKEN`.
 ### Session management
@ -96,7 +85,6 @@ Run via `architect/architect-run.sh`, which:
 - Acquires a poll-loop lock (via `acquire_lock`) and checks available memory
 - Cleans up per-issue scratch files from previous runs (`/tmp/architect-{project}-scratch-*.md`)
 - Sources shared libraries (env.sh, formula-session.sh)
 - Exports `FORGE_TOKEN_OVERRIDE="${FORGE_ARCHITECT_TOKEN}"` BEFORE sourcing env.sh, ensuring architect-bot identity survives re-sourcing (#762)
 - Uses FORGE_ARCHITECT_TOKEN for authentication
 - Processes existing architect PRs via bash-driven design phase
 - Loads the formula and builds context from VISION.md, AGENTS.md, and ops repo
@ -106,9 +94,7 @@ Run via `architect/architect-run.sh`, which:
  - Selects up to `pitch_budget` (3 - open architect PRs) remaining vision issues
  - For each selected issue, invokes stateless `claude -p` with issue body + context
  - Creates PRs directly from pitch content (no scratch files)
- Agent is invoked for stateless pitch generation and response processing (ACCEPT/REJECT handling)
+- Agent is invoked only for response processing (ACCEPT/REJECT handling)
 - NOTE: architect-bot is read-only on the project repo (#764) — sub-issue filing
  and in-progress label management are handled by filer-bot after sprint PR merge
 **Multi-sprint pitching**: The architect pitches up to 3 sprints per run. Bash handles all state management:
 - Fetches Forgejo API data (vision issues, open PRs, merged PRs)
@ -133,5 +119,4 @@ empty file not created, just document it).
 - #100: Architect formula — research + design fork identification
 - #101: Architect formula — sprint PR creation with questions
 - #102: Architect formula — answer parsing + sub-issue filing
 - #764: Permission scoping — architect read-only on project repo, filer-bot files sub-issues
 - #491: Refactor — bash-driven design phase with stateful session resumption
--- a/architect/architect-run.sh
+++ b/architect/architect-run.sh
@ -34,11 +34,10 @@ FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
 # Accept project config from argument; default to disinto
 export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 # Set override BEFORE sourcing env.sh so it survives any later re-source of
 # env.sh from nested shells / claude -p tools (#762, #747)
 export FORGE_TOKEN_OVERRIDE="${FORGE_ARCHITECT_TOKEN:-}"
 # shellcheck source=../lib/env.sh
 source "$FACTORY_ROOT/lib/env.sh"
 # Override FORGE_TOKEN with architect-bot's token (#747)
 FORGE_TOKEN="${FORGE_ARCHITECT_TOKEN:-${FORGE_TOKEN}}"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
@ -117,8 +116,8 @@ build_architect_prompt() {
 You are the architect agent for ${FORGE_REPO}. Work through the formula below.
 Your role: strategic decomposition of vision issues into development sprints.
-Propose sprints via PRs on the ops repo, converse with humans through PR comments.
+Propose sprints via PRs on the ops repo, converse with humans through PR comments,
-You are READ-ONLY on the project repo — sub-issues are filed by filer-bot after sprint PR merge (#764).
+and file sub-issues after design forks are resolved.
 ## Project context
 ${CONTEXT_BLOCK}
@ -145,8 +144,8 @@ build_architect_prompt_for_mode() {
 You are the architect agent for ${FORGE_REPO}. Work through the formula below.
 Your role: strategic decomposition of vision issues into development sprints.
-Propose sprints via PRs on the ops repo, converse with humans through PR comments.
+Propose sprints via PRs on the ops repo, converse with humans through PR comments,
-You are READ-ONLY on the project repo — sub-issues are filed by filer-bot after sprint PR merge (#764).
+and file sub-issues after design forks are resolved.
 ## CURRENT STATE: Approved PR awaiting initial design questions
@ -157,10 +156,10 @@ design conversation has not yet started. Your task is to:
 2. Identify the key design decisions that need human input
 3. Post initial design questions (Q1:, Q2:, etc.) as comments on the PR
 4. Add a `## Design forks` section to the PR body documenting the design decisions
-5. Update the ## Sub-issues section in the sprint spec if design decisions affect decomposition
+5. File sub-issues for each design fork path if applicable
 This is NOT a pitch phase — the pitch is already approved. This is the START
-of the design Q&A phase. Sub-issues are filed by filer-bot after sprint PR merge (#764).
+of the design Q&A phase.
 ## Project context
 ${CONTEXT_BLOCK}
@ -179,8 +178,8 @@ _PROMPT_EOF_
 You are the architect agent for ${FORGE_REPO}. Work through the formula below.
 Your role: strategic decomposition of vision issues into development sprints.
-Propose sprints via PRs on the ops repo, converse with humans through PR comments.
+Propose sprints via PRs on the ops repo, converse with humans through PR comments,
-You are READ-ONLY on the project repo — sub-issues are filed by filer-bot after sprint PR merge (#764).
+and file sub-issues after design forks are resolved.
 ## CURRENT STATE: Design Q&A in progress
@ -194,7 +193,7 @@ Your task is to:
 2. Read human answers from PR comments
 3. Parse the answers and determine next steps
 4. Post follow-up questions if needed (Q3:, Q4:, etc.)
-5. If all design forks are resolved, finalize the ## Sub-issues section in the sprint spec
+5. If all design forks are resolved, file sub-issues for each path
 6. Update the `## Design forks` section as you progress
 ## Project context
@ -418,10 +417,205 @@ fetch_vision_issues() {
    "${FORGE_API}/issues?labels=vision&state=open&limit=100" 2>/dev/null || echo '[]'
 }
-# NOTE: get_vision_subissues, all_subissues_closed, close_vision_issue,
+# ── Helper: Fetch all sub-issues for a vision issue ───────────────────────
-# check_and_close_completed_visions removed (#764) — architect-bot is read-only
+# Sub-issues are identified by:
-# on the project repo. Vision lifecycle (closing completed visions, adding
+#   1. Issues whose body contains "Decomposed from #N" pattern
-# in-progress labels) is now handled by filer-bot via lib/sprint-filer.sh.
+#   2. Issues referenced in merged sprint PR bodies
 # Returns: newline-separated list of sub-issue numbers (empty if none)
 # Args: vision_issue_number
 get_vision_subissues() {
  local vision_issue="$1"
  local subissues=()
  # Method 1: Find issues with "Decomposed from #N" in body
  local issues_json
  issues_json=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
    "${FORGE_API}/issues?limit=100" 2>/dev/null) || true
  if [ -n "$issues_json" ] && [ "$issues_json" != "null" ]; then
    while IFS= read -r subissue_num; do
      [ -z "$subissue_num" ] && continue
      subissues+=("$subissue_num")
    done <<< "$(printf '%s' "$issues_json" | jq -r --arg vid "$vision_issue" \
      '[.[] | select(.number != ($vid | tonumber)) | select(.body // "" | contains("Decomposed from #" + $vid))] | .[].number' 2>/dev/null)"
  fi
  # Method 2: Find issues referenced in merged sprint PR bodies
  local prs_json
  prs_json=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
    "${FORGE_API_BASE}/repos/${FORGE_OPS_REPO}/pulls?state=closed&limit=100" 2>/dev/null) || true
  if [ -n "$prs_json" ] && [ "$prs_json" != "null" ]; then
    while IFS= read -r pr_num; do
      [ -z "$pr_num" ] && continue
      # Check if PR is merged and references the vision issue
      local pr_details pr_body
      pr_details=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
        "${FORGE_API_BASE}/repos/${FORGE_OPS_REPO}/pulls/${pr_num}" 2>/dev/null) || continue
      local is_merged
      is_merged=$(printf '%s' "$pr_details" | jq -r '.merged // false') || continue
      if [ "$is_merged" != "true" ]; then
        continue
      fi
      pr_body=$(printf '%s' "$pr_details" | jq -r '.body // ""') || continue
      # Extract all issue numbers from PR body
      while IFS= read -r ref_issue; do
        [ -z "$ref_issue" ] && continue
        # Skip if already in list
        local found=false
        for existing in "${subissues[@]+"${subissues[@]}"}"; do
          [ "$existing" = "$ref_issue" ] && found=true && break
        done
        if [ "$found" = false ]; then
          subissues+=("$ref_issue")
        fi
      done <<< "$(printf '%s' "$pr_body" | grep -oE '#[0-9]+' | tr -d '#' | sort -u)"
    done <<< "$(printf '%s' "$prs_json" | jq -r '.[] | select(.title | contains("architect:")) | .number')"
  fi
  # Output unique sub-issues
  printf '%s\n' "${subissues[@]}" | sort -u | grep -v '^$' || true
 }
 # ── Helper: Check if all sub-issues of a vision issue are closed ───────────
 # Returns: 0 if all sub-issues are closed, 1 if any are still open
 # Args: vision_issue_number
 all_subissues_closed() {
  local vision_issue="$1"
  local subissues
  subissues=$(get_vision_subissues "$vision_issue")
  # If no sub-issues found, parent cannot be considered complete
  if [ -z "$subissues" ]; then
    return 1
  fi
  # Check each sub-issue state
  while IFS= read -r subissue_num; do
    [ -z "$subissue_num" ] && continue
    local sub_state
    sub_state=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
      "${FORGE_API}/issues/${subissue_num}" 2>/dev/null | jq -r '.state // "unknown"') || true
    if [ "$sub_state" != "closed" ]; then
      log "Sub-issue #${subissue_num} is ${sub_state} — vision issue #${vision_issue} not ready to close"
      return 1
    fi
  done <<< "$subissues"
  return 0
 }
 # ── Helper: Close vision issue with summary comment ────────────────────────
 # Posts a comment listing all completed sub-issues before closing.
 # Returns: 0 on success, 1 on failure
 # Args: vision_issue_number
 close_vision_issue() {
  local vision_issue="$1"
  local subissues
  subissues=$(get_vision_subissues "$vision_issue")
  # Build summary comment
  local summary=""
  local count=0
  while IFS= read -r subissue_num; do
    [ -z "$subissue_num" ] && continue
    local sub_title
    sub_title=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
      "${FORGE_API}/issues/${subissue_num}" 2>/dev/null | jq -r '.title // "Untitled"') || sub_title="Untitled"
    summary+="- #${subissue_num}: ${sub_title}"$'\n'
    count=$((count + 1))
  done <<< "$subissues"
  local comment
  comment=$(cat <<EOF
 ## Vision Issue Completed
 All sub-issues have been implemented and merged. This vision issue is now closed.
 ### Completed sub-issues (${count}):
 ${summary}
 ---
 *Automated closure by architect · $(date -u '+%Y-%m-%d %H:%M UTC')*
 EOF
 )
  # Post comment before closing
  local tmpfile tmpjson
  tmpfile=$(mktemp /tmp/vision-close-XXXXXX.md)
  tmpjson="${tmpfile}.json"
  printf '%s' "$comment" > "$tmpfile"
  jq -Rs '{body:.}' < "$tmpfile" > "$tmpjson"
  if ! curl -sf -X POST \
    -H "Authorization: token ${FORGE_TOKEN}" \
    -H "Content-Type: application/json" \
    "${FORGE_API}/issues/${vision_issue}/comments" \
    --data-binary @"$tmpjson" >/dev/null 2>&1; then
    log "WARNING: failed to post closure comment on vision issue #${vision_issue}"
    rm -f "$tmpfile" "$tmpjson"
    return 1
  fi
  rm -f "$tmpfile" "$tmpjson"
  # Clear assignee and close the issue
  curl -sf -X PATCH \
    -H "Authorization: token ${FORGE_TOKEN}" \
    -H "Content-Type: application/json" \
    "${FORGE_API}/issues/${vision_issue}" \
    -d '{"assignees":[]}' >/dev/null 2>&1 || true
  curl -sf -X PATCH \
    -H "Authorization: token ${FORGE_TOKEN}" \
    -H "Content-Type: application/json" \
    "${FORGE_API}/issues/${vision_issue}" \
    -d '{"state":"closed"}' >/dev/null 2>&1 || true
  log "Closed vision issue #${vision_issue} — all ${count} sub-issue(s) complete"
  return 0
 }
 # ── Lifecycle check: Close vision issues with all sub-issues complete ──────
 # Runs before picking new vision issues for decomposition.
 # Checks each open vision issue and closes it if all sub-issues are closed.
 check_and_close_completed_visions() {
  log "Checking for vision issues with all sub-issues complete..."
  local vision_issues_json
  vision_issues_json=$(fetch_vision_issues)
  if [ -z "$vision_issues_json" ] || [ "$vision_issues_json" = "null" ]; then
    log "No open vision issues found"
    return 0
  fi
  # Get all vision issue numbers
  local vision_issue_nums
  vision_issue_nums=$(printf '%s' "$vision_issues_json" | jq -r '.[].number' 2>/dev/null) || vision_issue_nums=""
  local closed_count=0
  while IFS= read -r vision_issue; do
    [ -z "$vision_issue" ] && continue
    if all_subissues_closed "$vision_issue"; then
      if close_vision_issue "$vision_issue"; then
        closed_count=$((closed_count + 1))
      fi
    fi
  done <<< "$vision_issue_nums"
  if [ "$closed_count" -gt 0 ]; then
    log "Closed ${closed_count} vision issue(s) with all sub-issues complete"
  else
    log "No vision issues ready for closure"
  fi
 }
 # ── Helper: Fetch open architect PRs from ops repo Forgejo API ───────────
 # Returns: JSON array of architect PR objects
@ -513,23 +707,7 @@ Instructions:
 ## Recommendation
 <architect's assessment: worth it / defer / alternative approach>
 ## Sub-issues
 <!-- filer:begin -->
 - id: <kebab-case-id>
  title: \"vision(#${issue_num}): <concise sub-issue title>\"
  labels: [backlog]
  depends_on: []
  body: |
    ## Goal
    <what this sub-issue accomplishes>
    ## Acceptance criteria
    - [ ] <criterion>
 <!-- filer:end -->
 IMPORTANT: Do NOT include design forks or questions. This is a go/no-go pitch.
 The ## Sub-issues block is parsed by the filer-bot pipeline after sprint PR merge.
 Each sub-issue between filer:begin/end markers becomes a Forgejo issue.
 ---
@ -638,8 +816,37 @@ post_pr_footer() {
  fi
 }
-# NOTE: add_inprogress_label removed (#764) — architect-bot is read-only on
+# ── Helper: Add in-progress label to vision issue ────────────────────────
-# project repo. in-progress label is now added by filer-bot via sprint-filer.sh.
+# Args: vision_issue_number
 add_inprogress_label() {
  local issue_num="$1"
  # Get label ID for 'in-progress'
  local labels_json
  labels_json=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
    "${FORGE_API}/labels" 2>/dev/null) || return 1
  local inprogress_label_id
  inprogress_label_id=$(printf '%s' "$labels_json" | jq -r --arg label "in-progress" '.[] | select(.name == $label) | .id' 2>/dev/null) || true
  if [ -z "$inprogress_label_id" ]; then
    log "WARNING: in-progress label not found"
    return 1
  fi
  # Add label to issue
  if curl -sf -X POST \
    -H "Authorization: token ${FORGE_TOKEN}" \
    -H "Content-Type: application/json" \
    "${FORGE_API}/issues/${issue_num}/labels" \
    -d "{\"labels\": [${inprogress_label_id}]}" >/dev/null 2>&1; then
    log "Added in-progress label to vision issue #${issue_num}"
    return 0
  else
    log "WARNING: failed to add in-progress label to vision issue #${issue_num}"
    return 1
  fi
 }
 # ── Precondition checks in bash before invoking the model ─────────────────
@ -689,7 +896,9 @@ if [ "${open_arch_prs:-0}" -ge 3 ]; then
  log "3 open architect PRs found but responses detected — processing"
 fi
-# NOTE: Vision lifecycle check (close completed visions) moved to filer-bot (#764)
+# ── Lifecycle check: Close vision issues with all sub-issues complete ──────
 # Run before picking new vision issues for decomposition
 check_and_close_completed_visions
 # ── Bash-driven state management: Select vision issues for pitching ───────
 # This logic is also documented in formulas/run-architect.toml preflight step
@ -825,7 +1034,8 @@ for vision_issue in "${ARCHITECT_TARGET_ISSUES[@]}"; do
  # Post footer comment
  post_pr_footer "$pr_number"
-  # NOTE: in-progress label is added by filer-bot after sprint PR merge (#764)
+  # Add in-progress label to vision issue
  add_inprogress_label "$vision_issue"
  pitch_count=$((pitch_count + 1))
  log "Completed pitch for vision issue #${vision_issue} — PR #${pr_number}"
--- a/bin/disinto
+++ b/bin/disinto
@ -31,6 +31,9 @@ FACTORY_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 export USER="${USER:-$(id -un)}"
 export HOME="${HOME:-$(eval echo "~${USER}")}"
 # Chat Claude identity directory — separate from factory agents (#707)
 export CHAT_CLAUDE_DIR="${CHAT_CLAUDE_DIR:-${HOME}/.claude-chat}"
 source "${FACTORY_ROOT}/lib/env.sh"
 source "${FACTORY_ROOT}/lib/ops-setup.sh"  # setup_ops_repo, migrate_ops_repo
 source "${FACTORY_ROOT}/lib/hire-agent.sh"
@ -81,13 +84,9 @@ Init options:
  --repo-root <path>   Local clone path (default: ~/name)
  --ci-id <n>          Woodpecker CI repo ID (default: 0 = no CI)
  --forge-url <url>    Forge base URL (default: http://localhost:3000)
  --backend <value>    Orchestration backend: docker (default) | nomad
  --empty              (nomad) Bring up cluster only, no jobs (S0.4)
  --bare               Skip compose generation (bare-metal setup)
  --build              Use local docker build instead of registry images (dev mode)
  --yes                Skip confirmation prompts
  --rotate-tokens      Force regeneration of all bot tokens/passwords (idempotent by default)
  --dry-run            Print every intended action without executing
 Hire an agent options:
  --formula <path>     Path to role formula TOML (default: formulas/<role>.toml)
@ -644,65 +643,100 @@ prompt_admin_password() {
  exit 1
 }
-# ── init command ─────────────────────────────────────────────────────────────
+# ── Chat identity prompt helper ───────────────────────────────────────────────
 # Prompts for separate Claude identity for disinto-chat (#707).
 # On yes, creates ~/.claude-chat/ and runs claude login in subshell.
 # Returns 0 on success, 1 on failure/skip.
 # Usage: prompt_chat_identity [<env_file>]
 prompt_chat_identity() {
  local env_file="${1:-${FACTORY_ROOT}/.env}"
  local chat_claude_dir="${CHAT_CLAUDE_DIR:-${HOME}/.claude-chat}"
  local config_dir="${chat_claude_dir}/config"
-# Nomad backend init — dispatcher (Nomad+Vault migration, S0.4, issue #824).
+  # Skip if ANTHROPIC_API_KEY is set (API key mode — no OAuth needed)
-#
+  if grep -q '^ANTHROPIC_API_KEY=' "$env_file" 2>/dev/null; then
-# Today `--empty` and the default (no flag) both bring up an empty
+    local api_key
-# single-node Nomad+Vault cluster via lib/init/nomad/cluster-up.sh. Step 1
+    api_key=$(grep '^ANTHROPIC_API_KEY=' "$env_file" | cut -d= -f2-)
-# will extend the default path to also deploy jobs; `--empty` will remain
+    if [ -n "$api_key" ]; then
-# the "cluster only, no workloads" escape hatch.
+      echo "Chat:    ANTHROPIC_API_KEY set — skipping OAuth identity setup"
-#
+      echo "Chat:    Chat will use API key authentication (no ~/.claude-chat mount)"
-# Uses `sudo -n` when not already root — cluster-up.sh mutates /etc/,
+      return 0
 # /srv/, and systemd state, so it has to run as root. The `-n` keeps the
 # failure mode legible (no hanging TTY-prompted sudo inside a factory
 # init run); operators running without sudo-NOPASSWD should invoke
 # `sudo disinto init ...` directly.
 _disinto_init_nomad() {
  local dry_run="${1:-false}" empty="${2:-false}"
  local cluster_up="${FACTORY_ROOT}/lib/init/nomad/cluster-up.sh"
  if [ ! -x "$cluster_up" ]; then
    echo "Error: ${cluster_up} not found or not executable" >&2
    exit 1
  fi
  # --empty and default both invoke cluster-up today. Log the requested
  # mode so the dispatch is visible in factory bootstrap logs — Step 1
  # will branch on $empty to gate the job-deployment path.
  if [ "$empty" = "true" ]; then
    echo "nomad backend: --empty (cluster-up only, no jobs)"
  else
    echo "nomad backend: default (cluster-up; jobs deferred to Step 1)"
  fi
  # Dry-run forwards straight through; cluster-up.sh prints its own step
  # list and exits 0 without touching the box.
  local -a cmd=("$cluster_up")
  if [ "$dry_run" = "true" ]; then
    cmd+=("--dry-run")
    "${cmd[@]}"
    exit $?
  fi
  # Real run — needs root. Invoke via sudo if we're not already root so
  # the command's exit code propagates directly. We don't distinguish
  # "sudo denied" from "cluster-up.sh failed" here; both surface as a
  # non-zero exit, and cluster-up.sh's own error messages cover the
  # latter case.
  local rc=0
  if [ "$(id -u)" -eq 0 ]; then
    "${cmd[@]}" || rc=$?
  else
    if ! command -v sudo >/dev/null 2>&1; then
      echo "Error: cluster-up.sh must run as root and sudo is not installed" >&2
      exit 1
    fi
    sudo -n -- "${cmd[@]}" || rc=$?
  fi
-  exit "$rc"
+
  # Check if ~/.claude-chat already exists and has credentials
  if [ -d "$chat_claude_dir" ]; then
    if [ -d "${config_dir}/credentials" ] && [ -n "$(ls -A "${config_dir}/credentials" 2>/dev/null)" ]; then
      echo "Chat:    ~/.claude-chat already configured (resuming from existing identity)"
      return 0
    fi
  fi
  # Non-interactive mode without explicit yes
  if [ "$auto_yes" = true ]; then
    echo "Chat:    --yes provided — creating ~/.claude-chat identity"
    # Create directory structure
    install -d -m 0700 -o "$USER" "$config_dir"
    # Skip claude login in non-interactive mode — user must run manually
    echo "Chat:    Identity directory created at ${chat_claude_dir}"
    echo "Chat:    Run 'CLAUDE_CONFIG_DIR=${config_dir} claude login' to authenticate"
    return 0
  fi
  # Interactive mode: prompt for separate identity
  if [ -t 0 ]; then
    echo ""
    echo "── Separate Claude identity for disinto-chat ──────────────────────"
    echo "Create a separate Claude identity for disinto-chat to avoid OAuth"
    echo "refresh races with factory agents sharing ~/.claude?"
    echo ""
    printf "Use separate identity for chat? [Y/n] "
    local response
    IFS= read -r response
    response="${response:-Y}"
    case "$response" in
      [Nn][Oo]|[Nn])
        echo "Chat:    Skipping separate identity — chat will use shared ~/.claude"
        echo "Chat:    Note: OAuth refresh races may occur with concurrent agents"
        return 1
        ;;
      *)
        # User said yes — create identity and run login
        echo ""
        echo "Chat:    Creating identity directory at ${chat_claude_dir}"
        install -d -m 0700 -o "$USER" "$config_dir"
        # Run claude login in subshell with CLAUDE_CONFIG_DIR set
        echo "Chat:    Launching Claude login for chat identity..."
        echo "Chat:    (This will authenticate to a separate Anthropic account)"
        echo ""
        CLAUDE_CONFIG_DIR="$config_dir" claude login
        local login_rc=$?
        if [ $login_rc -eq 0 ]; then
          echo ""
          echo "Chat:    Claude identity configured at ${chat_claude_dir}"
          echo "Chat:    Chat container will use: ${chat_claude_dir}:/home/chat/.claude-chat"
          return 0
        else
          echo "Error:   Claude login failed (exit code ${login_rc})" >&2
          echo "Chat:    Identity not configured"
          return 1
        fi
        ;;
    esac
  fi
  # Non-interactive, no TTY
  echo "Warning: cannot prompt for chat identity (no TTY)" >&2
  echo "Chat:    Skipping identity setup — run manually if needed"
  return 1
 }
 # ── init command ─────────────────────────────────────────────────────────────
 disinto_init() {
  local repo_url="${1:-}"
  if [ -z "$repo_url" ]; then
@ -713,49 +747,20 @@ disinto_init() {
  shift
  # Parse flags
-  local branch="" repo_root="" ci_id="0" auto_yes=false forge_url_flag="" bare=false rotate_tokens=false use_build=false dry_run=false backend="docker" empty=false
+  local branch="" repo_root="" ci_id="0" auto_yes=false forge_url_flag="" bare=false rotate_tokens=false
  while [ $# -gt 0 ]; do
    case "$1" in
      --branch)        branch="$2"; shift 2 ;;
      --repo-root)     repo_root="$2"; shift 2 ;;
      --ci-id)         ci_id="$2"; shift 2 ;;
      --forge-url)     forge_url_flag="$2"; shift 2 ;;
      --backend)       backend="$2"; shift 2 ;;
      --backend=*)     backend="${1#--backend=}"; shift ;;
      --bare)          bare=true; shift ;;
      --build)         use_build=true; shift ;;
      --empty)         empty=true; shift ;;
      --yes)           auto_yes=true; shift ;;
      --rotate-tokens) rotate_tokens=true; shift ;;
      --dry-run)       dry_run=true; shift ;;
      *) echo "Unknown option: $1" >&2; exit 1 ;;
    esac
  done
  # Validate backend
  case "$backend" in
    docker|nomad) ;;
    *) echo "Error: invalid --backend value '${backend}' (expected: docker|nomad)" >&2; exit 1 ;;
  esac
  # --empty is nomad-only today (the docker path has no concept of an
  # "empty cluster"). Reject explicitly rather than letting it silently
  # do nothing on --backend=docker.
  if [ "$empty" = true ] && [ "$backend" != "nomad" ]; then
    echo "Error: --empty is only valid with --backend=nomad" >&2
    exit 1
  fi
  # Dispatch on backend — the nomad path runs lib/init/nomad/cluster-up.sh
  # (S0.4). The default and --empty variants are identical today; Step 1
  # will branch on $empty to add job deployment to the default path.
  if [ "$backend" = "nomad" ]; then
    _disinto_init_nomad "$dry_run" "$empty"
    # shellcheck disable=SC2317  # _disinto_init_nomad always exits today;
    # `return` is defensive against future refactors.
    return
  fi
  # Export bare-metal flag for setup_forge
  export DISINTO_BARE="$bare"
@ -828,92 +833,12 @@ p.write_text(text)
    fi
  fi
  # ── Dry-run mode: report intended actions and exit ─────────────────────────
  if [ "$dry_run" = true ]; then
    echo ""
    echo "── Dry-run: intended actions ────────────────────────────"
    local env_file="${FACTORY_ROOT}/.env"
    local rr="${repo_root:-/home/${USER}/${project_name}}"
    if [ "$bare" = false ]; then
      [ -f "${FACTORY_ROOT}/docker-compose.yml" ] \
        && echo "[skip]   docker-compose.yml (exists)" \
        || echo "[create] docker-compose.yml"
    fi
    [ -f "$env_file" ] \
      && echo "[exists] .env" \
      || echo "[create] .env"
    # Report token state from .env
    if [ -f "$env_file" ]; then
      local _var
      for _var in FORGE_ADMIN_TOKEN HUMAN_TOKEN FORGE_TOKEN FORGE_REVIEW_TOKEN \
                  FORGE_PLANNER_TOKEN FORGE_GARDENER_TOKEN FORGE_VAULT_TOKEN \
                  FORGE_SUPERVISOR_TOKEN FORGE_PREDICTOR_TOKEN FORGE_ARCHITECT_TOKEN; do
        if grep -q "^${_var}=" "$env_file" 2>/dev/null; then
          echo "[keep]   ${_var} (preserved)"
        else
          echo "[create] ${_var}"
        fi
      done
    else
      echo "[create] all tokens and passwords"
    fi
    echo ""
    echo "[ensure] Forgejo admin user 'disinto-admin'"
    echo "[ensure] 8 bot users: dev-bot, review-bot, planner-bot, gardener-bot, vault-bot, supervisor-bot, predictor-bot, architect-bot"
    echo "[ensure] 2 llama bot users: dev-qwen, dev-qwen-nightly"
    echo "[ensure] .profile repos for all bots"
    echo "[ensure] repo ${forge_repo} on Forgejo with collaborators"
    echo "[run]    preflight checks"
    [ -d "${rr}/.git" ] \
      && echo "[skip]   clone ${rr} (exists)" \
      || echo "[clone]  ${repo_url} -> ${rr}"
    echo "[push]   to local Forgejo"
    echo "[ensure] ops repo disinto-admin/${project_name}-ops"
    echo "[ensure] branch protection on ${forge_repo}"
    [ "$toml_exists" = true ] \
      && echo "[skip]   ${toml_path} (exists)" \
      || echo "[create] ${toml_path}"
    if [ "$bare" = false ]; then
      echo "[ensure] Woodpecker OAuth2 app"
      echo "[ensure] Chat OAuth2 app"
      echo "[ensure] WOODPECKER_AGENT_SECRET in .env"
    fi
    echo "[ensure] labels on ${forge_repo}"
    [ -f "${rr}/VISION.md" ] \
      && echo "[skip]   VISION.md (exists)" \
      || echo "[create] VISION.md"
    echo "[copy]   issue templates"
    echo "[ensure] scheduling (cron or compose polling)"
    if [ "$bare" = false ]; then
      echo "[start]  docker compose stack"
      echo "[ensure] Woodpecker token + repo activation"
    fi
    echo "[ensure] CLAUDE_CONFIG_DIR"
    echo "[ensure] state files (.dev-active, .reviewer-active, .gardener-active)"
    echo ""
    echo "Dry run complete — no changes made."
    exit 0
  fi
  # Generate compose files (unless --bare)
  if [ "$bare" = false ]; then
    local forge_port
    forge_port=$(printf '%s' "$forge_url" | sed -E 's|.*:([0-9]+)/?$|\1|')
    forge_port="${forge_port:-3000}"
-    generate_compose "$forge_port" "$use_build"
+    generate_compose "$forge_port"
    generate_agent_docker
    generate_caddyfile
    generate_staging_index
@ -938,6 +863,9 @@ p.write_text(text)
  # This ensures the password is set before Forgejo user creation
  prompt_admin_password "${FACTORY_ROOT}/.env"
  # Prompt for separate chat identity (#707)
  prompt_chat_identity "${FACTORY_ROOT}/.env"
  # Set up local Forgejo instance (provision if needed, create users/tokens/repo)
  if [ "$rotate_tokens" = true ]; then
    echo "Note: Forcing token rotation (tokens/passwords will be regenerated)"
@ -1060,19 +988,6 @@ p.write_text(text)
    echo "Config:  CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=1 saved to .env"
  fi
  # Write local-Qwen dev agent env keys with safe defaults (#769)
  if ! grep -q '^ENABLE_LLAMA_AGENT=' "$env_file" 2>/dev/null; then
    cat >> "$env_file" <<'LLAMAENVEOF'
 # Local Qwen dev agent (optional) — set to 1 to enable
 ENABLE_LLAMA_AGENT=0
 FORGE_TOKEN_LLAMA=
 FORGE_PASS_LLAMA=
 ANTHROPIC_BASE_URL=
 LLAMAENVEOF
    echo "Config:  ENABLE_LLAMA_AGENT keys written to .env (disabled by default)"
  fi
  # Create labels on remote
  create_labels "$forge_repo" "$forge_url"
@ -1174,9 +1089,10 @@ p.write_text(text)
    exit 1
  fi
-  # Write CLAUDE_SHARED_DIR and CLAUDE_CONFIG_DIR to .env (idempotent)
+  # Write CLAUDE_SHARED_DIR, CLAUDE_CONFIG_DIR, and CHAT_CLAUDE_DIR to .env (idempotent)
  _env_set_idempotent "CLAUDE_SHARED_DIR" "$CLAUDE_SHARED_DIR" "$env_file"
  _env_set_idempotent "CLAUDE_CONFIG_DIR" "$CLAUDE_CONFIG_DIR" "$env_file"
  _env_set_idempotent "CHAT_CLAUDE_DIR" "$CHAT_CLAUDE_DIR" "$env_file"
  # Activate default agents (zero-cost when idle — they only invoke Claude
  # when there is actual work, so an empty project burns no LLM tokens)
@ -1204,15 +1120,24 @@ p.write_text(text)
  fi
  echo ""
  echo "── Claude authentication ──────────────────────────────"
-  echo "  OAuth (shared across containers):"
+  echo "  Factory agents (shared OAuth):"
  echo "    Run 'claude auth login' on the host once."
  echo "    Credentials in ${CLAUDE_CONFIG_DIR} are shared across containers."
  echo ""
  echo "  disinto-chat (separate identity #707):"
  echo "    Separate identity created at ${CHAT_CLAUDE_DIR:-${HOME}/.claude-chat}"
  echo "    Container mounts: ${CHAT_CLAUDE_DIR:-${HOME}/.claude-chat}:/home/chat/.claude-chat"
  echo "    CLAUDE_CONFIG_DIR=/home/chat/.claude-chat/config"
  echo ""
  echo "  API key (alternative — metered billing, no rotation issues):"
  echo "    Set ANTHROPIC_API_KEY in .env to skip OAuth entirely."
  echo "    Chat container will not mount identity directory."
  echo ""
-  echo "── Claude config directory ────────────────────────────"
+  echo "── Claude config directories ───────────────────────────"
-  echo "  CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR}"
+  echo "  Factory agents: CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR}"
-  echo "  Add this to your shell rc (~/.bashrc or ~/.zshrc):"
+  echo "  Chat service:   CLAUDE_CONFIG_DIR=/home/chat/.claude-chat/config"
  echo ""
  echo "  Add factory agents CLAUDE_CONFIG_DIR to your shell rc:"
  echo "    export CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR}"
  echo "  This ensures interactive Claude Code sessions on this host"
  echo "  share the same OAuth lock and token store as the factory."
@ -1301,6 +1226,8 @@ disinto_secrets() {
  local subcmd="${1:-}"
  local enc_file="${FACTORY_ROOT}/.env.enc"
  local env_file="${FACTORY_ROOT}/.env"
  local vault_enc_file="${FACTORY_ROOT}/.env.vault.enc"
  local vault_env_file="${FACTORY_ROOT}/.env.vault"
  # Shared helper: ensure sops+age and .sops.yaml exist
  _secrets_ensure_sops() {
@ -1346,51 +1273,30 @@ disinto_secrets() {
  case "$subcmd" in
    add)
-      # Parse flags
+      local name="${2:-}"
      local force=false
      shift  # consume 'add'
      while [ $# -gt 0 ]; do
        case "$1" in
          -f|--force) force=true; shift ;;
          -*) echo "Unknown flag: $1" >&2; exit 1 ;;
          *) break ;;
        esac
      done
      local name="${1:-}"
      if [ -z "$name" ]; then
-        echo "Usage: disinto secrets add [-f|--force] <NAME>" >&2
+        echo "Usage: disinto secrets add <NAME>" >&2
        exit 1
      fi
      _secrets_ensure_age_key
      mkdir -p "$secrets_dir"
      printf 'Enter value for %s: ' "$name" >&2
      local value
-      if [ -t 0 ]; then
+      IFS= read -rs value
-        # Interactive TTY — prompt with hidden input (original behavior)
+      echo >&2
        printf 'Enter value for %s: ' "$name" >&2
        IFS= read -rs value
        echo >&2
      else
        # Piped/redirected stdin — read raw bytes verbatim
        IFS= read -r -d '' value || true
      fi
      if [ -z "$value" ]; then
        echo "Error: empty value" >&2
        exit 1
      fi
      local enc_path="${secrets_dir}/${name}.enc"
-      if [ -f "$enc_path" ] && [ "$force" = false ]; then
+      if [ -f "$enc_path" ]; then
-        if [ -t 0 ]; then
+        printf 'Secret %s already exists. Overwrite? [y/N] ' "$name" >&2
-          printf 'Secret %s already exists. Overwrite? [y/N] ' "$name" >&2
+        local confirm
-          local confirm
+        read -r confirm
-          read -r confirm
+        if [ "$confirm" != "y" ] && [ "$confirm" != "Y" ]; then
-          if [ "$confirm" != "y" ] && [ "$confirm" != "Y" ]; then
+          echo "Aborted." >&2
            echo "Aborted." >&2
            exit 1
          fi
        else
          echo "Error: secret ${name} already exists (use -f to overwrite)" >&2
          exit 1
        fi
      fi
@ -1423,37 +1329,6 @@ disinto_secrets() {
        sops -d "$enc_file"
      fi
      ;;
    remove)
      local name="${2:-}"
      if [ -z "$name" ]; then
        echo "Usage: disinto secrets remove <NAME>" >&2
        exit 1
      fi
      local enc_path="${secrets_dir}/${name}.enc"
      if [ ! -f "$enc_path" ]; then
        echo "Error: ${enc_path} not found" >&2
        exit 1
      fi
      rm -f "$enc_path"
      echo "Removed: ${enc_path}"
      ;;
    list)
      if [ ! -d "$secrets_dir" ]; then
        echo "No secrets directory found." >&2
        exit 0
      fi
      local found=false
      for enc_file_path in "${secrets_dir}"/*.enc; do
        [ -f "$enc_file_path" ] || continue
        found=true
        local secret_name
        secret_name=$(basename "$enc_file_path" .enc)
        echo "$secret_name"
      done
      if [ "$found" = false ]; then
        echo "No secrets stored." >&2
      fi
      ;;
    edit)
      if [ ! -f "$enc_file" ]; then
        echo "Error: ${enc_file} not found. Run 'disinto secrets migrate' first." >&2
@ -1477,100 +1352,54 @@ disinto_secrets() {
      rm -f "$env_file"
      echo "Migrated: .env -> .env.enc (plaintext removed)"
      ;;
-    migrate-from-vault)
+    edit-vault)
-      # One-shot migration: split .env.vault.enc into secrets/<KEY>.enc files (#777)
+      if [ ! -f "$vault_enc_file" ]; then
-      local vault_enc_file="${FACTORY_ROOT}/.env.vault.enc"
+        echo "Error: ${vault_enc_file} not found. Run 'disinto secrets migrate-vault' first." >&2
      local vault_env_file="${FACTORY_ROOT}/.env.vault"
      local source_file=""
      if [ -f "$vault_enc_file" ] && command -v sops &>/dev/null; then
        source_file="$vault_enc_file"
      elif [ -f "$vault_env_file" ]; then
        source_file="$vault_env_file"
      else
        echo "Error: neither .env.vault.enc nor .env.vault found — nothing to migrate." >&2
        exit 1
      fi
-
+      sops "$vault_enc_file"
-      _secrets_ensure_age_key
+      ;;
-      mkdir -p "$secrets_dir"
+    show-vault)
-
+      if [ ! -f "$vault_enc_file" ]; then
-      # Decrypt vault to temp dotenv
+        echo "Error: ${vault_enc_file} not found." >&2
      local tmp_dotenv
      tmp_dotenv=$(mktemp /tmp/disinto-vault-migrate-XXXXXX)
      trap 'rm -f "$tmp_dotenv"' RETURN
      if [ "$source_file" = "$vault_enc_file" ]; then
        if ! sops -d --output-type dotenv "$vault_enc_file" > "$tmp_dotenv" 2>/dev/null; then
          rm -f "$tmp_dotenv"
          echo "Error: failed to decrypt .env.vault.enc" >&2
          exit 1
        fi
      else
        cp "$vault_env_file" "$tmp_dotenv"
      fi
      # Parse each KEY=VALUE and encrypt into secrets/<KEY>.enc
      local count=0
      local failed=0
      while IFS='=' read -r key value; do
        # Skip empty lines and comments
        [[ -z "$key" || "$key" =~ ^[[:space:]]*# ]] && continue
        # Trim whitespace from key
        key=$(echo "$key" | xargs)
        [ -z "$key" ] && continue
        local enc_path="${secrets_dir}/${key}.enc"
        if printf '%s' "$value" | age -r "$AGE_PUBLIC_KEY" -o "$enc_path" 2>/dev/null; then
          # Verify round-trip
          local check
          check=$(age -d -i "$age_key_file" "$enc_path" 2>/dev/null) || { failed=$((failed + 1)); echo "  FAIL (verify): ${key}" >&2; continue; }
          if [ "$check" = "$value" ]; then
            echo "  OK: ${key} -> secrets/${key}.enc"
            count=$((count + 1))
          else
            echo "  FAIL (mismatch): ${key}" >&2
            failed=$((failed + 1))
          fi
        else
          echo "  FAIL (encrypt): ${key}" >&2
          failed=$((failed + 1))
        fi
      done < "$tmp_dotenv"
      rm -f "$tmp_dotenv"
      if [ "$failed" -gt 0 ]; then
        echo "Error: ${failed} secret(s) failed migration. Vault files NOT removed." >&2
        exit 1
      fi
-
+      sops -d "$vault_enc_file"
-      if [ "$count" -eq 0 ]; then
+      ;;
-        echo "Warning: no secrets found in vault file." >&2
+    migrate-vault)
-      else
+      if [ ! -f "$vault_env_file" ]; then
-        echo "Migrated ${count} secret(s) to secrets/*.enc"
+        echo "Error: ${vault_env_file} not found — nothing to migrate." >&2
-        # Remove old vault files on success
+        echo "  Create .env.vault with vault secrets (GITHUB_TOKEN, deploy keys, etc.)" >&2
-        rm -f "$vault_enc_file" "$vault_env_file"
+        exit 1
        echo "Removed: .env.vault.enc / .env.vault"
      fi
      _secrets_ensure_sops
      encrypt_env_file "$vault_env_file" "$vault_enc_file"
      # Verify decryption works before removing plaintext
      if ! sops -d "$vault_enc_file" >/dev/null 2>&1; then
        echo "Error: failed to verify .env.vault.enc decryption" >&2
        rm -f "$vault_enc_file"
        exit 1
      fi
      rm -f "$vault_env_file"
      echo "Migrated: .env.vault -> .env.vault.enc (plaintext removed)"
      ;;
    *)
      cat <<EOF >&2
 Usage: disinto secrets <subcommand>
-Secrets (secrets/<NAME>.enc — age-encrypted, one file per key):
+Individual secrets (secrets/<NAME>.enc):
-  add <NAME>          Prompt for value, encrypt, store in secrets/<NAME>.enc
+  add <NAME>     Prompt for value, encrypt, store in secrets/<NAME>.enc
-  show <NAME>         Decrypt and print a secret
+  show <NAME>    Decrypt and print an individual secret
  remove <NAME>       Remove a secret
  list                List all stored secrets
-Agent secrets (.env.enc — sops-encrypted dotenv):
+Agent secrets (.env.enc):
-  edit                Edit agent secrets (FORGE_TOKEN, CLAUDE_API_KEY, etc.)
+  edit           Edit agent secrets (FORGE_TOKEN, CLAUDE_API_KEY, etc.)
-  show                Show decrypted agent secrets (no argument)
+  show           Show decrypted agent secrets (no argument)
-  migrate             Encrypt .env -> .env.enc
+  migrate        Encrypt .env -> .env.enc
-Migration:
+Vault secrets (.env.vault.enc):
-  migrate-from-vault  Split .env.vault.enc into secrets/<KEY>.enc (one-shot)
+  edit-vault     Edit vault secrets (GITHUB_TOKEN, deploy keys, etc.)
  show-vault     Show decrypted vault secrets
  migrate-vault  Encrypt .env.vault -> .env.vault.enc
 EOF
      exit 1
      ;;
@ -1582,8 +1411,7 @@ EOF
 disinto_run() {
  local action_id="${1:?Usage: disinto run <action-id>}"
  local compose_file="${FACTORY_ROOT}/docker-compose.yml"
-  local secrets_dir="${FACTORY_ROOT}/secrets"
+  local vault_enc="${FACTORY_ROOT}/.env.vault.enc"
  local age_key_file="${HOME}/.config/sops/age/keys.txt"
  if [ ! -f "$compose_file" ]; then
    echo "Error: docker-compose.yml not found" >&2
@ -1591,42 +1419,29 @@ disinto_run() {
    exit 1
  fi
-  if [ ! -d "$secrets_dir" ]; then
+  if [ ! -f "$vault_enc" ]; then
-    echo "Error: secrets/ directory not found — create secrets first" >&2
+    echo "Error: .env.vault.enc not found — create vault secrets first" >&2
-    echo "  Run 'disinto secrets add <NAME>' to add secrets" >&2
+    echo "  Run 'disinto secrets migrate-vault' after creating .env.vault" >&2
    exit 1
  fi
-  if ! command -v age &>/dev/null; then
+  if ! command -v sops &>/dev/null; then
-    echo "Error: age not found — required to decrypt secrets" >&2
+    echo "Error: sops not found — required to decrypt vault secrets" >&2
    exit 1
  fi
-  if [ ! -f "$age_key_file" ]; then
+  # Decrypt vault secrets to temp file
    echo "Error: age key not found at ${age_key_file}" >&2
    exit 1
  fi
  # Decrypt all secrets/*.enc into a temp env file for the runner
  local tmp_env
-  tmp_env=$(mktemp /tmp/disinto-secrets-XXXXXX)
+  tmp_env=$(mktemp /tmp/disinto-vault-XXXXXX)
  trap 'rm -f "$tmp_env"' EXIT
-  local count=0
+  if ! sops -d --output-type dotenv "$vault_enc" > "$tmp_env" 2>/dev/null; then
-  for enc_path in "${secrets_dir}"/*.enc; do
+    rm -f "$tmp_env"
-    [ -f "$enc_path" ] || continue
+    echo "Error: failed to decrypt .env.vault.enc" >&2
-    local key
+    exit 1
-    key=$(basename "$enc_path" .enc)
+  fi
    local val
    val=$(age -d -i "$age_key_file" "$enc_path" 2>/dev/null) || {
      echo "Warning: failed to decrypt ${enc_path}" >&2
      continue
    }
    printf '%s=%s\n' "$key" "$val" >> "$tmp_env"
    count=$((count + 1))
  done
-  echo "Decrypted ${count} secret(s) to tmpfile"
+  echo "Vault secrets decrypted to tmpfile"
  # Run action in ephemeral runner container
  local rc=0
@ -1697,96 +1512,21 @@ download_agent_binaries() {
 # ── up command ────────────────────────────────────────────────────────────────
 # Regenerate a file idempotently: run the generator, compare output, backup if changed.
 # Usage: _regen_file <target_file> <generator_fn> [args...]
 _regen_file() {
  local target="$1"; shift
  local generator="$1"; shift
  local basename
  basename=$(basename "$target")
  # Move existing file aside so the generator (which skips if file exists)
  # produces a fresh copy.
  local stashed=""
  if [ -f "$target" ]; then
    stashed=$(mktemp "${target}.stash.XXXXXX")
    mv "$target" "$stashed"
  fi
  # Run the generator — it writes $target from scratch.
  # If the generator fails, restore the stashed original so it is not stranded.
  if ! "$generator" "$@"; then
    if [ -n "$stashed" ]; then
      mv "$stashed" "$target"
    fi
    return 1
  fi
  if [ -z "$stashed" ]; then
    # No previous file — first generation
    echo "regenerated: ${basename} (new)"
    return
  fi
  if cmp -s "$stashed" "$target"; then
    # Content unchanged — restore original to preserve mtime
    mv "$stashed" "$target"
    echo "unchanged: ${basename}"
  else
    # Content changed — keep new, save old as .prev
    mv "$stashed" "${target}.prev"
    echo "regenerated: ${basename} (previous saved as ${basename}.prev)"
  fi
 }
 disinto_up() {
  local compose_file="${FACTORY_ROOT}/docker-compose.yml"
  local caddyfile="${FACTORY_ROOT}/docker/Caddyfile"
  if [ ! -f "$compose_file" ]; then
    echo "Error: docker-compose.yml not found" >&2
    echo "  Run 'disinto init <repo-url>' first (without --bare)" >&2
    exit 1
  fi
-  # Parse --no-regen flag; remaining args pass through to docker compose
+  # Pre-build: download binaries to docker/agents/bin/ to avoid network calls during docker build
-  local no_regen=false
+  echo "── Pre-build: downloading agent binaries ────────────────────────"
-  local -a compose_args=()
+  if ! download_agent_binaries; then
-  for arg in "$@"; do
+    echo "Error: failed to download agent binaries" >&2
-    case "$arg" in
+    exit 1
      --no-regen) no_regen=true ;;
      *)          compose_args+=("$arg") ;;
    esac
  done
  # ── Regenerate compose & Caddyfile from generators ──────────────────────
  if [ "$no_regen" = true ]; then
    echo "Warning: running with unmanaged compose — hand-edits will drift" >&2
  else
    # Determine forge_port from FORGE_URL (same logic as init)
    local forge_url="${FORGE_URL:-http://localhost:3000}"
    local forge_port
    forge_port=$(printf '%s' "$forge_url" | sed -E 's|.*:([0-9]+)/?$|\1|')
    forge_port="${forge_port:-3000}"
    # Detect build mode from existing compose
    local use_build=false
    if grep -q '^\s*build:' "$compose_file"; then
      use_build=true
    fi
    _regen_file "$compose_file" generate_compose "$forge_port" "$use_build"
    _regen_file "$caddyfile"    generate_caddyfile
  fi
  # Pre-build: download binaries only when compose uses local build
  if grep -q '^\s*build:' "$compose_file"; then
    echo "── Pre-build: downloading agent binaries ────────────────────────"
    if ! download_agent_binaries; then
      echo "Error: failed to download agent binaries" >&2
      exit 1
    fi
    echo ""
  fi
  echo ""
  # Decrypt secrets to temp .env if SOPS available and .env.enc exists
  local tmp_env=""
@ -1799,7 +1539,7 @@ disinto_up() {
    echo "Decrypted secrets for compose"
  fi
-  docker compose -f "$compose_file" up -d --build --remove-orphans ${compose_args[@]+"${compose_args[@]}"}
+  docker compose -f "$compose_file" up -d "$@"
  echo "Stack is up"
  # Clean up temp .env (also handled by EXIT trap if compose fails)
--- a/dev/AGENTS.md
+++ b/dev/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Dev Agent
 **Role**: Implement issues autonomously — write code, push branches, address
@ -29,11 +29,7 @@ stale checks (vision issues are managed by the architect). If the issue is assig
 `REQUEST_CHANGES`, spawns the dev-agent to address it before setting `BLOCKED_BY_INPROGRESS=true`;
 otherwise just sets blocked. If assigned to another agent, logs and falls through (does not
 block). If no assignee, no open PR, and no agent lock file — removes `in-progress`, adds
-`blocked` with a human-triage comment. **Post-crash self-assigned recovery (#749)**: when the
+`blocked` with a human-triage comment. **Per-agent open-PR gate**: before starting new work,
 issue is self-assigned (this bot) but there is no open PR, dev-poll now checks for a lock
 file (`/tmp/dev-impl-summary-$PROJECT_NAME-$ISSUE_NUM.txt`) AND a remote branch
 (`fix/issue-$ISSUE_NUM`) before declaring "my thread is busy". If neither exists after a cold
 boot, it spawns a fresh dev-agent for recovery instead of looping forever. **Per-agent open-PR gate**: before starting new work,
 filters open waiting PRs to only those assigned to this agent (`$BOT_USER`). Other agents'
 PRs do not block this agent's pipeline (#358, #369). **Pre-lock merge scan own-PRs only**:
 the direct-merge scan only merges PRs whose linked issue is assigned to this agent — skips
@ -55,12 +51,6 @@ PRs owned by other bot users (#374).
 **Crash recovery**: on `PHASE:crashed` or non-zero exit, the worktree is **preserved** (not destroyed) for debugging. Location logged. Supervisor housekeeping removes stale crashed worktrees older than 24h.
 **Polling loop isolation (#753)**: `docker/agents/entrypoint.sh` now tracks fast-poll PIDs
 (`FAST_PIDS`) and calls `wait "${FAST_PIDS[@]}"` instead of `wait` (no-args). This means
 long-running dev-agent sessions no longer block the loop from launching the next iteration's
 fast polls — the loop only waits for review-poll and dev-poll (the fast agents), never for
 the dev-agent subprocess itself.
 **Lifecycle**: dev-poll.sh (invoked by polling loop, `check_active dev`) → dev-agent.sh →
 tmux session → phase file drives CI/review loop → merge + `mirror_push()` → close issue.
 On respawn after `PHASE:escalate`, the stale phase file is cleared first so the session
--- a/dev/dev-poll.sh
+++ b/dev/dev-poll.sh
@ -476,19 +476,8 @@ if [ "$ORPHAN_COUNT" -gt 0 ]; then
          BLOCKED_BY_INPROGRESS=true
        fi
      else
-        # No open PR — check if a thread is actually alive (lock file or remote branch)
+        log "issue #${ISSUE_NUM} assigned to me — my thread is busy"
-        LOCK_FILE="/tmp/dev-impl-summary-${PROJECT_NAME}-${ISSUE_NUM}.txt"
+        BLOCKED_BY_INPROGRESS=true
        REMOTE_BRANCH_EXISTS=$(git ls-remote --exit-code origin "fix/issue-${ISSUE_NUM}" >/dev/null 2>&1 && echo yes || echo no)
        if [ -f "$LOCK_FILE" ] || [ "$REMOTE_BRANCH_EXISTS" = "yes" ]; then
          log "issue #${ISSUE_NUM} assigned to me — my thread is busy (lock=$([ -f "$LOCK_FILE" ] && echo y || echo n) remote_branch=$REMOTE_BRANCH_EXISTS)"
          BLOCKED_BY_INPROGRESS=true
        else
          log "issue #${ISSUE_NUM} self-assigned but orphaned (no lock, no branch, no PR) — recovering"
          nohup "${SCRIPT_DIR}/dev-agent.sh" "$ISSUE_NUM" >> "$LOGFILE" 2>&1 &
          log "started dev-agent PID $! for issue #${ISSUE_NUM} (post-crash recovery)"
          BLOCKED_BY_INPROGRESS=true
        fi
      fi
    else
      log "issue #${ISSUE_NUM} assigned to ${assignee} — their thread, not blocking"
--- a/docker-compose.yml
+++ b/docker-compose.yml
@ -14,10 +14,10 @@ services:
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
-      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
+      - ${HOME}/.claude.json:/home/agent/.claude.json:ro
-      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
+      - CLAUDE_BIN_PLACEHOLDER:/usr/local/bin/claude:ro
-      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
+      - ${HOME}/.ssh:/home/agent/.ssh:ro
-      - ${SOPS_AGE_DIR:-${HOME}/.config/sops/age}:/home/agent/.config/sops/age:ro
+      - ${HOME}/.config/sops/age:/home/agent/.config/sops/age:ro
      - woodpecker-data:/woodpecker-data:ro
    environment:
      - FORGE_URL=http://forgejo:3000
@ -30,7 +30,6 @@ services:
      - FORGE_SUPERVISOR_TOKEN=${FORGE_SUPERVISOR_TOKEN:-}
      - FORGE_PREDICTOR_TOKEN=${FORGE_PREDICTOR_TOKEN:-}
      - FORGE_ARCHITECT_TOKEN=${FORGE_ARCHITECT_TOKEN:-}
      - FORGE_FILER_TOKEN=${FORGE_FILER_TOKEN:-}
      - FORGE_BOT_USERNAMES=${FORGE_BOT_USERNAMES:-}
      - WOODPECKER_TOKEN=${WOODPECKER_TOKEN:-}
      - CLAUDE_TIMEOUT=${CLAUDE_TIMEOUT:-7200}
@ -49,18 +48,9 @@ services:
      - GARDENER_INTERVAL=${GARDENER_INTERVAL:-21600}
      - ARCHITECT_INTERVAL=${ARCHITECT_INTERVAL:-21600}
      - PLANNER_INTERVAL=${PLANNER_INTERVAL:-43200}
      - SUPERVISOR_INTERVAL=${SUPERVISOR_INTERVAL:-1200}
    healthcheck:
      test: ["CMD", "pgrep", "-f", "entrypoint.sh"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 30s
    depends_on:
-      forgejo:
+      - forgejo
-        condition: service_healthy
+      - woodpecker
      woodpecker:
        condition: service_started
    networks:
      - disinto-net
@ -77,10 +67,10 @@ services:
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
-      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
+      - ${HOME}/.claude.json:/home/agent/.claude.json:ro
-      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
+      - CLAUDE_BIN_PLACEHOLDER:/usr/local/bin/claude:ro
-      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
+      - ${HOME}/.ssh:/home/agent/.ssh:ro
-      - ${SOPS_AGE_DIR:-${HOME}/.config/sops/age}:/home/agent/.config/sops/age:ro
+      - ${HOME}/.config/sops/age:/home/agent/.config/sops/age:ro
      - woodpecker-data:/woodpecker-data:ro
    environment:
      - FORGE_URL=http://forgejo:3000
@ -110,85 +100,9 @@ services:
      - CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}
      - POLL_INTERVAL=${POLL_INTERVAL:-300}
      - AGENT_ROLES=dev
    healthcheck:
      test: ["CMD", "pgrep", "-f", "entrypoint.sh"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 30s
    depends_on:
-      forgejo:
+      - forgejo
-        condition: service_healthy
+      - woodpecker
      woodpecker:
        condition: service_started
    networks:
      - disinto-net
  agents-llama-all:
    build:
      context: .
      dockerfile: docker/agents/Dockerfile
    image: disinto/agents-llama:latest
    container_name: disinto-agents-llama-all
    restart: unless-stopped
    profiles: ["agents-llama-all"]
    security_opt:
      - apparmor=unconfined
    volumes:
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
      - ${SOPS_AGE_DIR:-${HOME}/.config/sops/age}:/home/agent/.config/sops/age:ro
      - woodpecker-data:/woodpecker-data:ro
    environment:
      - FORGE_URL=http://forgejo:3000
      - FORGE_REPO=${FORGE_REPO:-disinto-admin/disinto}
      - FORGE_TOKEN=${FORGE_TOKEN_LLAMA:-}
      - FORGE_PASS=${FORGE_PASS_LLAMA:-}
      - FORGE_REVIEW_TOKEN=${FORGE_REVIEW_TOKEN:-}
      - FORGE_PLANNER_TOKEN=${FORGE_PLANNER_TOKEN:-}
      - FORGE_GARDENER_TOKEN=${FORGE_GARDENER_TOKEN:-}
      - FORGE_VAULT_TOKEN=${FORGE_VAULT_TOKEN:-}
      - FORGE_SUPERVISOR_TOKEN=${FORGE_SUPERVISOR_TOKEN:-}
      - FORGE_PREDICTOR_TOKEN=${FORGE_PREDICTOR_TOKEN:-}
      - FORGE_ARCHITECT_TOKEN=${FORGE_ARCHITECT_TOKEN:-}
      - FORGE_FILER_TOKEN=${FORGE_FILER_TOKEN:-}
      - FORGE_BOT_USERNAMES=${FORGE_BOT_USERNAMES:-}
      - WOODPECKER_TOKEN=${WOODPECKER_TOKEN:-}
      - CLAUDE_TIMEOUT=${CLAUDE_TIMEOUT:-7200}
      - CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC=${CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC:-1}
      - CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=60
      - CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1
      - ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY:-}
      - ANTHROPIC_BASE_URL=${ANTHROPIC_BASE_URL:-}
      - FORGE_ADMIN_PASS=${FORGE_ADMIN_PASS:-}
      - DISINTO_CONTAINER=1
      - PROJECT_TOML=projects/disinto.toml
      - PROJECT_NAME=${PROJECT_NAME:-project}
      - PROJECT_REPO_ROOT=/home/agent/repos/${PROJECT_NAME:-project}
      - WOODPECKER_DATA_DIR=/woodpecker-data
      - WOODPECKER_REPO_ID=${WOODPECKER_REPO_ID:-}
      - CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}
      - POLL_INTERVAL=${POLL_INTERVAL:-300}
      - GARDENER_INTERVAL=${GARDENER_INTERVAL:-21600}
      - ARCHITECT_INTERVAL=${ARCHITECT_INTERVAL:-21600}
      - PLANNER_INTERVAL=${PLANNER_INTERVAL:-43200}
      - SUPERVISOR_INTERVAL=${SUPERVISOR_INTERVAL:-1200}
      - AGENT_ROLES=review,dev,gardener,architect,planner,predictor,supervisor
    healthcheck:
      test: ["CMD", "pgrep", "-f", "entrypoint.sh"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 30s
    depends_on:
      forgejo:
        condition: service_healthy
      woodpecker:
        condition: service_started
    networks:
      - disinto-net
@ -203,9 +117,9 @@ services:
      - /var/run/docker.sock:/var/run/docker.sock
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
-      - ${CLAUDE_DIR:-${HOME}/.claude}:/home/agent/.claude
+      - ${HOME}/.claude:/home/agent/.claude
-      - ${CLAUDE_BIN_DIR:-/usr/local/bin/claude}:/usr/local/bin/claude:ro
+      - /usr/local/bin/claude:/usr/local/bin/claude:ro
-      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
+      - ${HOME}/.ssh:/home/agent/.ssh:ro
    env_file:
      - .env
@ -219,9 +133,9 @@ services:
      - apparmor=unconfined
    volumes:
      - /var/run/docker.sock:/var/run/docker.sock
-      - ${CLAUDE_BIN_DIR:-/usr/local/bin/claude}:/usr/local/bin/claude:ro
+      - /usr/local/bin/claude:/usr/local/bin/claude:ro
-      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/root/.claude.json:ro
+      - ${HOME}/.claude.json:/root/.claude.json:ro
-      - ${CLAUDE_DIR:-${HOME}/.claude}:/root/.claude:ro
+      - ${HOME}/.claude:/root/.claude:ro
      - disinto-logs:/opt/disinto-logs
    environment:
      - FORGE_SUPERVISOR_TOKEN=${FORGE_SUPERVISOR_TOKEN:-}
@ -237,12 +151,6 @@ services:
    ports:
      - "80:80"
      - "443:443"
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:2019/config/"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 15s
    depends_on:
      - forgejo
    networks:
@ -263,12 +171,6 @@ services:
      - FORGEJO__security__INSTALL_LOCK=true
      - FORGEJO__service__DISABLE_REGISTRATION=true
      - FORGEJO__webhook__ALLOWED_HOST_LIST=private
    healthcheck:
      test: ["CMD", "curl", "-sf", "http://localhost:3000/api/v1/version"]
      interval: 5s
      timeout: 3s
      retries: 30
      start_period: 30s
    ports:
      - "3000:3000"
    networks:
--- a/docker/agents/Dockerfile
+++ b/docker/agents/Dockerfile
@ -28,9 +28,6 @@ RUN chmod +x /entrypoint.sh
 # Entrypoint runs polling loop directly, dropping to agent user via gosu.
 # All scripts execute as the agent user (UID 1000) while preserving env vars.
 VOLUME /home/agent/data
 VOLUME /home/agent/repos
 WORKDIR /home/agent/disinto
 ENTRYPOINT ["/entrypoint.sh"]
--- a/docker/agents/entrypoint.sh
+++ b/docker/agents/entrypoint.sh
@ -7,15 +7,14 @@ set -euo pipefail
 # poll scripts.  All Docker Compose env vars are inherited (PATH, FORGE_TOKEN,
 # ANTHROPIC_API_KEY, etc.).
 #
-# AGENT_ROLES env var controls which scripts run: "review,dev,gardener,architect,planner,predictor,supervisor"
+# AGENT_ROLES env var controls which scripts run: "review,dev,gardener,architect,planner,predictor"
-# (default: all seven). Uses while-true loop with staggered intervals:
+# (default: all six). Uses while-true loop with staggered intervals:
 #   - review-poll: every 5 minutes (offset by 0s)
 #   - dev-poll: every 5 minutes (offset by 2 minutes)
 #   - gardener: every GARDENER_INTERVAL seconds (default: 21600 = 6 hours)
 #   - architect: every ARCHITECT_INTERVAL seconds (default: 21600 = 6 hours)
 #   - planner: every PLANNER_INTERVAL seconds (default: 43200 = 12 hours)
 #   - predictor: every 24 hours (288 iterations * 5 min)
 #   - supervisor: every SUPERVISOR_INTERVAL seconds (default: 1200 = 20 min)
 DISINTO_BAKED="/home/agent/disinto"
 DISINTO_LIVE="/home/agent/repos/_factory"
@ -50,7 +49,7 @@ source "${DISINTO_BAKED}/lib/git-creds.sh"
 # Wrapper that calls the shared configure_git_creds with agent-specific paths,
 # then repairs any legacy baked-credential URLs in existing clones.
 _setup_git_creds() {
-  _GIT_CREDS_LOG_FN=log configure_git_creds "/home/agent" "gosu agent"
+  configure_git_creds "/home/agent" "gosu agent"
  if [ -n "${FORGE_PASS:-}" ] && [ -n "${FORGE_URL:-}" ]; then
    log "Git credential helper configured (password auth)"
  fi
@ -62,21 +61,16 @@ _setup_git_creds() {
 # Configure git author identity for commits made by this container.
 # Derives identity from the resolved bot user (BOT_USER) to ensure commits
 # are visibly attributable to the correct bot in the forge timeline.
 # BOT_USER is normally set by configure_git_creds() (#741); this function
 # only falls back to its own API call if BOT_USER was not already resolved.
 configure_git_identity() {
-  # Resolve BOT_USER from FORGE_TOKEN if not already set (configure_git_creds
+  # Resolve BOT_USER from FORGE_TOKEN if not already set
  # exports BOT_USER on success, so this is a fallback for edge cases only).
  if [ -z "${BOT_USER:-}" ] && [ -n "${FORGE_TOKEN:-}" ]; then
    BOT_USER=$(curl -sf --max-time 10 \
      -H "Authorization: token ${FORGE_TOKEN}" \
      "${FORGE_URL:-http://localhost:3000}/api/v1/user" 2>/dev/null | jq -r '.login // empty') || true
  fi
-  if [ -z "${BOT_USER:-}" ]; then
+  # Default to dev-bot if resolution fails
-    log "WARNING: Could not resolve bot username for git identity — commits will use fallback"
+  BOT_USER="${BOT_USER:-dev-bot}"
    BOT_USER="agent"
  fi
  # Configure git identity for all repositories
  gosu agent git config --global user.name "${BOT_USER}"
@ -329,7 +323,7 @@ init_state_dir
 # Parse AGENT_ROLES env var (default: all agents)
 # Expected format: comma-separated list like "review,dev,gardener"
-AGENT_ROLES="${AGENT_ROLES:-review,dev,gardener,architect,planner,predictor,supervisor}"
+AGENT_ROLES="${AGENT_ROLES:-review,dev,gardener,architect,planner,predictor}"
 log "Agent roles configured: ${AGENT_ROLES}"
 # Poll interval in seconds (5 minutes default)
@ -339,10 +333,9 @@ POLL_INTERVAL="${POLL_INTERVAL:-300}"
 GARDENER_INTERVAL="${GARDENER_INTERVAL:-21600}"
 ARCHITECT_INTERVAL="${ARCHITECT_INTERVAL:-21600}"
 PLANNER_INTERVAL="${PLANNER_INTERVAL:-43200}"
 SUPERVISOR_INTERVAL="${SUPERVISOR_INTERVAL:-1200}"
 log "Entering polling loop (interval: ${POLL_INTERVAL}s, roles: ${AGENT_ROLES})"
-log "Gardener interval: ${GARDENER_INTERVAL}s, Architect interval: ${ARCHITECT_INTERVAL}s, Planner interval: ${PLANNER_INTERVAL}s, Supervisor interval: ${SUPERVISOR_INTERVAL}s"
+log "Gardener interval: ${GARDENER_INTERVAL}s, Architect interval: ${ARCHITECT_INTERVAL}s, Planner interval: ${PLANNER_INTERVAL}s"
 # Main polling loop using iteration counter for gardener scheduling
 iteration=0
@ -387,13 +380,11 @@ print(cfg.get('primary_branch', 'main'))
    log "Processing project TOML: ${toml}"
    # --- Fast agents: run in background, wait before slow agents ---
    FAST_PIDS=()
    # Review poll (every iteration)
    if [[ ",${AGENT_ROLES}," == *",review,"* ]]; then
      log "Running review-poll (iteration ${iteration}) for ${toml}"
      gosu agent bash -c "cd ${DISINTO_DIR} && bash review/review-poll.sh \"${toml}\"" >> "${DISINTO_LOG_DIR}/review-poll.log" 2>&1 &
      FAST_PIDS+=($!)
    fi
    sleep 2  # stagger fast polls
@ -402,14 +393,10 @@ print(cfg.get('primary_branch', 'main'))
    if [[ ",${AGENT_ROLES}," == *",dev,"* ]]; then
      log "Running dev-poll (iteration ${iteration}) for ${toml}"
      gosu agent bash -c "cd ${DISINTO_DIR} && bash dev/dev-poll.sh \"${toml}\"" >> "${DISINTO_LOG_DIR}/dev-poll.log" 2>&1 &
      FAST_PIDS+=($!)
    fi
-    # Wait only for THIS iteration's fast polls — long-running gardener/dev-agent
+    # Wait for fast polls to finish before launching slow agents
-    # from prior iterations must not block us.
+    wait
    if [ ${#FAST_PIDS[@]} -gt 0 ]; then
      wait "${FAST_PIDS[@]}"
    fi
    # --- Slow agents: run in background with pgrep guard ---
@ -465,19 +452,6 @@ print(cfg.get('primary_branch', 'main'))
        fi
      fi
    fi
    # Supervisor (interval configurable via SUPERVISOR_INTERVAL env var, default 20 min)
    if [[ ",${AGENT_ROLES}," == *",supervisor,"* ]]; then
      supervisor_iteration=$((iteration * POLL_INTERVAL))
      if [ $((supervisor_iteration % SUPERVISOR_INTERVAL)) -eq 0 ] && [ "$now" -ge "$supervisor_iteration" ]; then
        if ! pgrep -f "supervisor-run.sh" >/dev/null; then
          log "Running supervisor (iteration ${iteration}, ${SUPERVISOR_INTERVAL}s interval) for ${toml}"
          gosu agent bash -c "cd ${DISINTO_DIR} && bash supervisor/supervisor-run.sh \"${toml}\"" >> "${DISINTO_LOG_DIR}/supervisor.log" 2>&1 &
        else
          log "Skipping supervisor — already running"
        fi
      fi
    fi
  done
  sleep "${POLL_INTERVAL}"
--- a/docker/chat/Dockerfile
+++ b/docker/chat/Dockerfile
@ -15,7 +15,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
    python3 \
    && rm -rf /var/lib/apt/lists/*
-# Non-root user — fixed UID 10001 for sandbox hardening (#706)
+# Non-root user — fixed UID 10001 for sandbox hardening (#706, #707)
 RUN useradd -m -u 10001 -s /bin/bash chat
 # Copy application files
@ -25,11 +25,18 @@ COPY ui/ /var/chat/ui/
 RUN chmod +x /entrypoint-chat.sh /usr/local/bin/server.py
 # Create and set ownership of chat identity directory for #707
 RUN install -d -m 0700 /home/chat/.claude-chat/config/credentials \
    && chown -R chat:chat /home/chat/.claude-chat
 USER chat
 WORKDIR /var/chat
 # Declare volume for chat identity — mounted from host at runtime (#707)
 VOLUME /home/chat/.claude-chat
 EXPOSE 8080
 HEALTHCHECK --interval=30s --timeout=5s --start-period=10s --retries=3 \
-  CMD python3 -c "import urllib.request; urllib.request.urlopen('http://localhost:8080/health')" || exit 1
+  CMD python3 -c "import urllib.request; urllib.request.urlopen('http://localhost:8080/')" || exit 1
 ENTRYPOINT ["/entrypoint-chat.sh"]
--- a/docker/chat/server.py
+++ b/docker/chat/server.py
@ -3,16 +3,15 @@
 disinto-chat server — minimal HTTP backend for Claude chat UI.
 Routes:
-    GET /chat/auth/verify    -> Caddy forward_auth callback (returns 200+X-Forwarded-User or 401)
+    GET /chat/login          → 302 to Forgejo OAuth authorize
-    GET /chat/login          -> 302 to Forgejo OAuth authorize
+    GET /chat/oauth/callback → exchange code for token, validate user, set session
-    GET /chat/oauth/callback -> exchange code for token, validate user, set session
+    GET /chat/               → serves index.html (session required)
-    GET /chat/               -> serves index.html (session required)
+    GET /chat/static/*       → serves static assets (session required)
-    GET /chat/static/*       -> serves static assets (session required)
+    POST /chat               → spawns `claude --print` with user message (session required)
-    POST /chat               -> spawns `claude --print` with user message (session required)
+    GET /ws                  → reserved for future streaming upgrade (returns 501)
    GET /ws                  -> reserved for future streaming upgrade (returns 501)
 OAuth flow:
-    1. User hits any /chat/* route without a valid session cookie -> 302 /chat/login
+    1. User hits any /chat/* route without a valid session cookie → 302 /chat/login
    2. /chat/login redirects to Forgejo /login/oauth/authorize
    3. Forgejo redirects back to /chat/oauth/callback with ?code=...&state=...
    4. Server exchanges code for access token, fetches /api/v1/user
@ -22,10 +21,8 @@ OAuth flow:
 The claude binary is expected to be mounted from the host at /usr/local/bin/claude.
 """
 import datetime
 import json
 import os
 import re
 import secrets
 import subprocess
 import sys
@ -46,18 +43,7 @@ CHAT_OAUTH_CLIENT_ID = os.environ.get("CHAT_OAUTH_CLIENT_ID", "")
 CHAT_OAUTH_CLIENT_SECRET = os.environ.get("CHAT_OAUTH_CLIENT_SECRET", "")
 EDGE_TUNNEL_FQDN = os.environ.get("EDGE_TUNNEL_FQDN", "")
-# Shared secret for Caddy forward_auth verify endpoint (#709).
+# Allowed users — disinto-admin always allowed; CSV allowlist extends it
 # When set, only requests carrying this value in X-Forward-Auth-Secret are
 # allowed to call /chat/auth/verify.  When empty the endpoint is unrestricted
 # (acceptable during local dev; production MUST set this).
 FORWARD_AUTH_SECRET = os.environ.get("FORWARD_AUTH_SECRET", "")
 # Rate limiting / cost caps (#711)
 CHAT_MAX_REQUESTS_PER_HOUR = int(os.environ.get("CHAT_MAX_REQUESTS_PER_HOUR", 60))
 CHAT_MAX_REQUESTS_PER_DAY = int(os.environ.get("CHAT_MAX_REQUESTS_PER_DAY", 500))
 CHAT_MAX_TOKENS_PER_DAY = int(os.environ.get("CHAT_MAX_TOKENS_PER_DAY", 1000000))
 # Allowed users - disinto-admin always allowed; CSV allowlist extends it
 _allowed_csv = os.environ.get("DISINTO_CHAT_ALLOWED_USERS", "")
 ALLOWED_USERS = {"disinto-admin"}
 if _allowed_csv:
@ -69,24 +55,12 @@ SESSION_COOKIE = "disinto_chat_session"
 # Session TTL: 24 hours
 SESSION_TTL = 24 * 60 * 60
-# Chat history directory (bind-mounted from host)
+# In-memory session store: token → {"user": str, "expires": float}
 CHAT_HISTORY_DIR = os.environ.get("CHAT_HISTORY_DIR", "/var/lib/chat/history")
 # Regex for valid conversation_id (12-char hex, no slashes)
 CONVERSATION_ID_PATTERN = re.compile(r"^[0-9a-f]{12}$")
 # In-memory session store: token -> {"user": str, "expires": float}
 _sessions = {}
-# Pending OAuth state tokens: state -> expires (float)
+# Pending OAuth state tokens: state → expires (float)
 _oauth_states = {}
 # Per-user rate limiting state (#711)
 # user -> list of request timestamps (for sliding-window hourly/daily caps)
 _request_log = {}
 # user -> {"tokens": int, "date": "YYYY-MM-DD"}
 _daily_tokens = {}
 # MIME types for static files
 MIME_TYPES = {
    ".html": "text/html; charset=utf-8",
@ -126,7 +100,7 @@ def _validate_session(cookie_header):
            session = _sessions.get(token)
            if session and session["expires"] > time.time():
                return session["user"]
-            # Expired - clean up
+            # Expired — clean up
            _sessions.pop(token, None)
            return None
    return None
@ -187,242 +161,6 @@ def _fetch_user(access_token):
        return None
 # =============================================================================
 # Rate Limiting Functions (#711)
 # =============================================================================
 def _check_rate_limit(user):
    """Check per-user rate limits. Returns (allowed, retry_after, reason) (#711).
    Checks hourly request cap, daily request cap, and daily token cap.
    """
    now = time.time()
    one_hour_ago = now - 3600
    today = datetime.date.today().isoformat()
    # Prune old entries from request log
    timestamps = _request_log.get(user, [])
    timestamps = [t for t in timestamps if t > now - 86400]
    _request_log[user] = timestamps
    # Hourly request cap
    hourly = [t for t in timestamps if t > one_hour_ago]
    if len(hourly) >= CHAT_MAX_REQUESTS_PER_HOUR:
        oldest_in_window = min(hourly)
        retry_after = int(oldest_in_window + 3600 - now) + 1
        return False, max(retry_after, 1), "hourly request limit"
    # Daily request cap
    start_of_day = time.mktime(datetime.date.today().timetuple())
    daily = [t for t in timestamps if t >= start_of_day]
    if len(daily) >= CHAT_MAX_REQUESTS_PER_DAY:
        next_day = start_of_day + 86400
        retry_after = int(next_day - now) + 1
        return False, max(retry_after, 1), "daily request limit"
    # Daily token cap
    token_info = _daily_tokens.get(user, {"tokens": 0, "date": today})
    if token_info["date"] != today:
        token_info = {"tokens": 0, "date": today}
        _daily_tokens[user] = token_info
    if token_info["tokens"] >= CHAT_MAX_TOKENS_PER_DAY:
        next_day = start_of_day + 86400
        retry_after = int(next_day - now) + 1
        return False, max(retry_after, 1), "daily token limit"
    return True, 0, ""
 def _record_request(user):
    """Record a request timestamp for the user (#711)."""
    _request_log.setdefault(user, []).append(time.time())
 def _record_tokens(user, tokens):
    """Record token usage for the user (#711)."""
    today = datetime.date.today().isoformat()
    token_info = _daily_tokens.get(user, {"tokens": 0, "date": today})
    if token_info["date"] != today:
        token_info = {"tokens": 0, "date": today}
    token_info["tokens"] += tokens
    _daily_tokens[user] = token_info
 def _parse_stream_json(output):
    """Parse stream-json output from claude --print (#711).
    Returns (text_content, total_tokens).  Falls back gracefully if the
    usage event is absent or malformed.
    """
    text_parts = []
    total_tokens = 0
    for line in output.splitlines():
        line = line.strip()
        if not line:
            continue
        try:
            event = json.loads(line)
        except json.JSONDecodeError:
            continue
        etype = event.get("type", "")
        # Collect assistant text
        if etype == "content_block_delta":
            delta = event.get("delta", {})
            if delta.get("type") == "text_delta":
                text_parts.append(delta.get("text", ""))
        elif etype == "assistant":
            # Full assistant message (non-streaming)
            content = event.get("content", "")
            if isinstance(content, str) and content:
                text_parts.append(content)
            elif isinstance(content, list):
                for block in content:
                    if isinstance(block, dict) and block.get("text"):
                        text_parts.append(block["text"])
        # Parse usage from result event
        if etype == "result":
            usage = event.get("usage", {})
            total_tokens = usage.get("input_tokens", 0) + usage.get("output_tokens", 0)
        elif "usage" in event:
            usage = event["usage"]
            if isinstance(usage, dict):
                total_tokens = usage.get("input_tokens", 0) + usage.get("output_tokens", 0)
    return "".join(text_parts), total_tokens
 # =============================================================================
 # Conversation History Functions (#710)
 # =============================================================================
 def _generate_conversation_id():
    """Generate a new conversation ID (12-char hex string)."""
    return secrets.token_hex(6)
 def _validate_conversation_id(conv_id):
    """Validate that conversation_id matches the required format."""
    return bool(CONVERSATION_ID_PATTERN.match(conv_id))
 def _get_user_history_dir(user):
    """Get the history directory path for a user."""
    return os.path.join(CHAT_HISTORY_DIR, user)
 def _get_conversation_path(user, conv_id):
    """Get the full path to a conversation file."""
    user_dir = _get_user_history_dir(user)
    return os.path.join(user_dir, f"{conv_id}.ndjson")
 def _ensure_user_dir(user):
    """Ensure the user's history directory exists."""
    user_dir = _get_user_history_dir(user)
    os.makedirs(user_dir, exist_ok=True)
    return user_dir
 def _write_message(user, conv_id, role, content):
    """Append a message to a conversation file in NDJSON format."""
    conv_path = _get_conversation_path(user, conv_id)
    _ensure_user_dir(user)
    record = {
        "ts": time.strftime("%Y-%m-%dT%H:%M:%SZ", time.gmtime()),
        "user": user,
        "role": role,
        "content": content,
    }
    with open(conv_path, "a", encoding="utf-8") as f:
        f.write(json.dumps(record, ensure_ascii=False) + "\n")
 def _read_conversation(user, conv_id):
    """Read all messages from a conversation file."""
    conv_path = _get_conversation_path(user, conv_id)
    messages = []
    if not os.path.exists(conv_path):
        return None
    try:
        with open(conv_path, "r", encoding="utf-8") as f:
            for line in f:
                line = line.strip()
                if line:
                    try:
                        messages.append(json.loads(line))
                    except json.JSONDecodeError:
                        # Skip malformed lines
                        continue
    except IOError:
        return None
    return messages
 def _list_user_conversations(user):
    """List all conversation files for a user with first message preview."""
    user_dir = _get_user_history_dir(user)
    conversations = []
    if not os.path.exists(user_dir):
        return conversations
    try:
        for filename in os.listdir(user_dir):
            if not filename.endswith(".ndjson"):
                continue
            conv_id = filename[:-7]  # Remove .ndjson extension
            if not _validate_conversation_id(conv_id):
                continue
            conv_path = os.path.join(user_dir, filename)
            messages = _read_conversation(user, conv_id)
            if messages:
                first_msg = messages[0]
                preview = first_msg.get("content", "")[:50]
                if len(first_msg.get("content", "")) > 50:
                    preview += "..."
                conversations.append({
                    "id": conv_id,
                    "created_at": first_msg.get("ts", ""),
                    "preview": preview,
                    "message_count": len(messages),
                })
            else:
                # Empty conversation file
                conversations.append({
                    "id": conv_id,
                    "created_at": "",
                    "preview": "(empty)",
                    "message_count": 0,
                })
    except OSError:
        pass
    # Sort by created_at descending
    conversations.sort(key=lambda x: x["created_at"] or "", reverse=True)
    return conversations
 def _delete_conversation(user, conv_id):
    """Delete a conversation file."""
    conv_path = _get_conversation_path(user, conv_id)
    if os.path.exists(conv_path):
        os.remove(conv_path)
        return True
    return False
 class ChatHandler(BaseHTTPRequestHandler):
    """HTTP request handler for disinto-chat with Forgejo OAuth."""
@ -448,52 +186,11 @@ class ChatHandler(BaseHTTPRequestHandler):
        self.end_headers()
        return None
    def _check_forwarded_user(self, session_user):
        """Defense-in-depth: verify X-Forwarded-User matches session user (#709).
        Returns True if the request may proceed, False if a 403 was sent.
        When X-Forwarded-User is absent (forward_auth removed from Caddy),
        the request is rejected - fail-closed by design.
        """
        forwarded = self.headers.get("X-Forwarded-User")
        if not forwarded:
            rid = self.headers.get("X-Request-Id", "-")
            print(
                f"WARN: missing X-Forwarded-User for session_user={session_user} "
                f"req_id={rid} - fail-closed (#709)",
                file=sys.stderr,
            )
            self.send_error_page(403, "Forbidden: missing forwarded-user header")
            return False
        if forwarded != session_user:
            rid = self.headers.get("X-Request-Id", "-")
            print(
                f"WARN: X-Forwarded-User mismatch: header={forwarded} "
                f"session={session_user} req_id={rid} (#709)",
                file=sys.stderr,
            )
            self.send_error_page(403, "Forbidden: user identity mismatch")
            return False
        return True
    def do_GET(self):
        """Handle GET requests."""
        parsed = urlparse(self.path)
        path = parsed.path
        # Health endpoint (no auth required) — used by Docker healthcheck
        if path == "/health":
            self.send_response(200)
            self.send_header("Content-Type", "text/plain")
            self.end_headers()
            self.wfile.write(b"ok\n")
            return
        # Verify endpoint for Caddy forward_auth (#709)
        if path == "/chat/auth/verify":
            self.handle_auth_verify()
            return
        # OAuth routes (no session required)
        if path == "/chat/login":
            self.handle_login()
@ -503,43 +200,16 @@ class ChatHandler(BaseHTTPRequestHandler):
            self.handle_oauth_callback(parsed.query)
            return
        # Conversation list endpoint: GET /chat/history
        if path == "/chat/history":
            user = self._require_session()
            if not user:
                return
            if not self._check_forwarded_user(user):
                return
            self.handle_conversation_list(user)
            return
        # Single conversation endpoint: GET /chat/history/<id>
        if path.startswith("/chat/history/"):
            user = self._require_session()
            if not user:
                return
            if not self._check_forwarded_user(user):
                return
            conv_id = path[len("/chat/history/"):]
            self.handle_conversation_get(user, conv_id)
            return
        # Serve index.html at root
        if path in ("/", "/chat", "/chat/"):
-            user = self._require_session()
+            if not self._require_session():
            if not user:
                return
            if not self._check_forwarded_user(user):
                return
            self.serve_index()
            return
        # Serve static files
        if path.startswith("/chat/static/") or path.startswith("/static/"):
-            user = self._require_session()
+            if not self._require_session():
            if not user:
                return
            if not self._check_forwarded_user(user):
                return
            self.serve_static(path)
            return
@ -557,59 +227,16 @@ class ChatHandler(BaseHTTPRequestHandler):
        parsed = urlparse(self.path)
        path = parsed.path
        # New conversation endpoint (session required)
        if path == "/chat/new":
            user = self._require_session()
            if not user:
                return
            if not self._check_forwarded_user(user):
                return
            self.handle_new_conversation(user)
            return
        # Chat endpoint (session required)
        if path in ("/chat", "/chat/"):
-            user = self._require_session()
+            if not self._require_session():
            if not user:
                return
-            if not self._check_forwarded_user(user):
+            self.handle_chat()
                return
            self.handle_chat(user)
            return
        # 404 for unknown paths
        self.send_error_page(404, "Not found")
    def handle_auth_verify(self):
        """Caddy forward_auth callback - validate session and return X-Forwarded-User (#709).
        Caddy calls this endpoint for every /chat/* request.  If the session
        cookie is valid the endpoint returns 200 with the X-Forwarded-User
        header set to the session username.  Otherwise it returns 401 so Caddy
        knows the request is unauthenticated.
        Access control: when FORWARD_AUTH_SECRET is configured, the request must
        carry a matching X-Forward-Auth-Secret header (shared secret between
        Caddy and the chat backend).
        """
        # Shared-secret gate
        if FORWARD_AUTH_SECRET:
            provided = self.headers.get("X-Forward-Auth-Secret", "")
            if not secrets.compare_digest(provided, FORWARD_AUTH_SECRET):
                self.send_error_page(403, "Forbidden: invalid forward-auth secret")
                return
        user = _validate_session(self.headers.get("Cookie"))
        if not user:
            self.send_error_page(401, "Unauthorized: no valid session")
            return
        self.send_response(200)
        self.send_header("X-Forwarded-User", user)
        self.send_header("Content-Type", "text/plain; charset=utf-8")
        self.end_headers()
        self.wfile.write(b"ok")
    def handle_login(self):
        """Redirect to Forgejo OAuth authorize endpoint."""
        _gc_sessions()
@ -736,33 +363,11 @@ class ChatHandler(BaseHTTPRequestHandler):
        except IOError as e:
            self.send_error_page(500, f"Error reading file: {e}")
-    def _send_rate_limit_response(self, retry_after, reason):
+    def handle_chat(self):
        """Send a 429 response with Retry-After header and HTMX fragment (#711)."""
        body = (
            f'<div class="rate-limit-error">'
            f"Rate limit exceeded: {reason}. "
            f"Please try again in {retry_after} seconds."
            f"</div>"
        )
        self.send_response(429)
        self.send_header("Retry-After", str(retry_after))
        self.send_header("Content-Type", "text/html; charset=utf-8")
        self.send_header("Content-Length", str(len(body.encode("utf-8"))))
        self.end_headers()
        self.wfile.write(body.encode("utf-8"))
    def handle_chat(self, user):
        """
        Handle chat requests by spawning `claude --print` with the user message.
-        Enforces per-user rate limits and tracks token usage (#711).
+        Returns the response as plain text.
        """
        # Check rate limits before processing (#711)
        allowed, retry_after, reason = _check_rate_limit(user)
        if not allowed:
            self._send_rate_limit_response(retry_after, reason)
            return
        # Read request body
        content_length = int(self.headers.get("Content-Length", 0))
        if content_length == 0:
@ -775,7 +380,6 @@ class ChatHandler(BaseHTTPRequestHandler):
            body_str = body.decode("utf-8")
            params = parse_qs(body_str)
            message = params.get("message", [""])[0]
            conv_id = params.get("conversation_id", [None])[0]
        except (UnicodeDecodeError, KeyError):
            self.send_error_page(400, "Invalid message format")
            return
@ -784,150 +388,46 @@ class ChatHandler(BaseHTTPRequestHandler):
            self.send_error_page(400, "Empty message")
            return
        # Get user from session
        user = _validate_session(self.headers.get("Cookie"))
        if not user:
            self.send_error_page(401, "Unauthorized")
            return
        # Validate Claude binary exists
        if not os.path.exists(CLAUDE_BIN):
            self.send_error_page(500, "Claude CLI not found")
            return
        # Generate new conversation ID if not provided
        if not conv_id or not _validate_conversation_id(conv_id):
            conv_id = _generate_conversation_id()
        # Record request for rate limiting (#711)
        _record_request(user)
        try:
-            # Save user message to history
+            # Spawn claude --print with text output format
            _write_message(user, conv_id, "user", message)
            # Spawn claude --print with stream-json for token tracking (#711)
            proc = subprocess.Popen(
-                [CLAUDE_BIN, "--print", "--output-format", "stream-json", message],
+                [CLAUDE_BIN, "--print", message],
                stdout=subprocess.PIPE,
                stderr=subprocess.PIPE,
                text=True,
            )
-            raw_output = proc.stdout.read()
+            # Read response as text (Claude outputs plain text when not using stream-json)
            response = proc.stdout.read()
            # Read stderr (should be minimal, mostly for debugging)
            error_output = proc.stderr.read()
            if error_output:
                print(f"Claude stderr: {error_output}", file=sys.stderr)
            # Wait for process to complete
            proc.wait()
            # Check for errors
            if proc.returncode != 0:
                self.send_error_page(500, f"Claude CLI failed with exit code {proc.returncode}")
                return
            # Parse stream-json for text and token usage (#711)
            response, total_tokens = _parse_stream_json(raw_output)
            # Track token usage - does not block *this* request (#711)
            if total_tokens > 0:
                _record_tokens(user, total_tokens)
                print(
                    f"Token usage: user={user} tokens={total_tokens}",
                    file=sys.stderr,
                )
            # Fall back to raw output if stream-json parsing yielded no text
            if not response:
                response = raw_output
            # Save assistant response to history
            _write_message(user, conv_id, "assistant", response)
            self.send_response(200)
-            self.send_header("Content-Type", "application/json; charset=utf-8")
+            self.send_header("Content-Type", "text/plain; charset=utf-8")
            self.send_header("Content-Length", len(response.encode("utf-8")))
            self.end_headers()
-            self.wfile.write(json.dumps({
+            self.wfile.write(response.encode("utf-8"))
                "response": response,
                "conversation_id": conv_id,
            }, ensure_ascii=False).encode("utf-8"))
        except FileNotFoundError:
            self.send_error_page(500, "Claude CLI not found")
        except Exception as e:
            self.send_error_page(500, f"Error: {e}")
    # =======================================================================
    # Conversation History Handlers
    # =======================================================================
    def handle_conversation_list(self, user):
        """List all conversations for the logged-in user."""
        conversations = _list_user_conversations(user)
        self.send_response(200)
        self.send_header("Content-Type", "application/json; charset=utf-8")
        self.end_headers()
        self.wfile.write(json.dumps(conversations, ensure_ascii=False).encode("utf-8"))
    def handle_conversation_get(self, user, conv_id):
        """Get a specific conversation for the logged-in user."""
        # Validate conversation_id format
        if not _validate_conversation_id(conv_id):
            self.send_error_page(400, "Invalid conversation ID")
            return
        messages = _read_conversation(user, conv_id)
        if messages is None:
            self.send_error_page(404, "Conversation not found")
            return
        self.send_response(200)
        self.send_header("Content-Type", "application/json; charset=utf-8")
        self.end_headers()
        self.wfile.write(json.dumps(messages, ensure_ascii=False).encode("utf-8"))
    def handle_conversation_delete(self, user, conv_id):
        """Delete a specific conversation for the logged-in user."""
        # Validate conversation_id format
        if not _validate_conversation_id(conv_id):
            self.send_error_page(400, "Invalid conversation ID")
            return
        if _delete_conversation(user, conv_id):
            self.send_response(204)  # No Content
            self.end_headers()
        else:
            self.send_error_page(404, "Conversation not found")
    def handle_new_conversation(self, user):
        """Create a new conversation and return its ID."""
        conv_id = _generate_conversation_id()
        self.send_response(200)
        self.send_header("Content-Type", "application/json; charset=utf-8")
        self.end_headers()
        self.wfile.write(json.dumps({"conversation_id": conv_id}, ensure_ascii=False).encode("utf-8"))
    def do_DELETE(self):
        """Handle DELETE requests."""
        parsed = urlparse(self.path)
        path = parsed.path
        # Delete conversation endpoint
        if path.startswith("/chat/history/"):
            user = self._require_session()
            if not user:
                return
            if not self._check_forwarded_user(user):
                return
            conv_id = path[len("/chat/history/"):]
            self.handle_conversation_delete(user, conv_id)
            return
        # 404 for unknown paths
        self.send_error_page(404, "Not found")
 def main():
    """Start the HTTP server."""
@ -939,17 +439,7 @@ def main():
        print(f"OAuth enabled (client_id={CHAT_OAUTH_CLIENT_ID[:8]}...)", file=sys.stderr)
        print(f"Allowed users: {', '.join(sorted(ALLOWED_USERS))}", file=sys.stderr)
    else:
-        print("WARNING: CHAT_OAUTH_CLIENT_ID not set - OAuth disabled", file=sys.stderr)
+        print("WARNING: CHAT_OAUTH_CLIENT_ID not set — OAuth disabled", file=sys.stderr)
    if FORWARD_AUTH_SECRET:
        print("forward_auth secret configured (#709)", file=sys.stderr)
    else:
        print("WARNING: FORWARD_AUTH_SECRET not set - verify endpoint unrestricted", file=sys.stderr)
    print(
        f"Rate limits (#711): {CHAT_MAX_REQUESTS_PER_HOUR}/hr, "
        f"{CHAT_MAX_REQUESTS_PER_DAY}/day, "
        f"{CHAT_MAX_TOKENS_PER_DAY} tokens/day",
        file=sys.stderr,
    )
    httpd.serve_forever()
--- a/docker/chat/ui/index.html
+++ b/docker/chat/ui/index.html
@ -17,93 +17,7 @@
            color: #eaeaea;
            min-height: 100vh;
            display: flex;
        }
        /* Sidebar styles */
        .sidebar {
            width: 280px;
            background: #16213e;
            border-right: 1px solid #0f3460;
            display: flex;
            flex-direction: column;
            height: 100vh;
            position: fixed;
            left: 0;
            top: 0;
            z-index: 100;
        }
        .sidebar-header {
            padding: 1rem;
            border-bottom: 1px solid #0f3460;
        }
        .sidebar-header h1 {
            font-size: 1.25rem;
            font-weight: 600;
            margin-bottom: 0.5rem;
        }
        .new-chat-btn {
            width: 100%;
            background: #e94560;
            color: white;
            border: none;
            border-radius: 6px;
            padding: 0.75rem 1rem;
            font-size: 0.9rem;
            font-weight: 600;
            cursor: pointer;
            transition: background 0.2s;
        }
        .new-chat-btn:hover {
            background: #d63447;
        }
        .new-chat-btn:disabled {
            background: #555;
            cursor: not-allowed;
        }
        .conversations-list {
            flex: 1;
            overflow-y: auto;
            padding: 0.5rem;
        }
        .conversation-item {
            padding: 0.75rem 1rem;
            border-radius: 6px;
            cursor: pointer;
            margin-bottom: 0.25rem;
            transition: background 0.2s;
            border: 1px solid transparent;
        }
        .conversation-item:hover {
            background: #1a1a2e;
        }
        .conversation-item.active {
            background: #0f3460;
            border-color: #e94560;
        }
        .conversation-item .preview {
            font-size: 0.875rem;
            white-space: nowrap;
            overflow: hidden;
            text-overflow: ellipsis;
            opacity: 0.9;
        }
        .conversation-item .meta {
            font-size: 0.75rem;
            opacity: 0.6;
            margin-top: 0.25rem;
        }
        .conversation-item .message-count {
            float: right;
            font-size: 0.7rem;
            background: #0f3460;
            padding: 0.125rem 0.5rem;
            border-radius: 10px;
        }
        .main-content {
            margin-left: 280px;
            display: flex;
            flex-direction: column;
            width: 100%;
            height: 100vh;
        }
        header {
            background: #16213e;
@ -205,202 +119,30 @@
        .loading {
            opacity: 0.6;
        }
        .empty-state {
            display: flex;
            flex-direction: column;
            align-items: center;
            justify-content: center;
            height: 100%;
            color: #888;
            text-align: center;
        }
        .empty-state p {
            margin-top: 1rem;
        }
        /* Responsive sidebar toggle */
        .sidebar-toggle {
            display: none;
            position: fixed;
            top: 1rem;
            left: 1rem;
            z-index: 200;
            background: #e94560;
            color: white;
            border: none;
            border-radius: 6px;
            padding: 0.5rem;
            cursor: pointer;
        }
        @media (max-width: 768px) {
            .sidebar {
                transform: translateX(-100%);
                transition: transform 0.3s;
            }
            .sidebar.open {
                transform: translateX(0);
            }
            .sidebar-toggle {
                display: block;
            }
            .main-content {
                margin-left: 0;
            }
        }
    </style>
 </head>
 <body>
-    <button class="sidebar-toggle" id="sidebar-toggle">☰</button>
+    <header>
-    <aside class="sidebar" id="sidebar">
+        <h1>disinto-chat</h1>
-        <div class="sidebar-header">
+    </header>
-            <h1>disinto-chat</h1>
+    <main>
-            <button class="new-chat-btn" id="new-chat-btn">+ New Chat</button>
+        <div id="messages">
-        </div>
+            <div class="message system">
-        <div class="conversations-list" id="conversations-list">
+                <div class="role">system</div>
-            <!-- Conversations will be loaded here -->
+                <div class="content">Welcome to disinto-chat. Type a message to start chatting with Claude.</div>
        </div>
    </aside>
    <div class="main-content">
        <header>
            <h1>disinto-chat</h1>
        </header>
        <main>
            <div id="messages">
                <div class="message system">
                    <div class="role">system</div>
                    <div class="content">Welcome to disinto-chat. Type a message to start chatting with Claude.</div>
                </div>
            </div>
-            <form class="input-area" id="chat-form">
+        </div>
-                <textarea name="message" placeholder="Type your message..." required></textarea>
+        <form class="input-area">
-                <button type="submit" id="send-btn">Send</button>
+            <textarea name="message" placeholder="Type your message..." required></textarea>
-            </form>
+            <button type="submit" id="send-btn">Send</button>
-        </main>
+        </form>
-    </div>
+    </main>
    <script>
        // State
        let currentConversationId = null;
        let conversations = [];
        // DOM elements
        const messagesDiv = document.getElementById('messages');
        const sendBtn = document.getElementById('send-btn');
        const textarea = document.querySelector('textarea');
        const conversationsList = document.getElementById('conversations-list');
        const newChatBtn = document.getElementById('new-chat-btn');
        const sidebar = document.getElementById('sidebar');
        const sidebarToggle = document.getElementById('sidebar-toggle');
        // Load conversations list
        async function loadConversations() {
            try {
                const response = await fetch('/chat/history');
                if (response.ok) {
                    conversations = await response.json();
                    renderConversationsList();
                }
            } catch (error) {
                console.error('Failed to load conversations:', error);
            }
        }
        // Render conversations list
        function renderConversationsList() {
            conversationsList.innerHTML = '';
            if (conversations.length === 0) {
                conversationsList.innerHTML = '<div style="padding: 1rem; color: #888; text-align: center; font-size: 0.875rem;">No conversations yet</div>';
                return;
            }
            conversations.forEach(conv => {
                const item = document.createElement('div');
                item.className = 'conversation-item';
                if (conv.id === currentConversationId) {
                    item.classList.add('active');
                }
                item.dataset.conversationId = conv.id;
                const previewDiv = document.createElement('div');
                previewDiv.className = 'preview';
                previewDiv.textContent = conv.preview || '(empty)';
                const metaDiv = document.createElement('div');
                metaDiv.className = 'meta';
                const date = conv.created_at ? new Date(conv.created_at).toLocaleDateString() : '';
                metaDiv.innerHTML = `${date} <span class="message-count">${conv.message_count || 0} msg${conv.message_count !== 1 ? 's' : ''}</span>`;
                item.appendChild(previewDiv);
                item.appendChild(metaDiv);
                item.addEventListener('click', () => loadConversation(conv.id));
                conversationsList.appendChild(item);
            });
        }
        // Load a specific conversation
        async function loadConversation(convId) {
            // Early-return if already showing this conversation
            if (convId === currentConversationId) {
                return;
            }
            // Clear messages
            messagesDiv.innerHTML = '';
            // Update active state in sidebar
            document.querySelectorAll('.conversation-item').forEach(item => {
                item.classList.remove('active');
            });
            document.querySelector(`[data-conversation-id="${convId}"]`)?.classList.add('active');
            currentConversationId = convId;
            try {
                const response = await fetch(`/chat/history/${convId}`);
                if (response.ok) {
                    const messages = await response.json();
                    if (messages && messages.length > 0) {
                        messages.forEach(msg => {
                            addMessage(msg.role, msg.content);
                        });
                    } else {
                        addSystemMessage('This conversation is empty');
                    }
                } else {
                    addSystemMessage('Failed to load conversation');
                }
            } catch (error) {
                console.error('Failed to load conversation:', error);
                addSystemMessage('Error loading conversation');
            }
            // Close sidebar on mobile
            if (window.innerWidth <= 768) {
                sidebar.classList.remove('open');
            }
        }
        // Create a new conversation
        async function createNewConversation() {
            try {
                const response = await fetch('/chat/new', { method: 'POST' });
                if (response.ok) {
                    const data = await response.json();
                    currentConversationId = data.conversation_id;
                    messagesDiv.innerHTML = '';
                    addSystemMessage('New conversation started');
                    await loadConversations();
                } else {
                    addSystemMessage('Failed to create new conversation');
                }
            } catch (error) {
                console.error('Failed to create new conversation:', error);
                addSystemMessage('Error creating new conversation');
            }
        }
        // Add message to display
        function addMessage(role, content, streaming = false) {
            const msgDiv = document.createElement('div');
            msgDiv.className = `message ${role}`;
@ -413,17 +155,6 @@
            return msgDiv.querySelector('.content');
        }
        function addSystemMessage(content) {
            const msgDiv = document.createElement('div');
            msgDiv.className = 'message system';
            msgDiv.innerHTML = `
                <div class="role">system</div>
                <div class="content">${escapeHtml(content)}</div>
            `;
            messagesDiv.appendChild(msgDiv);
            messagesDiv.scrollTop = messagesDiv.scrollHeight;
        }
        function escapeHtml(text) {
            const div = document.createElement('div');
            div.textContent = text;
@ -431,7 +162,7 @@
        }
        // Send message handler
-        async function sendMessage() {
+        sendBtn.addEventListener('click', async () => {
            const message = textarea.value.trim();
            if (!message) return;
@ -444,16 +175,10 @@
            addMessage('user', message);
            textarea.value = '';
            // If no conversation ID, create one
            if (!currentConversationId) {
                await createNewConversation();
            }
            try {
                // Use fetch with URLSearchParams for application/x-www-form-urlencoded
                const params = new URLSearchParams();
                params.append('message', message);
                params.append('conversation_id', currentConversationId);
                const response = await fetch('/chat', {
                    method: 'POST',
@ -467,55 +192,31 @@
                    throw new Error(`HTTP ${response.status}`);
                }
-                // Read the response as JSON (now returns JSON with response and conversation_id)
+                // Read the response as text and add assistant message
-                const data = await response.json();
+                const content = await response.text();
-                addMessage('assistant', data.response);
+                addMessage('assistant', content);
            } catch (error) {
-                addSystemMessage(`Error: ${error.message}`);
+                addMessage('system', `Error: ${error.message}`);
            } finally {
                textarea.disabled = false;
                sendBtn.disabled = false;
                sendBtn.textContent = 'Send';
                textarea.focus();
                messagesDiv.scrollTop = messagesDiv.scrollHeight;
                // Refresh conversations list
                await loadConversations();
            }
-        }
+        });
        // Event listeners
        sendBtn.addEventListener('click', sendMessage);
        newChatBtn.addEventListener('click', createNewConversation);
        // Handle Enter key in textarea
        textarea.addEventListener('keydown', (e) => {
            if (e.key === 'Enter' && !e.shiftKey) {
                e.preventDefault();
-                sendMessage();
+                sendBtn.click();
            }
        });
        // Sidebar toggle for mobile
        sidebarToggle.addEventListener('click', () => {
            sidebar.classList.toggle('open');
        });
        // Close sidebar when clicking outside on mobile
        document.addEventListener('click', (e) => {
            if (window.innerWidth <= 768) {
                if (!sidebar.contains(e.target) && !sidebarToggle.contains(e.target)) {
                    sidebar.classList.remove('open');
                }
            }
        });
        // Initial focus
        textarea.focus();
        // Load conversations on page load
        loadConversations();
    </script>
 </body>
 </html>
--- a/docker/edge/Dockerfile
+++ b/docker/edge/Dockerfile
@ -1,7 +1,4 @@
 FROM caddy:latest
 RUN apk add --no-cache bash jq curl git docker-cli python3 openssh-client autossh
 COPY entrypoint-edge.sh /usr/local/bin/entrypoint-edge.sh
 VOLUME /data
 ENTRYPOINT ["bash", "/usr/local/bin/entrypoint-edge.sh"]
--- a/docker/edge/dispatcher.sh
+++ b/docker/edge/dispatcher.sh
@ -8,8 +8,8 @@
 # 2. Scan vault/actions/ for TOML files without .result.json
 # 3. Verify TOML arrived via merged PR with admin merger (Forgejo API)
 # 4. Validate TOML using vault-env.sh validator
-# 5. Decrypt declared secrets via load_secret (lib/env.sh)
+# 5. Decrypt .env.vault.enc and extract only declared secrets
-# 6. Launch: delegate to _launch_runner_{docker,nomad} backend
+# 6. Launch: docker run --rm disinto/agents:latest <action-id>
 # 7. Write <action-id>.result.json with exit code, timestamp, logs summary
 #
 # Part of #76.
@ -19,7 +19,7 @@ set -euo pipefail
 # Resolve script root (parent of lib/)
 SCRIPT_ROOT="$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)"
-# Source shared environment (provides load_secret, log helpers, etc.)
+# Source shared environment
 source "${SCRIPT_ROOT}/../lib/env.sh"
 # Project TOML location: prefer mounted path, fall back to cloned path
@ -27,18 +27,26 @@ source "${SCRIPT_ROOT}/../lib/env.sh"
 # the shallow clone only has .toml.example files.
 PROJECTS_DIR="${PROJECTS_DIR:-${FACTORY_ROOT:-/opt/disinto}-projects}"
-# -----------------------------------------------------------------------------
+# Load vault secrets after env.sh (env.sh unsets them for agent security)
-# Backend selection: DISPATCHER_BACKEND={docker,nomad}
+# Vault secrets must be available to the dispatcher
-# Default: docker.  nomad lands as a pure addition during migration Step 5.
+if [ -f "$FACTORY_ROOT/.env.vault.enc" ] && command -v sops &>/dev/null; then
-# -----------------------------------------------------------------------------
+  set -a
-DISPATCHER_BACKEND="${DISPATCHER_BACKEND:-docker}"
+  eval "$(sops -d --output-type dotenv "$FACTORY_ROOT/.env.vault.enc" 2>/dev/null)" \
    || echo "Warning: failed to decrypt .env.vault.enc — vault secrets not loaded" >&2
  set +a
 elif [ -f "$FACTORY_ROOT/.env.vault" ]; then
  set -a
  # shellcheck source=/dev/null
  source "$FACTORY_ROOT/.env.vault"
  set +a
 fi
 # Ops repo location (vault/actions directory)
 OPS_REPO_ROOT="${OPS_REPO_ROOT:-/home/debian/disinto-ops}"
 VAULT_ACTIONS_DIR="${OPS_REPO_ROOT}/vault/actions"
 # Vault action validation
-VAULT_ENV="${SCRIPT_ROOT}/../action-vault/vault-env.sh"
+VAULT_ENV="${SCRIPT_ROOT}/../vault/vault-env.sh"
 # Admin users who can merge vault PRs (from issue #77)
 # Comma-separated list of Forgejo usernames with admin role
@ -342,113 +350,73 @@ get_dispatch_mode() {
  fi
 }
-# Commit result.json to the ops repo via git push (portable, no bind-mount).
+# Write result file for an action
 #
 # Clones the ops repo into a scratch directory, writes the result file,
 # commits as vault-bot, and pushes to the primary branch.
 # Idempotent: skips if result.json already exists upstream.
 # Retries on push conflict with rebase-and-push (handles concurrent merges).
 #
 # Usage: commit_result_via_git <action_id> <exit_code> <logs>
 commit_result_via_git() {
  local action_id="$1"
  local exit_code="$2"
  local logs="$3"
  local result_relpath="vault/actions/${action_id}.result.json"
  local ops_clone_url="${FORGE_URL}/${FORGE_OPS_REPO}.git"
  local branch="${PRIMARY_BRANCH:-main}"
  local scratch_dir
  scratch_dir=$(mktemp -d /tmp/dispatcher-result-XXXXXX)
  # shellcheck disable=SC2064
  trap "rm -rf '${scratch_dir}'" RETURN
  # Shallow clone of the ops repo — only the primary branch
  if ! git clone --depth 1 --branch "$branch" \
    "$ops_clone_url" "$scratch_dir" 2>/dev/null; then
    log "ERROR: Failed to clone ops repo for result commit (action ${action_id})"
    return 1
  fi
  # Idempotency: skip if result.json already exists upstream
  if [ -f "${scratch_dir}/${result_relpath}" ]; then
    log "Result already exists upstream for ${action_id} — skipping commit"
    return 0
  fi
  # Configure git identity as vault-bot
  git -C "$scratch_dir" config user.name "vault-bot"
  git -C "$scratch_dir" config user.email "vault-bot@disinto.local"
  # Truncate logs if too long (keep last 1000 chars)
  if [ ${#logs} -gt 1000 ]; then
    logs="${logs: -1000}"
  fi
  # Write result JSON via jq (never string-interpolate into JSON)
  mkdir -p "$(dirname "${scratch_dir}/${result_relpath}")"
  jq -n \
    --arg id "$action_id" \
    --argjson exit_code "$exit_code" \
    --arg timestamp "$(date -u '+%Y-%m-%dT%H:%M:%SZ')" \
    --arg logs "$logs" \
    '{id: $id, exit_code: $exit_code, timestamp: $timestamp, logs: $logs}' \
    > "${scratch_dir}/${result_relpath}"
  git -C "$scratch_dir" add "$result_relpath"
  git -C "$scratch_dir" commit -q -m "vault: result for ${action_id}"
  # Push with retry on conflict (rebase-and-push pattern).
  # Common case: admin merges another action PR between our clone and push.
  local attempt
  for attempt in 1 2 3; do
    if git -C "$scratch_dir" push origin "$branch" 2>/dev/null; then
      log "Result committed and pushed for ${action_id} (attempt ${attempt})"
      return 0
    fi
    log "Push conflict for ${action_id} (attempt ${attempt}/3) — rebasing"
    if ! git -C "$scratch_dir" pull --rebase origin "$branch" 2>/dev/null; then
      # Rebase conflict — check if result was pushed by another process
      git -C "$scratch_dir" rebase --abort 2>/dev/null || true
      if git -C "$scratch_dir" fetch origin "$branch" 2>/dev/null && \
         git -C "$scratch_dir" show "origin/${branch}:${result_relpath}" >/dev/null 2>&1; then
        log "Result already exists upstream for ${action_id} (pushed by another process)"
        return 0
      fi
    fi
  done
  log "ERROR: Failed to push result for ${action_id} after 3 attempts"
  return 1
 }
 # Write result file for an action via git push to the ops repo.
 # Usage: write_result <action_id> <exit_code> <logs>
 write_result() {
  local action_id="$1"
  local exit_code="$2"
  local logs="$3"
-  commit_result_via_git "$action_id" "$exit_code" "$logs"
+  local result_file="${VAULT_ACTIONS_DIR}/${action_id}.result.json"
  # Truncate logs if too long (keep last 1000 chars)
  if [ ${#logs} -gt 1000 ]; then
    logs="${logs: -1000}"
  fi
  # Write result JSON
  jq -n \
    --arg id "$action_id" \
    --argjson exit_code "$exit_code" \
    --arg timestamp "$(date -u '+%Y-%m-%dT%H:%M:%SZ')" \
    --arg logs "$logs" \
    '{id: $id, exit_code: $exit_code, timestamp: $timestamp, logs: $logs}' \
    > "$result_file"
  log "Result written: ${result_file}"
 }
-# -----------------------------------------------------------------------------
+# Launch runner for the given action
-# Pluggable launcher backends
+# Usage: launch_runner <toml_file>
-# -----------------------------------------------------------------------------
+launch_runner() {
  local toml_file="$1"
  local action_id
  action_id=$(basename "$toml_file" .toml)
-# _launch_runner_docker ACTION_ID SECRETS_CSV MOUNTS_CSV
+  log "Launching runner for action: ${action_id}"
 #
 # Builds and executes a `docker run` command for the vault runner.
 # Secrets are resolved via load_secret (lib/env.sh).
 # Returns: exit code of the docker run.  Stdout/stderr are captured to a temp
 #          log file whose path is printed to stdout (caller reads it).
 _launch_runner_docker() {
  local action_id="$1"
  local secrets_csv="$2"
  local mounts_csv="$3"
  # Validate TOML
  if ! validate_action "$toml_file"; then
    log "ERROR: Action validation failed for ${action_id}"
    write_result "$action_id" 1 "Validation failed: see logs above"
    return 1
  fi
  # Check dispatch mode to determine if admin verification is needed
  local dispatch_mode
  dispatch_mode=$(get_dispatch_mode "$toml_file")
  if [ "$dispatch_mode" = "direct" ]; then
    log "Action ${action_id}: tier=${VAULT_TIER:-unknown}, dispatch_mode=${dispatch_mode} — skipping admin merge verification (direct commit)"
  else
    # Verify admin merge for PR-based actions
    log "Action ${action_id}: tier=${VAULT_TIER:-unknown}, dispatch_mode=${dispatch_mode} — verifying admin merge"
    if ! verify_admin_merged "$toml_file"; then
      log "ERROR: Admin merge verification failed for ${action_id}"
      write_result "$action_id" 1 "Admin merge verification failed: see logs above"
      return 1
    fi
    log "Action ${action_id}: admin merge verified"
  fi
  # Extract secrets from validated action
  local secrets_array
  secrets_array="${VAULT_ACTION_SECRETS:-}"
  # Build docker run command (self-contained, no compose context needed).
  # The edge container has the Docker socket but not the host's compose project,
  # so docker compose run would fail with exit 125. docker run is self-contained:
  # the dispatcher knows the image, network, env vars, and entrypoint.
  local -a cmd=(docker run --rm
    --name "vault-runner-${action_id}"
    --network host
@ -483,27 +451,29 @@ _launch_runner_docker() {
    cmd+=(-v "${runtime_home}/.claude.json:/home/agent/.claude.json:ro")
  fi
-  # Add environment variables for secrets (resolved via load_secret)
+  # Add environment variables for secrets (if any declared)
-  if [ -n "$secrets_csv" ]; then
+  if [ -n "$secrets_array" ]; then
-    local secret
+    for secret in $secrets_array; do
    for secret in $(echo "$secrets_csv" | tr ',' ' '); do
      secret=$(echo "$secret" | xargs)
-      [ -n "$secret" ] || continue
+      if [ -n "$secret" ]; then
-      local secret_val
+        # Verify secret exists in vault
-      secret_val=$(load_secret "$secret") || true
+        if [ -z "${!secret:-}" ]; then
-      if [ -z "$secret_val" ]; then
+          log "ERROR: Secret '${secret}' not found in vault for action ${action_id}"
-        log "ERROR: Secret '${secret}' could not be resolved for action ${action_id}"
+          write_result "$action_id" 1 "Secret not found in vault: ${secret}"
-        write_result "$action_id" 1 "Secret not found: ${secret}"
+          return 1
-        return 1
+        fi
        cmd+=(-e "${secret}=${!secret}")
      fi
      cmd+=(-e "${secret}=${secret_val}")
    done
  else
    log "Action ${action_id} has no secrets declared — runner will execute without extra env vars"
  fi
-  # Add volume mounts for file-based credentials
+  # Add volume mounts for file-based credentials (if any declared)
-  if [ -n "$mounts_csv" ]; then
+  local mounts_array
-    local mount_alias
+  mounts_array="${VAULT_ACTION_MOUNTS:-}"
-    for mount_alias in $(echo "$mounts_csv" | tr ',' ' '); do
+  if [ -n "$mounts_array" ]; then
    for mount_alias in $mounts_array; do
      mount_alias=$(echo "$mount_alias" | xargs)
      [ -n "$mount_alias" ] || continue
      case "$mount_alias" in
@ -531,7 +501,7 @@ _launch_runner_docker() {
  # Image and entrypoint arguments: runner entrypoint + action-id
  cmd+=(disinto/agents:latest /home/agent/disinto/docker/runner/entrypoint-runner.sh "$action_id")
-  log "Running: docker run --rm vault-runner-${action_id} (secrets: ${secrets_csv:-none}, mounts: ${mounts_csv:-none})"
+  log "Running: docker run --rm vault-runner-${action_id} (secrets: ${secrets_array:-none}, mounts: ${mounts_array:-none})"
  # Create temp file for logs
  local log_file
@ -539,6 +509,7 @@ _launch_runner_docker() {
  trap 'rm -f "$log_file"' RETURN
  # Execute with array expansion (safe from shell injection)
  # Capture stdout and stderr to log file
  "${cmd[@]}" > "$log_file" 2>&1
  local exit_code=$?
@ -558,137 +529,6 @@ _launch_runner_docker() {
  return $exit_code
 }
 # _launch_runner_nomad ACTION_ID SECRETS_CSV MOUNTS_CSV
 #
 # Nomad backend stub — will be implemented in migration Step 5.
 _launch_runner_nomad() {
  echo "nomad backend not yet implemented" >&2
  return 1
 }
 # Launch runner for the given action (backend-agnostic orchestrator)
 # Usage: launch_runner <toml_file>
 launch_runner() {
  local toml_file="$1"
  local action_id
  action_id=$(basename "$toml_file" .toml)
  log "Launching runner for action: ${action_id}"
  # Validate TOML
  if ! validate_action "$toml_file"; then
    log "ERROR: Action validation failed for ${action_id}"
    write_result "$action_id" 1 "Validation failed: see logs above"
    return 1
  fi
  # Check dispatch mode to determine if admin verification is needed
  local dispatch_mode
  dispatch_mode=$(get_dispatch_mode "$toml_file")
  if [ "$dispatch_mode" = "direct" ]; then
    log "Action ${action_id}: tier=${VAULT_TIER:-unknown}, dispatch_mode=${dispatch_mode} — skipping admin merge verification (direct commit)"
  else
    # Verify admin merge for PR-based actions
    log "Action ${action_id}: tier=${VAULT_TIER:-unknown}, dispatch_mode=${dispatch_mode} — verifying admin merge"
    if ! verify_admin_merged "$toml_file"; then
      log "ERROR: Admin merge verification failed for ${action_id}"
      write_result "$action_id" 1 "Admin merge verification failed: see logs above"
      return 1
    fi
    log "Action ${action_id}: admin merge verified"
  fi
  # Build CSV lists from validated action metadata
  local secrets_csv=""
  if [ -n "${VAULT_ACTION_SECRETS:-}" ]; then
    # Convert space-separated to comma-separated
    secrets_csv=$(echo "${VAULT_ACTION_SECRETS}" | xargs | tr ' ' ',')
  fi
  local mounts_csv=""
  if [ -n "${VAULT_ACTION_MOUNTS:-}" ]; then
    mounts_csv=$(echo "${VAULT_ACTION_MOUNTS}" | xargs | tr ' ' ',')
  fi
  # Delegate to the selected backend
  "_launch_runner_${DISPATCHER_BACKEND}" "$action_id" "$secrets_csv" "$mounts_csv"
 }
 # -----------------------------------------------------------------------------
 # Pluggable sidecar launcher (reproduce / triage / verify)
 # -----------------------------------------------------------------------------
 # _dispatch_sidecar_docker CONTAINER_NAME ISSUE_NUM PROJECT_TOML IMAGE [FORMULA]
 #
 # Launches a sidecar container via docker run (background, pid-tracked).
 # Prints the background PID to stdout.
 _dispatch_sidecar_docker() {
  local container_name="$1"
  local issue_number="$2"
  local project_toml="$3"
  local image="$4"
  local formula="${5:-}"
  local -a cmd=(docker run --rm
    --name "${container_name}"
    --network host
    --security-opt apparmor=unconfined
    -v /var/run/docker.sock:/var/run/docker.sock
    -v agent-data:/home/agent/data
    -v project-repos:/home/agent/repos
    -e "FORGE_URL=${FORGE_URL}"
    -e "FORGE_TOKEN=${FORGE_TOKEN}"
    -e "FORGE_REPO=${FORGE_REPO}"
    -e "PRIMARY_BRANCH=${PRIMARY_BRANCH:-main}"
    -e DISINTO_CONTAINER=1
  )
  # Set formula if provided
  if [ -n "$formula" ]; then
    cmd+=(-e "DISINTO_FORMULA=${formula}")
  fi
  # Pass through ANTHROPIC_API_KEY if set
  if [ -n "${ANTHROPIC_API_KEY:-}" ]; then
    cmd+=(-e "ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}")
  fi
  # Mount shared Claude config dir and ~/.ssh from the runtime user's home
  local runtime_home="${HOME:-/home/debian}"
  if [ -d "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}" ]; then
    cmd+=(-v "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}")
    cmd+=(-e "CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}")
  fi
  if [ -f "${runtime_home}/.claude.json" ]; then
    cmd+=(-v "${runtime_home}/.claude.json:/home/agent/.claude.json:ro")
  fi
  if [ -d "${runtime_home}/.ssh" ]; then
    cmd+=(-v "${runtime_home}/.ssh:/home/agent/.ssh:ro")
  fi
  if [ -f /usr/local/bin/claude ]; then
    cmd+=(-v /usr/local/bin/claude:/usr/local/bin/claude:ro)
  fi
  # Mount the project TOML into the container at a stable path
  local container_toml="/home/agent/project.toml"
  cmd+=(-v "${project_toml}:${container_toml}:ro")
  cmd+=("${image}" "$container_toml" "$issue_number")
  # Launch in background
  "${cmd[@]}" &
  echo $!
 }
 # _dispatch_sidecar_nomad CONTAINER_NAME ISSUE_NUM PROJECT_TOML IMAGE [FORMULA]
 #
 # Nomad sidecar backend stub — will be implemented in migration Step 5.
 _dispatch_sidecar_nomad() {
  echo "nomad backend not yet implemented" >&2
  return 1
 }
 # -----------------------------------------------------------------------------
 # Reproduce dispatch — launch sidecar for bug-report issues
 # -----------------------------------------------------------------------------
@ -767,13 +607,52 @@ dispatch_reproduce() {
  log "Dispatching reproduce-agent for issue #${issue_number} (project: ${project_toml})"
-  local bg_pid
+  # Build docker run command using array (safe from injection)
-  bg_pid=$("_dispatch_sidecar_${DISPATCHER_BACKEND}" \
+  local -a cmd=(docker run --rm
-    "disinto-reproduce-${issue_number}" \
+    --name "disinto-reproduce-${issue_number}"
-    "$issue_number" \
+    --network host
-    "$project_toml" \
+    --security-opt apparmor=unconfined
-    "disinto-reproduce:latest")
+    -v /var/run/docker.sock:/var/run/docker.sock
    -v agent-data:/home/agent/data
    -v project-repos:/home/agent/repos
    -e "FORGE_URL=${FORGE_URL}"
    -e "FORGE_TOKEN=${FORGE_TOKEN}"
    -e "FORGE_REPO=${FORGE_REPO}"
    -e "PRIMARY_BRANCH=${PRIMARY_BRANCH:-main}"
    -e DISINTO_CONTAINER=1
  )
  # Pass through ANTHROPIC_API_KEY if set
  if [ -n "${ANTHROPIC_API_KEY:-}" ]; then
    cmd+=(-e "ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}")
  fi
  # Mount shared Claude config dir and ~/.ssh from the runtime user's home if available
  local runtime_home="${HOME:-/home/debian}"
  if [ -d "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}" ]; then
    cmd+=(-v "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}")
    cmd+=(-e "CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}")
  fi
  if [ -f "${runtime_home}/.claude.json" ]; then
    cmd+=(-v "${runtime_home}/.claude.json:/home/agent/.claude.json:ro")
  fi
  if [ -d "${runtime_home}/.ssh" ]; then
    cmd+=(-v "${runtime_home}/.ssh:/home/agent/.ssh:ro")
  fi
  # Mount claude CLI binary if present on host
  if [ -f /usr/local/bin/claude ]; then
    cmd+=(-v /usr/local/bin/claude:/usr/local/bin/claude:ro)
  fi
  # Mount the project TOML into the container at a stable path
  local container_toml="/home/agent/project.toml"
  cmd+=(-v "${project_toml}:${container_toml}:ro")
  cmd+=(disinto-reproduce:latest "$container_toml" "$issue_number")
  # Launch in background; write pid-file so we don't double-launch
  "${cmd[@]}" &
  local bg_pid=$!
  echo "$bg_pid" > "$(_reproduce_lockfile "$issue_number")"
  log "Reproduce container launched (pid ${bg_pid}) for issue #${issue_number}"
 }
@ -853,14 +732,53 @@ dispatch_triage() {
  log "Dispatching triage-agent for issue #${issue_number} (project: ${project_toml})"
-  local bg_pid
+  # Build docker run command using array (safe from injection)
-  bg_pid=$("_dispatch_sidecar_${DISPATCHER_BACKEND}" \
+  local -a cmd=(docker run --rm
-    "disinto-triage-${issue_number}" \
+    --name "disinto-triage-${issue_number}"
-    "$issue_number" \
+    --network host
-    "$project_toml" \
+    --security-opt apparmor=unconfined
-    "disinto-reproduce:latest" \
+    -v /var/run/docker.sock:/var/run/docker.sock
-    "triage")
+    -v agent-data:/home/agent/data
    -v project-repos:/home/agent/repos
    -e "FORGE_URL=${FORGE_URL}"
    -e "FORGE_TOKEN=${FORGE_TOKEN}"
    -e "FORGE_REPO=${FORGE_REPO}"
    -e "PRIMARY_BRANCH=${PRIMARY_BRANCH:-main}"
    -e DISINTO_CONTAINER=1
    -e DISINTO_FORMULA=triage
  )
  # Pass through ANTHROPIC_API_KEY if set
  if [ -n "${ANTHROPIC_API_KEY:-}" ]; then
    cmd+=(-e "ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}")
  fi
  # Mount shared Claude config dir and ~/.ssh from the runtime user's home if available
  local runtime_home="${HOME:-/home/debian}"
  if [ -d "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}" ]; then
    cmd+=(-v "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}")
    cmd+=(-e "CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}")
  fi
  if [ -f "${runtime_home}/.claude.json" ]; then
    cmd+=(-v "${runtime_home}/.claude.json:/home/agent/.claude.json:ro")
  fi
  if [ -d "${runtime_home}/.ssh" ]; then
    cmd+=(-v "${runtime_home}/.ssh:/home/agent/.ssh:ro")
  fi
  # Mount claude CLI binary if present on host
  if [ -f /usr/local/bin/claude ]; then
    cmd+=(-v /usr/local/bin/claude:/usr/local/bin/claude:ro)
  fi
  # Mount the project TOML into the container at a stable path
  local container_toml="/home/agent/project.toml"
  cmd+=(-v "${project_toml}:${container_toml}:ro")
  cmd+=(disinto-reproduce:latest "$container_toml" "$issue_number")
  # Launch in background; write pid-file so we don't double-launch
  "${cmd[@]}" &
  local bg_pid=$!
  echo "$bg_pid" > "$(_triage_lockfile "$issue_number")"
  log "Triage container launched (pid ${bg_pid}) for issue #${issue_number}"
 }
@ -1016,14 +934,53 @@ dispatch_verify() {
  log "Dispatching verification-agent for issue #${issue_number} (project: ${project_toml})"
-  local bg_pid
+  # Build docker run command using array (safe from injection)
-  bg_pid=$("_dispatch_sidecar_${DISPATCHER_BACKEND}" \
+  local -a cmd=(docker run --rm
-    "disinto-verify-${issue_number}" \
+    --name "disinto-verify-${issue_number}"
-    "$issue_number" \
+    --network host
-    "$project_toml" \
+    --security-opt apparmor=unconfined
-    "disinto-reproduce:latest" \
+    -v /var/run/docker.sock:/var/run/docker.sock
-    "verify")
+    -v agent-data:/home/agent/data
    -v project-repos:/home/agent/repos
    -e "FORGE_URL=${FORGE_URL}"
    -e "FORGE_TOKEN=${FORGE_TOKEN}"
    -e "FORGE_REPO=${FORGE_REPO}"
    -e "PRIMARY_BRANCH=${PRIMARY_BRANCH:-main}"
    -e DISINTO_CONTAINER=1
    -e DISINTO_FORMULA=verify
  )
  # Pass through ANTHROPIC_API_KEY if set
  if [ -n "${ANTHROPIC_API_KEY:-}" ]; then
    cmd+=(-e "ANTHROPIC_API_KEY=${ANTHROPIC_API_KEY}")
  fi
  # Mount shared Claude config dir and ~/.ssh from the runtime user's home if available
  local runtime_home="${HOME:-/home/debian}"
  if [ -d "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}" ]; then
    cmd+=(-v "${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}")
    cmd+=(-e "CLAUDE_CONFIG_DIR=${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}")
  fi
  if [ -f "${runtime_home}/.claude.json" ]; then
    cmd+=(-v "${runtime_home}/.claude.json:/home/agent/.claude.json:ro")
  fi
  if [ -d "${runtime_home}/.ssh" ]; then
    cmd+=(-v "${runtime_home}/.ssh:/home/agent/.ssh:ro")
  fi
  # Mount claude CLI binary if present on host
  if [ -f /usr/local/bin/claude ]; then
    cmd+=(-v /usr/local/bin/claude:/usr/local/bin/claude:ro)
  fi
  # Mount the project TOML into the container at a stable path
  local container_toml="/home/agent/project.toml"
  cmd+=(-v "${project_toml}:${container_toml}:ro")
  cmd+=(disinto-reproduce:latest "$container_toml" "$issue_number")
  # Launch in background; write pid-file so we don't double-launch
  "${cmd[@]}" &
  local bg_pid=$!
  echo "$bg_pid" > "$(_verify_lockfile "$issue_number")"
  log "Verification container launched (pid ${bg_pid}) for issue #${issue_number}"
 }
@ -1045,25 +1002,10 @@ ensure_ops_repo() {
 # Main dispatcher loop
 main() {
-  log "Starting dispatcher (backend=${DISPATCHER_BACKEND})..."
+  log "Starting dispatcher..."
  log "Polling ops repo: ${VAULT_ACTIONS_DIR}"
  log "Admin users: ${ADMIN_USERS}"
  # Validate backend selection at startup
  case "$DISPATCHER_BACKEND" in
    docker) ;;
    nomad)
      log "ERROR: nomad backend not yet implemented"
      echo "nomad backend not yet implemented" >&2
      exit 1
      ;;
    *)
      log "ERROR: unknown DISPATCHER_BACKEND=${DISPATCHER_BACKEND}"
      echo "unknown DISPATCHER_BACKEND=${DISPATCHER_BACKEND} (expected: docker, nomad)" >&2
      exit 1
      ;;
  esac
  while true; do
    # Refresh ops repo at the start of each poll cycle
    ensure_ops_repo
--- a/docker/edge/entrypoint-edge.sh
+++ b/docker/edge/entrypoint-edge.sh
@ -173,67 +173,6 @@ PROJECT_TOML="${PROJECT_TOML:-projects/disinto.toml}"
  sleep 1200  # 20 minutes
 done) &
 # ── Load required secrets from secrets/*.enc (#777) ────────────────────
 # Edge container declares its required secrets; missing ones cause a hard fail.
 _AGE_KEY_FILE="${HOME}/.config/sops/age/keys.txt"
 _SECRETS_DIR="/opt/disinto/secrets"
 EDGE_REQUIRED_SECRETS="CADDY_SSH_KEY CADDY_SSH_HOST CADDY_SSH_USER CADDY_ACCESS_LOG"
 _edge_decrypt_secret() {
  local enc_path="${_SECRETS_DIR}/${1}.enc"
  [ -f "$enc_path" ] || return 1
  age -d -i "$_AGE_KEY_FILE" "$enc_path" 2>/dev/null
 }
 if [ -f "$_AGE_KEY_FILE" ] && [ -d "$_SECRETS_DIR" ]; then
  _missing=""
  for _secret_name in $EDGE_REQUIRED_SECRETS; do
    _val=$(_edge_decrypt_secret "$_secret_name") || { _missing="${_missing} ${_secret_name}"; continue; }
    export "$_secret_name=$_val"
  done
  if [ -n "$_missing" ]; then
    echo "FATAL: required secrets missing from secrets/*.enc:${_missing}" >&2
    echo "  Run 'disinto secrets add <NAME>' for each missing secret." >&2
    echo "  If migrating from .env.vault.enc, run 'disinto secrets migrate-from-vault' first." >&2
    exit 1
  fi
  echo "edge: loaded required secrets: ${EDGE_REQUIRED_SECRETS}" >&2
 else
  echo "FATAL: age key (${_AGE_KEY_FILE}) or secrets dir (${_SECRETS_DIR}) not found — cannot load required secrets" >&2
  echo "  Ensure age is installed and secrets/*.enc files are present." >&2
  exit 1
 fi
 # Start daily engagement collection cron loop in background (#745)
 # Runs collect-engagement.sh daily at ~23:50 UTC via a sleep loop that
 # calculates seconds until the next 23:50 window. SSH key from secrets/*.enc (#777).
 (while true; do
  # Calculate seconds until next 23:50 UTC
  _now=$(date -u +%s)
  _target=$(date -u -d "today 23:50" +%s 2>/dev/null || date -u -d "23:50" +%s 2>/dev/null || echo 0)
  if [ "$_target" -le "$_now" ]; then
    _target=$(( _target + 86400 ))
  fi
  _sleep_secs=$(( _target - _now ))
  echo "edge: collect-engagement scheduled in ${_sleep_secs}s (next 23:50 UTC)" >&2
  sleep "$_sleep_secs"
  _fetch_log="/tmp/caddy-access-log-fetch.log"
  _ssh_key_file=$(mktemp)
  printf '%s\n' "$CADDY_SSH_KEY" > "$_ssh_key_file"
  chmod 0600 "$_ssh_key_file"
  scp -i "$_ssh_key_file" -o StrictHostKeyChecking=accept-new -o ConnectTimeout=10 -o BatchMode=yes \
    "${CADDY_SSH_USER}@${CADDY_SSH_HOST}:${CADDY_ACCESS_LOG}" \
    "$_fetch_log" 2>&1 | tee -a /opt/disinto-logs/collect-engagement.log || true
  rm -f "$_ssh_key_file"
  if [ -s "$_fetch_log" ]; then
    CADDY_ACCESS_LOG="$_fetch_log" bash /opt/disinto/site/collect-engagement.sh 2>&1 \
      | tee -a /opt/disinto-logs/collect-engagement.log || true
  else
    echo "edge: collect-engagement: fetched log is empty, skipping parse" >&2
  fi
  rm -f "$_fetch_log"
 done) &
 # Caddy as main process — run in foreground via wait so background jobs survive
 # (exec replaces the shell, which can orphan backgrounded subshells)
 caddy run --config /etc/caddy/Caddyfile --adapter caddyfile &
--- a/docker/reproduce/Dockerfile
+++ b/docker/reproduce/Dockerfile
@ -7,8 +7,5 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
 RUN useradd -m -u 1000 -s /bin/bash agent
 COPY docker/reproduce/entrypoint-reproduce.sh /entrypoint-reproduce.sh
 RUN chmod +x /entrypoint-reproduce.sh
 VOLUME /home/agent/data
 VOLUME /home/agent/repos
 WORKDIR /home/agent
 ENTRYPOINT ["/entrypoint-reproduce.sh"]
--- a/docs/VAULT.md
+++ b/docs/VAULT.md
@ -26,8 +26,8 @@ The `main` branch on the ops repo (`johba/disinto-ops`) is protected via Forgejo
 ## Vault PR Lifecycle
-1. **Request** — Agent calls `lib/action-vault.sh:vault_request()` with action TOML content
+1. **Request** — Agent calls `lib/vault.sh:vault_request()` with action TOML content
-2. **Validation** — TOML is validated against the schema in `action-vault/vault-env.sh`
+2. **Validation** — TOML is validated against the schema in `vault/vault-env.sh`
 3. **PR Creation** — A PR is created on `disinto-ops` with:
   - Branch: `vault/<action-id>`
   - Title: `vault: <action-id>`
@ -90,12 +90,12 @@ To verify the protection is working:
 - #73 — Vault redesign proposal
 - #74 — Vault action TOML schema
- #75 — Vault PR creation helper (`lib/action-vault.sh`)
+- #75 — Vault PR creation helper (`lib/vault.sh`)
 - #76 — Dispatcher rewrite (poll for merged vault PRs)
 - #77 — Branch protection on ops repo (this issue)
 ## See Also
- [`lib/action-vault.sh`](../lib/action-vault.sh) — Vault PR creation helper
+- [`lib/vault.sh`](../lib/vault.sh) — Vault PR creation helper
- [`action-vault/vault-env.sh`](../action-vault/vault-env.sh) — TOML validation
+- [`vault/vault-env.sh`](../vault/vault-env.sh) — TOML validation
 - [`lib/branch-protection.sh`](../lib/branch-protection.sh) — Branch protection helper
--- a/docs/agents-llama.md
+++ b/docs/agents-llama.md
@ -1,59 +0,0 @@
 # agents-llama — Local-Qwen Agents
 The `agents-llama` service is an optional compose service that runs agents
 backed by a local llama-server instance (e.g. Qwen) instead of the Anthropic
 API. It uses the same Docker image as the main `agents` service but connects to
 a local inference endpoint via `ANTHROPIC_BASE_URL`.
 Two profiles are available:
 | Profile | Service | Roles | Use case |
 |---------|---------|-------|----------|
 | _(default)_ | `agents-llama` | `dev` only | Conservative: single-role soak test |
 | `agents-llama-all` | `agents-llama-all` | all 7 (review, dev, gardener, architect, planner, predictor, supervisor) | Pre-migration: validate every role on llama before Nomad cutover |
 ## Enabling
 Set `ENABLE_LLAMA_AGENT=1` in `.env` (or `.env.enc`) and provide the required
 credentials:
 ```env
 ENABLE_LLAMA_AGENT=1
 FORGE_TOKEN_LLAMA=<dev-qwen API token>
 FORGE_PASS_LLAMA=<dev-qwen password>
 ANTHROPIC_BASE_URL=http://host.docker.internal:8081   # llama-server endpoint
 ```
 Then regenerate the compose file (`disinto init ...`) and bring the stack up.
 ### Running all 7 roles (agents-llama-all)
 ```bash
 docker compose --profile agents-llama-all up -d
 ```
 This starts the `agents-llama-all` container with all 7 bot roles against the
 local llama endpoint. The per-role forge tokens (`FORGE_REVIEW_TOKEN`,
 `FORGE_GARDENER_TOKEN`, etc.) must be set in `.env` — they are the same tokens
 used by the Claude-backed `agents` container.
 ## Prerequisites
 - **llama-server** (or compatible OpenAI-API endpoint) running on the host,
  reachable from inside Docker at the URL set in `ANTHROPIC_BASE_URL`.
 - A Forgejo bot user (e.g. `dev-qwen`) with its own API token and password,
  stored as `FORGE_TOKEN_LLAMA` / `FORGE_PASS_LLAMA`.
 ## Behaviour
 - `agents-llama`: `AGENT_ROLES=dev` — only picks up dev work.
 - `agents-llama-all`: `AGENT_ROLES=review,dev,gardener,architect,planner,predictor,supervisor` — runs all 7 roles.
 - `CLAUDE_AUTOCOMPACT_PCT_OVERRIDE=60` — more aggressive compaction for smaller
  context windows.
 - Serialises on the llama-server's single KV cache (AD-002).
 ## Disabling
 Set `ENABLE_LLAMA_AGENT=0` (or leave it unset) and regenerate. The service
 block is omitted entirely from `docker-compose.yml`; the stack starts cleanly
 without it.
--- a/docs/edge-routing-fallback.md
+++ b/docs/edge-routing-fallback.md
@ -1,149 +0,0 @@
 # Edge Routing Fallback: Per-Project Subdomains
 > **Status:** Contingency plan. Only implement if subpath routing (#704 / #708)
 > proves unworkable.
 ## Context
 The primary approach routes services under subpaths of `<project>.disinto.ai`:
 | Service    | Primary (subpath)                          |
 |------------|--------------------------------------------|
 | Forgejo    | `<project>.disinto.ai/forge/`              |
 | Woodpecker | `<project>.disinto.ai/ci/`                 |
 | Chat       | `<project>.disinto.ai/chat/`               |
 | Staging    | `<project>.disinto.ai/staging/`            |
 The fallback uses per-service subdomains instead:
 | Service    | Fallback (subdomain)                       |
 |------------|--------------------------------------------|
 | Forgejo    | `forge.<project>.disinto.ai/`              |
 | Woodpecker | `ci.<project>.disinto.ai/`                 |
 | Chat       | `chat.<project>.disinto.ai/`               |
 | Staging    | `<project>.disinto.ai/`  (root)            |
 The wildcard cert from #621 already covers `*.<project>.disinto.ai` — no new
 DNS records or certs are needed for sub-subdomains because `*.disinto.ai`
 matches one level deep. For sub-subdomains like `forge.<project>.disinto.ai`
 we would need to add a second wildcard (`*.*.disinto.ai`) or explicit DNS
 records per project. Both are straightforward with the existing Gandi DNS-01
 setup.
 ## Pivot Decision Criteria
 **Pivot if:**
 - Forgejo `ROOT_URL` under a subpath (`/forge/`) causes redirect loops that
  cannot be fixed with `X-Forwarded-Prefix` or Caddy `uri strip_prefix`.
 - Woodpecker's `WOODPECKER_HOST` does not honour subpath prefixes, causing
  OAuth callback mismatches that persist after adjusting redirect URIs.
 - Forward-auth on `/chat/*` conflicts with Forgejo's own OAuth flow when both
  share the same origin (cookie collision, CSRF token mismatch).
 **Do NOT pivot if:**
 - Forgejo login redirects to `/` instead of `/forge/` — fixable with Caddy
  `handle_path` + `uri prefix` rewrite.
 - Woodpecker UI assets 404 under `/ci/` — fixable with asset prefix config
  (`WOODPECKER_ROOT_PATH`).
 - A single OAuth app needs a second redirect URI — Forgejo supports multiple
  `redirect_uris` in the same app.
 ## Fallback Topology
 ### Caddyfile
 Replace the single `:80` block with four host blocks:
 ```caddy
 # Main project domain — staging / landing
 <project>.disinto.ai {
    reverse_proxy staging:80
 }
 # Forgejo — root path, no subpath rewrite needed
 forge.<project>.disinto.ai {
    reverse_proxy forgejo:3000
 }
 # Woodpecker CI — root path
 ci.<project>.disinto.ai {
    reverse_proxy woodpecker:8000
 }
 # Chat — with forward_auth (same as #709, but on its own host)
 chat.<project>.disinto.ai {
    handle /login {
        reverse_proxy chat:8080
    }
    handle /oauth/callback {
        reverse_proxy chat:8080
    }
    handle /* {
        forward_auth chat:8080 {
            uri /auth/verify
            copy_headers X-Forwarded-User
            header_up X-Forward-Auth-Secret {$FORWARD_AUTH_SECRET}
        }
        reverse_proxy chat:8080
    }
 }
 ```
 **Current file:** `docker/Caddyfile` (generated by `lib/generators.sh:_generate_caddyfile_impl`, line ~596).
 ### Service Configuration Changes
 | Variable / Setting         | Current (subpath)                              | Fallback (subdomain)                            | File                        |
 |----------------------------|------------------------------------------------|-------------------------------------------------|-----------------------------|
 | Forgejo `ROOT_URL`         | `https://<project>.disinto.ai/forge/`          | `https://forge.<project>.disinto.ai/`           | forgejo `app.ini`           |
 | `WOODPECKER_HOST`          | `http://localhost:8000` (subpath via proxy)     | `https://ci.<project>.disinto.ai`               | `lib/ci-setup.sh` line ~164 |
 | Woodpecker OAuth redirect  | `https://<project>.disinto.ai/ci/authorize`    | `https://ci.<project>.disinto.ai/authorize`     | `lib/ci-setup.sh` line ~153 |
 | Chat OAuth redirect        | `https://<project>.disinto.ai/chat/oauth/callback` | `https://chat.<project>.disinto.ai/oauth/callback` | `lib/ci-setup.sh` line ~188 |
 | `EDGE_TUNNEL_FQDN`         | `<project>.disinto.ai`                         | unchanged (main domain)                         | `lib/generators.sh` line ~432 |
 ### New Environment Variables (pivot only)
 These would be added to `lib/generators.sh` `_generate_compose_impl()` in the
 edge service environment block (currently line ~415):
 | Variable                     | Value                                  |
 |------------------------------|----------------------------------------|
 | `EDGE_TUNNEL_FQDN_FORGE`    | `forge.<project>.disinto.ai`           |
 | `EDGE_TUNNEL_FQDN_CI`       | `ci.<project>.disinto.ai`              |
 | `EDGE_TUNNEL_FQDN_CHAT`     | `chat.<project>.disinto.ai`            |
 ### DNS
 No new records needed if the registrar supports `*.*.disinto.ai` wildcards.
 Otherwise, add explicit A/CNAME records per project:
 ```
 forge.<project>.disinto.ai  → edge server IP
 ci.<project>.disinto.ai     → edge server IP
 chat.<project>.disinto.ai   → edge server IP
 ```
 The edge server already handles TLS via Caddy's automatic HTTPS with the
 existing ACME / DNS-01 challenge.
 ### Edge Control (`tools/edge-control/register.sh`)
 Currently `do_register()` creates a single route for `<project>.disinto.ai`.
 The fallback would need to register four routes (or accept a `--subdomain`
 parameter). See the TODO in `register.sh`.
 ## Files to Change on Pivot
 | File                              | What changes                                                    |
 |-----------------------------------|-----------------------------------------------------------------|
 | `docker/Caddyfile`               | Replace single host block → four host blocks (see above)        |
 | `lib/generators.sh`              | Add `EDGE_TUNNEL_FQDN_{FORGE,CI,CHAT}` env vars to compose     |
 | `lib/ci-setup.sh` ~line 153      | Woodpecker OAuth redirect URI → `ci.<project>` subdomain        |
 | `lib/ci-setup.sh` ~line 188      | Chat OAuth redirect URI → `chat.<project>` subdomain            |
 | `tools/edge-control/register.sh` | Register four routes per project instead of one                 |
 | `tools/edge-control/lib/caddy.sh`| `add_route()` gains subdomain support                           |
 | forgejo `app.ini`                 | `ROOT_URL` → `https://forge.<project>.disinto.ai/`             |
 Estimated effort for a full pivot: **under one day** given this plan.
--- a/docs/mirror-bootstrap.md
+++ b/docs/mirror-bootstrap.md
@ -1,59 +0,0 @@
 # Mirror Bootstrap — Pull-Mirror Cutover Path
 How to populate an empty Forgejo repo from an external source using
 `lib/mirrors.sh`'s `mirror_pull_register()`.
 ## Prerequisites
 | Variable | Example | Purpose |
 |---|---|---|
 | `FORGE_URL` | `http://forgejo:3000` | Forgejo instance base URL |
 | `FORGE_API_BASE` | `${FORGE_URL}/api/v1` | Global API base (set by `lib/env.sh`) |
 | `FORGE_TOKEN` | (admin or org-owner token) | Must have `repo:create` scope |
 The target org/user must already exist on the Forgejo instance.
 ## Command
 ```bash
 source lib/env.sh
 source lib/mirrors.sh
 # Register a pull mirror — creates the repo and starts the first sync.
 mirror_pull_register \
  "https://codeberg.org/johba/disinto.git" \   # source URL
  "disinto-admin" \                             # target owner
  "disinto" \                                   # target repo name
  "8h0m0s"                                      # sync interval (optional, default 8h)
 ```
 The function calls `POST /api/v1/repos/migrate` with `mirror: true`.
 Forgejo creates the repo and immediately queues the first sync.
 ## Verifying the sync
 ```bash
 # Check mirror status via API
 forge_api GET "/repos/disinto-admin/disinto" | jq '.mirror, .mirror_interval'
 # Confirm content arrived — should list branches
 forge_api GET "/repos/disinto-admin/disinto/branches" | jq '.[].name'
 ```
 The first sync typically completes within a few seconds for small-to-medium
 repos.  For large repos, poll the branches endpoint until content appears.
 ## Cutover scenario (Nomad migration)
 At cutover to the Nomad box:
 1. Stand up fresh Forgejo on the Nomad cluster (empty instance).
 2. Create the `disinto-admin` org via `disinto init` or API.
 3. Run `mirror_pull_register` pointing at the Codeberg source.
 4. Wait for sync to complete (check branches endpoint).
 5. Once content is confirmed, proceed with `disinto init` against the
   now-populated repo — all subsequent `mirror_push` calls will push
   to any additional mirrors configured in `projects/*.toml`.
 No manual `git clone` + `git push` step is needed.  The Forgejo pull-mirror
 handles the entire transfer.
--- a/formulas/collect-engagement.toml
+++ b/formulas/collect-engagement.toml
@ -1,172 +0,0 @@
 # formulas/collect-engagement.toml — Collect website engagement data
 #
 # Daily formula: SSH into Caddy host, fetch access log, parse locally,
 # commit evidence JSON to ops repo via Forgejo API.
 #
 # Triggered by cron in the edge container entrypoint (daily at 23:50 UTC).
 # Design choices from #426: Q1=A (fetch raw log, process locally),
 # Q2=A (direct cron in edge container), Q3=B (dedicated purpose-limited SSH key).
 #
 # Steps: fetch-log → parse-engagement → commit-evidence
 name        = "collect-engagement"
 description = "SSH-fetch Caddy access log, parse engagement metrics, commit evidence"
 version     = 1
 [context]
 files = ["AGENTS.md"]
 [vars.caddy_host]
 description = "SSH host for the Caddy server"
 required    = false
 default     = "${CADDY_SSH_HOST:-disinto.ai}"
 [vars.caddy_user]
 description = "SSH user on the Caddy host"
 required    = false
 default     = "${CADDY_SSH_USER:-debian}"
 [vars.caddy_log_path]
 description = "Path to Caddy access log on the remote host"
 required    = false
 default     = "${CADDY_ACCESS_LOG:-/var/log/caddy/access.log}"
 [vars.local_log_path]
 description = "Local path to store fetched access log"
 required    = false
 default     = "/tmp/caddy-access-log-fetch.log"
 [vars.evidence_dir]
 description = "Evidence output directory in the ops repo"
 required    = false
 default     = "evidence/engagement"
 # ── Step 1: SSH fetch ────────────────────────────────────────────────
 [[steps]]
 id          = "fetch-log"
 title       = "Fetch Caddy access log from remote host via SSH"
 description = """
 Fetch today's Caddy access log segment from the remote host using SCP.
 The SSH key is read from the environment (CADDY_SSH_KEY), which is
 decrypted from secrets/CADDY_SSH_KEY.enc by the edge entrypoint. It is NEVER hardcoded.
 1. Write the SSH key to a temporary file with restricted permissions:
     _ssh_key_file=$(mktemp)
     trap 'rm -f "$_ssh_key_file"' EXIT
     printf '%s\n' "$CADDY_SSH_KEY" > "$_ssh_key_file"
     chmod 0600 "$_ssh_key_file"
 2. Verify connectivity:
     ssh -i "$_ssh_key_file" -o StrictHostKeyChecking=accept-new \
       -o ConnectTimeout=10 -o BatchMode=yes \
       {{caddy_user}}@{{caddy_host}} 'echo ok'
 3. Fetch the access log via scp:
     scp -i "$_ssh_key_file" -o StrictHostKeyChecking=accept-new \
       -o ConnectTimeout=10 -o BatchMode=yes \
       "{{caddy_user}}@{{caddy_host}}:{{caddy_log_path}}" \
       "{{local_log_path}}"
 4. Verify the fetched file is non-empty:
     if [ ! -s "{{local_log_path}}" ]; then
       echo "WARNING: fetched access log is empty — site may have no traffic"
     else
       echo "Fetched $(wc -l < "{{local_log_path}}") lines from {{caddy_host}}"
     fi
 5. Clean up the temporary key file:
     rm -f "$_ssh_key_file"
 """
 # ── Step 2: Parse engagement ─────────────────────────────────────────
 [[steps]]
 id          = "parse-engagement"
 title       = "Run collect-engagement.sh against the local log copy"
 description = """
 Run the engagement parser against the locally fetched access log.
 1. Set CADDY_ACCESS_LOG to point at the local copy so collect-engagement.sh
   reads from it instead of the default path:
     export CADDY_ACCESS_LOG="{{local_log_path}}"
 2. Run the parser:
     bash "$FACTORY_ROOT/site/collect-engagement.sh"
 3. Verify the evidence JSON was written:
     REPORT_DATE=$(date -u +%Y-%m-%d)
     EVIDENCE_FILE="${OPS_REPO_ROOT}/{{evidence_dir}}/${REPORT_DATE}.json"
     if [ -f "$EVIDENCE_FILE" ]; then
       echo "Evidence written: $EVIDENCE_FILE"
       jq . "$EVIDENCE_FILE"
     else
       echo "ERROR: evidence file not found at $EVIDENCE_FILE"
       exit 1
     fi
 4. Clean up the fetched log:
     rm -f "{{local_log_path}}"
 """
 needs       = ["fetch-log"]
 # ── Step 3: Commit evidence ──────────────────────────────────────────
 [[steps]]
 id          = "commit-evidence"
 title       = "Commit evidence JSON to ops repo via Forgejo API"
 description = """
 Commit the dated evidence JSON to the ops repo so the planner can
 consume it during gap analysis.
 1. Read the evidence file:
     REPORT_DATE=$(date -u +%Y-%m-%d)
     EVIDENCE_FILE="${OPS_REPO_ROOT}/{{evidence_dir}}/${REPORT_DATE}.json"
     CONTENT=$(base64 < "$EVIDENCE_FILE")
 2. Check if the file already exists in the ops repo (update vs create):
     OPS_OWNER="${OPS_FORGE_OWNER:-${FORGE_REPO%%/*}}"
     OPS_REPO="${OPS_FORGE_REPO:-${PROJECT_NAME:-disinto}-ops}"
     FILE_PATH="{{evidence_dir}}/${REPORT_DATE}.json"
     EXISTING=$(curl -sf \
       -H "Authorization: token ${FORGE_TOKEN}" \
       "${FORGE_URL}/api/v1/repos/${OPS_OWNER}/${OPS_REPO}/contents/${FILE_PATH}" \
       2>/dev/null || echo "")
 3. Create or update the file via Forgejo API:
     if [ -n "$EXISTING" ] && printf '%s' "$EXISTING" | jq -e '.sha' >/dev/null 2>&1; then
       # Update existing file
       SHA=$(printf '%s' "$EXISTING" | jq -r '.sha')
       curl -sf -X PUT \
         -H "Authorization: token ${FORGE_TOKEN}" \
         -H "Content-Type: application/json" \
         "${FORGE_URL}/api/v1/repos/${OPS_OWNER}/${OPS_REPO}/contents/${FILE_PATH}" \
         -d "$(jq -nc --arg content "$CONTENT" --arg sha "$SHA" --arg msg "evidence: engagement ${REPORT_DATE}" \
           '{message: $msg, content: $content, sha: $sha}')"
       echo "Updated existing evidence file in ops repo"
     else
       # Create new file
       curl -sf -X POST \
         -H "Authorization: token ${FORGE_TOKEN}" \
         -H "Content-Type: application/json" \
         "${FORGE_URL}/api/v1/repos/${OPS_OWNER}/${OPS_REPO}/contents/${FILE_PATH}" \
         -d "$(jq -nc --arg content "$CONTENT" --arg msg "evidence: engagement ${REPORT_DATE}" \
           '{message: $msg, content: $content}')"
       echo "Created evidence file in ops repo"
     fi
 4. Verify the commit landed:
     VERIFY=$(curl -sf \
       -H "Authorization: token ${FORGE_TOKEN}" \
       "${FORGE_URL}/api/v1/repos/${OPS_OWNER}/${OPS_REPO}/contents/${FILE_PATH}" \
       | jq -r '.name // empty')
     if [ "$VERIFY" = "${REPORT_DATE}.json" ]; then
       echo "Evidence committed: ${FILE_PATH}"
     else
       echo "ERROR: could not verify evidence commit"
       exit 1
     fi
 """
 needs       = ["parse-engagement"]
--- a/formulas/rent-a-human-caddy-ssh.toml
+++ b/formulas/rent-a-human-caddy-ssh.toml
@ -1,161 +0,0 @@
 # formulas/rent-a-human-caddy-ssh.toml — Provision SSH key for Caddy log collection
 #
 # "Rent a Human" — walk the operator through provisioning a purpose-limited
 # SSH keypair so collect-engagement.sh can fetch Caddy access logs remotely.
 #
 # The key uses a `command=` restriction so it can ONLY cat the access log.
 # No interactive shell, no port forwarding, no agent forwarding.
 #
 # Parent vision issue: #426
 # Sprint: website-observability-wire-up (ops PR #10)
 # Consumed by: site/collect-engagement.sh (issue #745)
 name        = "rent-a-human-caddy-ssh"
 description = "Provision a purpose-limited SSH keypair for remote Caddy log collection"
 version     = 1
 # ── Step 1: Generate keypair ─────────────────────────────────────────────────
 [[steps]]
 id    = "generate-keypair"
 title = "Generate a dedicated ed25519 keypair"
 description = """
 Generate a purpose-limited SSH keypair for Caddy log collection.
 Run on your local machine (NOT the Caddy host):
 ```
 ssh-keygen -t ed25519 -f caddy-collect -N '' -C 'disinto-collect-engagement'
 ```
 This produces two files:
  - caddy-collect      (private key — goes into the vault)
  - caddy-collect.pub  (public key — goes onto the Caddy host)
 Do NOT set a passphrase (-N '') — the factory runs unattended.
 """
 # ── Step 2: Install public key on Caddy host ─────────────────────────────────
 [[steps]]
 id    = "install-public-key"
 title = "Install the public key on the Caddy host with command= restriction"
 needs = ["generate-keypair"]
 description = """
 Install the public key on the Caddy host with a strict command= restriction
 so this key can ONLY read the access log.
 1. SSH into the Caddy host as the user who owns /var/log/caddy/access.log.
 2. Open (or create) ~/.ssh/authorized_keys:
     mkdir -p ~/.ssh && chmod 700 ~/.ssh
     nano ~/.ssh/authorized_keys
 3. Add this line (all on ONE line — do not wrap):
     command="cat /var/log/caddy/access.log",no-port-forwarding,no-X11-forwarding,no-agent-forwarding ssh-ed25519 AAAA... disinto-collect-engagement
   Replace "AAAA..." with the contents of caddy-collect.pub.
   To build the line automatically:
     echo "command=\"cat /var/log/caddy/access.log\",no-port-forwarding,no-X11-forwarding,no-agent-forwarding $(cat caddy-collect.pub)"
 4. Set permissions:
     chmod 600 ~/.ssh/authorized_keys
 What the restrictions do:
  - command="cat /var/log/caddy/access.log"
      Forces this key to only execute `cat /var/log/caddy/access.log`,
      regardless of what the client requests.
  - no-port-forwarding    — blocks SSH tunnels
  - no-X11-forwarding     — blocks X11
  - no-agent-forwarding   — blocks agent forwarding
 If the access log is at a different path, update the command= restriction
 AND set CADDY_ACCESS_LOG in the factory environment to match.
 """
 # ── Step 3: Add private key to vault secrets ─────────────────────────────────
 [[steps]]
 id    = "store-private-key"
 title = "Add the private key as CADDY_SSH_KEY secret"
 needs = ["generate-keypair"]
 description = """
 Store the private key in the factory's encrypted secrets store.
 1. Add the private key using `disinto secrets add`:
     cat caddy-collect | disinto secrets add CADDY_SSH_KEY
   This encrypts the key with age and stores it as secrets/CADDY_SSH_KEY.enc.
 2. IMPORTANT: After storing, securely delete the local private key file:
     shred -u caddy-collect 2>/dev/null || rm -f caddy-collect
     rm -f caddy-collect.pub
   The public key is already installed on the Caddy host; the private key
   now lives only in secrets/CADDY_SSH_KEY.enc.
 Never commit the private key to any git repository.
 """
 # ── Step 4: Configure Caddy host address ─────────────────────────────────────
 [[steps]]
 id    = "store-caddy-host"
 title = "Add the Caddy host details as secrets"
 needs = ["install-public-key"]
 description = """
 Store the Caddy connection details so collect-engagement.sh knows
 where to SSH.
 1. Add each value using `disinto secrets add`:
     echo 'disinto.ai' | disinto secrets add CADDY_SSH_HOST
     echo 'debian' | disinto secrets add CADDY_SSH_USER
     echo '/var/log/caddy/access.log' | disinto secrets add CADDY_ACCESS_LOG
   Replace values with the actual SSH host, user, and log path for your setup.
 """
 # ── Step 5: Test the connection ──────────────────────────────────────────────
 [[steps]]
 id    = "test-connection"
 title = "Verify the SSH key works and returns the access log"
 needs = ["install-public-key", "store-private-key", "store-caddy-host"]
 description = """
 Test the end-to-end connection before the factory tries to use it.
 1. From the factory host (or anywhere with the private key), run:
     ssh -i caddy-collect -o StrictHostKeyChecking=accept-new user@caddy-host
   Expected behavior:
     - Outputs the contents of /var/log/caddy/access.log
     - Disconnects immediately (command= restriction forces this)
   If you already shredded the local key, decode it from the vault:
     echo "$CADDY_SSH_KEY" | base64 -d > /tmp/caddy-collect-test
     chmod 600 /tmp/caddy-collect-test
     ssh -i /tmp/caddy-collect-test -o StrictHostKeyChecking=accept-new user@caddy-host
     rm -f /tmp/caddy-collect-test
 2. Verify the output is Caddy structured JSON (one JSON object per line):
     ssh -i /tmp/caddy-collect-test user@caddy-host | head -1 | jq .
   You should see fields like: ts, request, status, duration.
 3. If the connection fails:
     - Permission denied → check authorized_keys format (must be one line)
     - Connection refused → check sshd is running on the Caddy host
     - Empty output → check /var/log/caddy/access.log exists and is readable
       by the SSH user
     - "jq: error" → Caddy may be using Combined Log Format instead of
       structured JSON; check Caddy's log configuration
 4. Once verified, the factory's collect-engagement.sh can use this key
   to fetch logs remotely via:
     ssh -i <decoded-key-path> $CADDY_HOST
 """
--- a/formulas/review-pr.toml
+++ b/formulas/review-pr.toml
@ -213,7 +213,7 @@ should file a vault item instead of executing directly.
 **Exceptions** (do NOT flag these):
 - Code inside `vault/` — the vault system itself is allowed to handle secrets
 - References in comments or documentation explaining the architecture
- `bin/disinto` setup commands that manage `secrets/*.enc` and the `run` subcommand
+- `bin/disinto` setup commands that manage `.env.vault.enc` and the `run` subcommand
 - Local operations (git push to forge, forge API calls with `FORGE_TOKEN`)
 ## 6. Re-review (if previous review is provided)
--- a/formulas/run-architect.toml
+++ b/formulas/run-architect.toml
@ -16,14 +16,7 @@
 #            - Bash creates the ops PR with pitch content
 #            - Bash posts the ACCEPT/REJECT footer comment
 #   Step 3: Sprint PR creation with questions (issue #101) (one PR per pitch)
-#   Step 4: Post-merge sub-issue filing via filer-bot (#764)
+#   Step 4: Answer parsing + sub-issue filing (issue #102)
 #
 # Permission model (#764):
 #   architect-bot: READ-ONLY on project repo (GET issues/PRs/labels for context).
 #     Cannot POST/PUT/PATCH/DELETE any project-repo resource.
 #     Write access ONLY on ops repo (branches, PRs, comments).
 #   filer-bot: issues:write on project repo. Files sub-issues from merged sprint
 #     PRs via ops-filer pipeline. Adds in-progress label to vision issues.
 #
 # Architecture:
 # - Bash script (architect-run.sh) handles ALL state management
@ -153,32 +146,15 @@ For each issue in ARCHITECT_TARGET_ISSUES, bash performs:
 ## Recommendation
 <architect's assessment: worth it / defer / alternative approach>
 ## Sub-issues
 <!-- filer:begin -->
 - id: <kebab-case-id>
  title: "vision(#N): <concise sub-issue title>"
  labels: [backlog]
  depends_on: []
  body: |
    ## Goal
    <what this sub-issue accomplishes>
    ## Acceptance criteria
    - [ ] <criterion>
 <!-- filer:end -->
 IMPORTANT: Do NOT include design forks or questions yet. The pitch is a go/no-go
 decision for the human. Questions come only after acceptance.
 The ## Sub-issues block is parsed by the filer-bot pipeline after sprint PR merge.
 Each sub-issue between filer:begin/end markers becomes a Forgejo issue on the
 project repo. The filer appends a decomposed-from marker to each body automatically.
 4. Bash creates PR:
   - Create branch: architect/sprint-{pitch-number}
   - Write sprint spec to sprints/{sprint-slug}.md
   - Create PR with pitch content as body
   - Post footer comment: "Reply ACCEPT to proceed with design questions, or REJECT: <reason> to decline."
-   - NOTE: in-progress label is added by filer-bot after sprint PR merge (#764)
+   - Add in-progress label to vision issue
 Output:
 - One PR per vision issue (up to 3 per run)
@ -209,9 +185,6 @@ This ensures approved PRs don't sit indefinitely without design conversation.
 Architecture:
 - Bash creates PRs during stateless pitch generation (step 2)
 - Model has no role in PR creation — no Forgejo API access
 - architect-bot is READ-ONLY on the project repo (#764) — all project-repo
  writes (sub-issue filing, in-progress label) are handled by filer-bot
  via the ops-filer pipeline after sprint PR merge
 - This step describes the PR format for reference
 PR Format (created by bash):
@ -228,29 +201,64 @@ PR Format (created by bash):
   - Head: architect/sprint-{pitch-number}
   - Footer comment: "Reply ACCEPT to proceed with design questions, or REJECT: <reason> to decline."
 4. Add in-progress label to vision issue:
   - Look up label ID: GET /repos/{owner}/{repo}/labels
   - Add label: POST /repos/{owner}/{repo}/issues/{issue_number}/labels
 After creating all PRs, signal PHASE:done.
 NOTE: in-progress label on the vision issue is added by filer-bot after sprint PR merge (#764).
-## Forgejo API Reference (ops repo only)
+## Forgejo API Reference
-All operations use the ops repo Forgejo API with `Authorization: token ${FORGE_TOKEN}` header.
+All operations use the Forgejo API with Authorization: token ${FORGE_TOKEN} header.
 architect-bot is READ-ONLY on the project repo — cannot POST/PUT/PATCH/DELETE project-repo resources (#764).
-### Create branch (ops repo)
+### Create branch
 ```
-POST /repos/{owner}/{repo-ops}/branches
+POST /repos/{owner}/{repo}/branches
 Body: {"new_branch_name": "architect/<sprint-slug>", "old_branch_name": "main"}
 ```
-### Create/update file (ops repo)
+### Create/update file
 ```
-PUT /repos/{owner}/{repo-ops}/contents/<path>
+PUT /repos/{owner}/{repo}/contents/<path>
 Body: {"message": "sprint: add <sprint-slug>.md", "content": "<base64-encoded-content>", "branch": "architect/<sprint-slug>"}
 ```
-### Create PR (ops repo)
+### Create PR
 ```
-POST /repos/{owner}/{repo-ops}/pulls
+POST /repos/{owner}/{repo}/pulls
 Body: {"title": "architect: <sprint summary>", "body": "<markdown-text>", "head": "architect/<sprint-slug>", "base": "main"}
 ```
 **Important: PR body format**
 - The body field must contain plain markdown text (the raw content from the model)
 - Do NOT JSON-encode or escape the body — pass it as a JSON string value
 - Newlines and markdown formatting (headings, lists, etc.) must be preserved as-is
 ### Add label to issue
 ```
 POST /repos/{owner}/{repo}/issues/{index}/labels
 Body: {"labels": [<label-id>]}
 ```
 ## Forgejo API Reference
 All operations use the Forgejo API with `Authorization: token ${FORGE_TOKEN}` header.
 ### Create branch
 ```
 POST /repos/{owner}/{repo}/branches
 Body: {"new_branch_name": "architect/<sprint-slug>", "old_branch_name": "main"}
 ```
 ### Create/update file
 ```
 PUT /repos/{owner}/{repo}/contents/<path>
 Body: {"message": "sprint: add <sprint-slug>.md", "content": "<base64-encoded-content>", "branch": "architect/<sprint-slug>"}
 ```
 ### Create PR
 ```
 POST /repos/{owner}/{repo}/pulls
 Body: {"title": "architect: <sprint summary>", "body": "<markdown-text>", "head": "architect/<sprint-slug>", "base": "main"}
 ```
@ -259,22 +267,30 @@ Body: {"title": "architect: <sprint summary>", "body": "<markdown-text>", "head"
 - Do NOT JSON-encode or escape the body — pass it as a JSON string value
 - Newlines and markdown formatting (headings, lists, etc.) must be preserved as-is
-### Close PR (ops repo)
+### Close PR
 ```
-PATCH /repos/{owner}/{repo-ops}/pulls/{index}
+PATCH /repos/{owner}/{repo}/pulls/{index}
 Body: {"state": "closed"}
 ```
-### Delete branch (ops repo)
+### Delete branch
 ```
-DELETE /repos/{owner}/{repo-ops}/git/branches/<branch-name>
+DELETE /repos/{owner}/{repo}/git/branches/<branch-name>
 ```
-### Read-only on project repo (context gathering)
+### Get labels (look up label IDs by name)
 ```
-GET /repos/{owner}/{repo}/issues          — list issues
+GET /repos/{owner}/{repo}/labels
-GET /repos/{owner}/{repo}/issues/{number} — read issue details
+```
-GET /repos/{owner}/{repo}/labels          — list labels
+
-GET /repos/{owner}/{repo}/pulls           — list PRs
+### Add label to issue (for in-progress on vision issue)
 ```
 POST /repos/{owner}/{repo}/issues/{index}/labels
 Body: {"labels": [<label-id>]}
 ```
 ### Remove label from issue (for in-progress removal on REJECT)
 ```
 DELETE /repos/{owner}/{repo}/issues/{index}/labels/{label-id}
 ```
 """
--- a/formulas/run-gardener.toml
+++ b/formulas/run-gardener.toml
@ -177,7 +177,7 @@ DUST (trivial — single-line edit, rename, comment, style, whitespace):
 VAULT (needs human decision or external resource):
  File a vault procurement item using vault_request():
-    source "$(dirname "$0")/../lib/action-vault.sh"
+    source "$(dirname "$0")/../lib/vault.sh"
    TOML_CONTENT="# Vault action: <action_id>
 context = \"<description of what decision/resource is needed>\"
 unblocks = [\"#NNN\"]
--- a/formulas/run-planner.toml
+++ b/formulas/run-planner.toml
@ -243,7 +243,7 @@ needs = ["preflight"]
 [[steps]]
 id    = "commit-ops-changes"
-title = "Write tree, memory, and journal; commit and push branch"
+title = "Write tree, memory, and journal; commit and push"
 description = """
 ### 1. Write prerequisite tree
 Write to: $OPS_REPO_ROOT/prerequisites.md
@ -256,16 +256,14 @@ If (count - N) >= 5 or planner-memory.md missing, write to:
 Include: run counter marker, date, constraint focus, patterns, direction.
 Keep under 100 lines. Replace entire file.
-### 3. Commit ops repo changes to the planner branch
+### 3. Commit ops repo changes
-Commit the ops repo changes (prerequisites, memory, vault items) and push the
+Commit the ops repo changes (prerequisites, memory, vault items):
 branch. Do NOT push directly to $PRIMARY_BRANCH — planner-run.sh will create a
 PR and walk it to merge via review-bot.
  cd "$OPS_REPO_ROOT"
  git add prerequisites.md knowledge/planner-memory.md vault/pending/
  git add -u
  if ! git diff --cached --quiet; then
    git commit -m "chore: planner run $(date -u +%Y-%m-%d)"
-    git push origin HEAD
+    git push origin "$PRIMARY_BRANCH"
  fi
  cd "$PROJECT_REPO_ROOT"
--- a/formulas/run-predictor.toml
+++ b/formulas/run-predictor.toml
@ -125,8 +125,8 @@ For each weakness you identify, choose one:
  The prediction explains the theory. The vault PR triggers the proof
  after human approval. When the planner runs next, evidence is already there.
-  Vault dispatch (requires lib/action-vault.sh):
+  Vault dispatch (requires lib/vault.sh):
-    source "$PROJECT_REPO_ROOT/lib/action-vault.sh"
+    source "$PROJECT_REPO_ROOT/lib/vault.sh"
    TOML_CONTENT="id = \"predict-<prediction_number>-<formula>\"
 context = \"Test prediction #<prediction_number>: <theory summary> — focus: <specific test>\"
@ -154,7 +154,7 @@ tea is pre-configured with login "$TEA_LOGIN" and repo "$FORGE_REPO".
       --title "<title>" --body "<body>" --labels "prediction/unreviewed"
 2. Dispatch formula via vault (if exploiting):
-     source "$PROJECT_REPO_ROOT/lib/action-vault.sh"
+     source "$PROJECT_REPO_ROOT/lib/vault.sh"
     PR_NUM=$(vault_request "predict-NNN-<formula>" "$TOML_CONTENT")
     # See EXPLOIT section above for TOML_CONTENT format
--- a/gardener/AGENTS.md
+++ b/gardener/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: c51cc9dba649ed543b910b561231a5c8bd2130bc -->
 # Gardener Agent
 **Role**: Backlog grooming — detect duplicate issues, missing acceptance
@ -32,7 +32,7 @@ the gardener runs as part of the polling loop alongside the planner, predictor,
  PR, reviewed alongside AGENTS.md changes, executed by gardener-run.sh after merge.
 **Environment variables consumed**:
- `FORGE_TOKEN`, `FORGE_GARDENER_TOKEN` (falls back to FORGE_TOKEN), `FORGE_REPO`, `FORGE_API`, `PROJECT_NAME`, `PROJECT_REPO_ROOT`. `FORGE_TOKEN_OVERRIDE` is exported to `$FORGE_GARDENER_TOKEN` before sourcing env.sh so the gardener-bot identity survives re-sourcing (#762).
+- `FORGE_TOKEN`, `FORGE_GARDENER_TOKEN` (falls back to FORGE_TOKEN), `FORGE_REPO`, `FORGE_API`, `PROJECT_NAME`, `PROJECT_REPO_ROOT`
 - `PRIMARY_BRANCH`, `CLAUDE_MODEL` (set to sonnet by gardener-run.sh)
 **Lifecycle**: gardener-run.sh (invoked by polling loop every 6h, `check_active gardener`) →
--- a/gardener/gardener-run.sh
+++ b/gardener/gardener-run.sh
@ -26,11 +26,10 @@ FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
 # Accept project config from argument; default to disinto
 export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 # Set override BEFORE sourcing env.sh so it survives any later re-source of
 # env.sh from nested shells / claude -p tools (#762, #747)
 export FORGE_TOKEN_OVERRIDE="${FORGE_GARDENER_TOKEN:-}"
 # shellcheck source=../lib/env.sh
 source "$FACTORY_ROOT/lib/env.sh"
 # Use gardener-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_GARDENER_TOKEN:-${FORGE_TOKEN}}"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
--- a/gardener/pending-actions.json
+++ b/gardener/pending-actions.json
@ -1,12 +1,37 @@
 [
  {
-    "action": "comment",
+    "action": "edit_body",
-    "issue": 623,
+    "issue": 707,
-    "body": "**Dependency check:** All blocking dependencies are now closed:\n- #620 ✓ closed\n- #621 ✓ closed  \n- #622 ✓ closed\n\nPer the issue description: *\"Once #620/#621/#622 are green, this issue should fork into at least three backlog children: subpath routing + Forgejo ROOT_URL / Woodpecker HOST, disinto-chat container scaffold with OAuth gate, and Claude Code sandbox envelope + working-dir scoping.\"*\n\nThis vision issue is ready for the planner to decompose into backlog children."
+    "body": "## Goal\n\nGive `disinto-chat` its own Claude identity mount so its OAuth refresh races cannot corrupt the factory agents' shared `~/.claude` credentials. Default to a separate `~/.claude-chat/` on the host; support `ANTHROPIC_API_KEY` as a fallback that skips OAuth entirely.\n\n## Why\n\n- #623 root-caused this: Claude Code's internal refresh lock in `~/.claude.lock` operates outside bind-mounted directories, so two containers sharing `~/.claude` can race during token refresh and invalidate each other. The factory has already had OAuth expiry incidents traced to multiple agents sharing credentials.\n- Scoping chat to its own identity dir means chat can be logged in as a different Anthropic account, or pinned to an API key, without touching agent credentials.\n\n## Scope\n\n### Files to touch\n\n- `lib/generators.sh` chat service block (from #705):\n  - Replace the throwaway named volume with `${CHAT_CLAUDE_DIR:-${HOME}/.claude-chat}:/home/chat/.claude-chat`.\n  - Env: `CLAUDE_CONFIG_DIR=/home/chat/.claude-chat/config`, `CLAUDE_CREDENTIALS_DIR=/home/chat/.claude-chat/config/credentials`.\n  - Conditional: if `ANTHROPIC_API_KEY` is set in `.env`, pass it through and **do not** mount `~/.claude-chat` at all (no credentials on disk in that mode).\n- `bin/disinto disinto_init()` — after #620's admin password prompt, add an optional prompt: `Use separate Anthropic identity for chat? (y/N)`. On yes, create `~/.claude-chat/` and invoke `claude login` in a subshell with `CLAUDE_CONFIG_DIR=~/.claude-chat/config`.\n- `lib/claude-config.sh` — factor out the existing `~/.claude` setup logic so a non-default `CLAUDE_CONFIG_DIR` is a first-class parameter. If it is already parameterised, just document it; if not, extract a helper `setup_claude_dir <dir>` and have the existing path call it with the default dir.\n- `docker/chat/Dockerfile` — declare `VOLUME /home/chat/.claude-chat`, set owner to the non-root chat user introduced in #706.\n\n### Out of scope\n\n- Cross-session lock coherence for multiple concurrent chat containers (single-chat-container assumption is fine for MVP).\n- Anthropic team / workspace support — single identity is enough.\n\n## Acceptance\n\n- [ ] Fresh `disinto init` with \"use separate chat identity\" answered yes creates `~/.claude-chat/` and logs in successfully.\n- [ ] With `ANTHROPIC_API_KEY=sk-ant-...` set in `.env`, chat starts without any `~/.claude-chat` mount (verified via `docker inspect disinto-chat`) and successfully completes a test prompt.\n- [ ] Running the factory agents AND chat simultaneously for 24h does not produce any OAuth refresh failures on either side (manual soak test — document result in PR).\n- [ ] `CLAUDE_CONFIG_DIR` and `CLAUDE_CREDENTIALS_DIR` inside the chat container resolve to `/home/chat/.claude-chat/config*`, not the shared factory path.\n\n## Depends on\n\n- #705 (chat scaffold).\n- #620 (admin password prompt — same init flow this adds a step to).\n\n## Notes\n\n- The factory's existing shared mount is `/var/lib/disinto/claude-shared` (see `lib/generators.sh:113,327,381,426`). Chat must NOT use this path.\n- `flock(\"${HOME}/.claude/session.lock\")` logic mentioned in #623 is load-bearing, not redundant — do not \"simplify\" it.\n- Prefer the API-key path for anyone running the factory on shared hardware; call this out in README updates.\n\n## Boundaries for dev-agent\n\n- Do not try to make chat share `~/.claude` with the agents \"just for convenience\". The whole point of this chunk is the opposite.\n- Do not add a third claude config dir. One for agents, one for chat, done.\n- Do not refactor `lib/claude-config.sh` beyond extracting a parameterised helper if needed.\n- Parent vision: #623.\n\n## Affected files\n- `lib/generators.sh` — chat service bind mount and env vars for CLAUDE_CONFIG_DIR\n- `bin/disinto` — disinto_init() prompt for separate chat identity\n- `lib/claude-config.sh` — factor out claude dir setup (new helper setup_claude_dir)\n- `docker/chat/Dockerfile` — declare VOLUME /home/chat/.claude-chat"
  },
  {
-    "action": "comment",
+    "action": "edit_body",
-    "issue": 758,
+    "issue": 708,
-    "body": "**Gardener flag:** This issue requires human admin action on Forgejo to resolve — changing branch protection settings on the ops repo. No automated formula can fix Forgejo admin settings.\n\nProposed options (from issue body):\n1. Add `planner-bot` to the merge whitelist in ops repo branch protection\n2. Remove branch protection from the ops repo (agents are primary writers)\n3. Create an admin-level service token for agents\n\nThis is blocking all ops repo writes (planner knowledge, sprint artifacts, vault items)."
+    "body": "## Goal\n\nGate `/chat/*` behind Forgejo OAuth. Register a second OAuth2 app on the internal Forgejo (alongside the existing Woodpecker one), implement the server-side authorization-code flow in the chat backend, and enforce that the logged-in user is `disinto-admin` (or a member of a configured allowlist).\n\n## Why\n\n- #623: the single `disinto-admin` password (bootstrap secret from #620) is the only auth credential; chat must reuse it via Forgejo OAuth, not invent a second password.\n- Keeps attack surface flat: exactly one identity provider for forge, CI, and chat.\n\n## Scope\n\n### Files to touch\n\n- `lib/ci-setup.sh` — generalise the Woodpecker OAuth-app creation helper `_create_woodpecker_oauth_impl()` (at `lib/ci-setup.sh:96`) into `_create_forgejo_oauth_app <name> <redirect_uri>` that both Woodpecker and chat can call. The existing Woodpecker callsite becomes a thin wrapper; no behaviour change for Woodpecker.\n- `bin/disinto disinto_init()` — after Woodpecker OAuth creation (around `bin/disinto:847`), add a call to create the chat OAuth app with redirect URI `https://${EDGE_TUNNEL_FQDN}/chat/oauth/callback`. Write `CHAT_OAUTH_CLIENT_ID` / `CHAT_OAUTH_CLIENT_SECRET` to `.env`.\n- `docker/chat/server.{py,go}` — new routes:\n  - `GET /chat/login` → 302 to Forgejo `/login/oauth/authorize?client_id=...&state=...`.\n  - `GET /chat/oauth/callback` → exchange code for token, fetch `/api/v1/user`, assert `login == \"disinto-admin\"` (or `DISINTO_CHAT_ALLOWED_USERS` CSV if set), set an `HttpOnly` session cookie, 302 to `/chat/`.\n  - Any other route with no valid session → 302 to `/chat/login`.\n  - Session store: in-memory map keyed by random token; TTL 24h; no persistence across container restarts (by design — forces re-auth after deploy).\n- `lib/generators.sh` chat service env: pass `FORGE_URL`, `CHAT_OAUTH_CLIENT_ID`, `CHAT_OAUTH_CLIENT_SECRET`, `EDGE_TUNNEL_FQDN`, `DISINTO_CHAT_ALLOWED_USERS`.\n\n### Out of scope\n\n- Defense-in-depth `Remote-User` header check — #709.\n- Team membership lookup via Forgejo API — start with a plain user allowlist, add teams later.\n- CSRF protection on `POST /chat` beyond the session cookie — add only if soak testing reveals a need.\n\n## Acceptance\n\n- [ ] `disinto init` on a fresh project creates TWO OAuth apps on Forgejo (woodpecker + chat), both visible in Forgejo admin UI.\n- [ ] `curl -c cookies -L http://edge/chat/` follows through the OAuth flow when seeded with `disinto-admin` credentials and lands on the chat UI.\n- [ ] Same flow as a non-admin Forgejo user lands on a \"not authorised\" page with a 403.\n- [ ] Expired / missing session cookie redirects to `/chat/login`.\n- [ ] `DISINTO_CHAT_ALLOWED_USERS=alice,bob` permits those users in addition to `disinto-admin`.\n\n## Depends on\n\n- #705 (chat scaffold).\n- #620 (admin password prompt — the password this auth leans on).\n\n## Notes\n\n- The existing Woodpecker OAuth exchange at `lib/ci-setup.sh:273-294` is the reference for the server-side code→token exchange. Read it before implementing; do not guess the Forgejo OAuth flow from docs.\n- Forgejo OAuth2 endpoints are `/login/oauth/authorize` and `/login/oauth/access_token`. Do not hit `/api/v1/...` for OAuth — that is the REST API, not the OAuth endpoints.\n- Session cookie must be `Secure; HttpOnly; SameSite=Lax` in prod; permit non-secure in local dev when `EDGE_TUNNEL_FQDN` is unset.\n\n## Boundaries for dev-agent\n\n- Do not add a new OAuth library dependency if the standard library of the chosen backend language has HTTP client + JSON — the flow is two HTTP calls.\n- Do not reuse the Woodpecker OAuth app. Create a second one with a distinct name and redirect URI. They are different principals.\n- Do not persist sessions to disk. In-memory is correct for MVP; persistence is a separate conversation.\n- Do not implement team membership this chunk. Static allowlist first; teams later.\n- Parent vision: #623.\n\n## Affected files\n- `lib/ci-setup.sh` — generalise _create_woodpecker_oauth_impl() into _create_forgejo_oauth_app helper\n- `bin/disinto` — disinto_init() to create chat OAuth app and write CHAT_OAUTH_CLIENT_ID/SECRET to .env\n- `docker/chat/server.{py,go}` — new OAuth routes: /chat/login, /chat/oauth/callback, session middleware\n- `lib/generators.sh` — chat service env vars: FORGE_URL, CHAT_OAUTH_CLIENT_ID, CHAT_OAUTH_CLIENT_SECRET, EDGE_TUNNEL_FQDN, DISINTO_CHAT_ALLOWED_USERS"
  },
  {
    "action": "edit_body",
    "issue": 709,
    "body": "## Goal\n\nAdd a second, independent auth check on top of #708: Caddy injects an `X-Forwarded-User` header from the validated Forgejo session, and the chat backend refuses any request whose session cookie disagrees with the header. This is the belt to #708's braces.\n\n## Why\n\n- #623 explicitly calls this out as defense-in-depth. If the chat backend session logic has a bug (forged cookie, state confusion), a correctly-configured Caddy `forward_auth` layer catches it — and vice versa.\n- Cheap to add on top of #704 and #708; expensive to bolt on after an incident.\n\n## Scope\n\n### Files to touch\n\n- `docker/Caddyfile` — the `/chat/*` block:\n  - Add `forward_auth chat:8080 { uri /chat/auth/verify; copy_headers X-Forwarded-User }`.\n  - Requests without a valid session are forwarded to `/chat/login` by chat itself; `forward_auth` just stamps the header when there is one.\n- `docker/chat/server.{py,go}`:\n  - New route `GET /chat/auth/verify` — reads the session cookie, returns 200 + `X-Forwarded-User: <login>` if valid, 401 otherwise.\n  - On `POST /chat` and other authenticated routes: read `X-Forwarded-User`, read the session cookie, assert both resolve to the same user. On mismatch: log a warning with the request ID and return 403.\n\n### Out of scope\n\n- Rewriting the session store. The verify endpoint reads the same in-memory map #708 introduced.\n\n## Acceptance\n\n- [ ] `curl http://edge/chat/` with a valid session cookie still works; chat backend logs show `X-Forwarded-User` matching the cookie user.\n- [ ] Editing the session cookie client-side to impersonate another user while keeping the forged cookie valid triggers a 403 with a clear log line (simulate by swapping cookies mid-session).\n- [ ] Removing the `forward_auth` block from Caddyfile and restarting causes the chat backend to fail-closed (all authenticated routes 403) — documented as the intended failure mode.\n- [ ] The verify endpoint does not accept arbitrary external requests from outside Caddy: the chat backend rejects calls to `/chat/auth/verify` that lack a shared-secret header (or whose origin IP is not the edge container).\n\n## Depends on\n\n- #704 (Caddy subpath routing).\n- #708 (chat OAuth gate — provides the session store this chunk reads).\n\n## Notes\n\n- Caddy `forward_auth` reference: https://caddyserver.com/docs/caddyfile/directives/forward_auth — stick to the documented directives, do not hand-roll header passing.\n- If network-level origin validation on `/chat/auth/verify` is fiddly, a shared-secret header between Caddy and chat is acceptable — but prefer network-level if possible.\n\n## Boundaries for dev-agent\n\n- Do not replace the #708 session store with something new. Read it, do not rewrite it.\n- Do not push the entire auth decision into Caddy. The chat backend is still the source of truth; Caddy adds a redundant check.\n- Parent vision: #623.\n\n## Affected files\n- `docker/Caddyfile` — add forward_auth block to /chat/* with X-Forwarded-User header copy\n- `docker/chat/server.{py,go}` — new GET /chat/auth/verify route; X-Forwarded-User vs session cookie check on authenticated routes"
  },
  {
    "action": "edit_body",
    "issue": 710,
    "body": "## Goal\n\nDecide and implement a conversation history persistence model for `disinto-chat`. MVP target: append-only per-user NDJSON files on a bind-mounted host volume, one file per conversation, with a simple history list endpoint and sidebar in the UI.\n\n## Why\n\n- Without history, every page refresh loses context. Claude is stateless per invocation; the chat UI is what makes it feel like a conversation.\n- A full database with search is overkill for a personal / small-team factory (#623 security posture). Flat files are enough and recoverable by `cat`.\n\n## Scope\n\n### Files to touch\n\n- `lib/generators.sh` chat service:\n  - Add a writable bind mount `${CHAT_HISTORY_DIR:-./state/chat-history}:/var/lib/chat/history` (one per-project host path; compose already pins the project root).\n  - Must coexist with #706's read-only rootfs (this is a separate mount, not part of rootfs — sanity-check the sandbox verify script still passes).\n- `docker/chat/server.{py,go}`:\n  - On each `POST /chat`, append one NDJSON line `{ts, user, role, content}` to `/var/lib/chat/history/<user>/<conversation_id>.ndjson`.\n  - `GET /chat/history` → returns the list of conversation ids and first-message previews for the logged-in user.\n  - `GET /chat/history/<id>` → returns the full conversation for the logged-in user; 404 if the file belongs to another user.\n  - New conversation: `POST /chat/new` → generates a fresh conversation_id (random 12-char hex) and returns it.\n  - UI: sidebar with conversation list, \"new chat\" button, load history into the log on click.\n- File naming: `<user>/<conversation_id>.ndjson` — user-scoped directory prevents cross-user leakage even if a bug leaks ids. `conversation_id` must match `^[0-9a-f]{12}$`, no slashes allowed.\n\n### Out of scope\n\n- Full-text search.\n- Database / SQLite.\n- History retention / rotation — unbounded for now.\n\n### In scope explicitly\n\n- Replaying prior turns back into the `claude --print` subprocess for follow-up turns: the backend must feed the prior NDJSON lines back into claude via whatever convention the agent code uses. Cross-check `docker/agents/entrypoint.sh` for how agents pass conversation state.\n\n## Acceptance\n\n- [ ] Sending 3 messages, refreshing the page, and clicking the conversation in the sidebar re-loads all 3 messages.\n- [ ] A new conversation starts with an empty context and does not see prior messages.\n- [ ] `ls state/chat-history/disinto-admin/` on the host shows one NDJSON file per conversation, each line is valid JSON.\n- [ ] A second user logging in via the #708 allowlist sees only their own conversations.\n- [ ] History endpoints are blocked for unauthenticated requests (inherits #708 / #709 auth).\n\n## Depends on\n\n- #705 (chat scaffold).\n\n## Notes\n\n- NDJSON, not JSON-array: append is O(1) and partial writes never corrupt prior lines. Mirrors the factory's CI log format at `lib/ci-log-reader.py`.\n- Per-user directory, not a single shared dir — path traversal via a crafted `conversation_id` is the main risk. The strict regex above is the mitigation.\n\n## Boundaries for dev-agent\n\n- Do not add SQLite, Postgres, or any database. Files.\n- Do not invent a conversation replay system. Whatever `claude --print` / the agents already do for context is the baseline — match it.\n- Do not store history inside the container's tmpfs — it has to survive container restarts.\n- Parent vision: #623.\n\n## Affected files\n- `lib/generators.sh` — chat service bind mount CHAT_HISTORY_DIR:/var/lib/chat/history\n- `docker/chat/server.{py,go}` — NDJSON append on POST /chat; GET /chat/history; GET /chat/history/<id>; POST /chat/new"
  },
  {
    "action": "edit_body",
    "issue": 711,
    "body": "## Goal\n\nAdd per-user cost and request caps to `disinto-chat` so a compromised session (or a wedged browser tab firing requests in a loop) cannot run up an unbounded Anthropic bill or starve the agents' token budget.\n\n## Why\n\n- #623 \"Open questions\" explicitly calls this out. Chat is the only user-facing surface that spawns Claude on demand; no other factory surface does.\n- Cheap to enforce (counter + bash-style dict), expensive to forget.\n\n## Scope\n\n### Files to touch\n\n- `docker/chat/server.{py,go}`:\n  - Per-user sliding-window request counter: `CHAT_MAX_REQUESTS_PER_HOUR` (default `60`), `CHAT_MAX_REQUESTS_PER_DAY` (default `500`).\n  - Per-user token-cost counter: after each `claude --print`, parse the final `usage` event from `--output-format stream-json` if present; track cumulative tokens per day; reject if over `CHAT_MAX_TOKENS_PER_DAY` (default `1000000`).\n  - Counters stored in-memory; reset on container restart (acceptable for MVP; file-based persistence is a follow-up).\n  - Rejection response: 429 with `Retry-After` header and a friendly HTMX fragment explaining which cap was hit.\n- `lib/generators.sh` chat env: expose the three caps as overridable env vars with sane defaults baked in.\n\n### Out of scope\n\n- Billing dashboard.\n- Cross-container token budget coordination with the agents.\n- Cost tracking via Anthropic's billing API (not stable enough to depend on).\n\n## Acceptance\n\n- [ ] Sending 61 requests in an hour trips the hourly cap and returns 429 with `Retry-After: <seconds>`.\n- [ ] A single large completion that pushes daily tokens over the cap blocks the *next* request, not the current one (atomic check-then-consume is OK to skip for MVP).\n- [ ] Resetting the container clears counters (verified manually).\n- [ ] Caps are configurable via `.env` without rebuilding the image.\n\n## Depends on\n\n- #705 (chat scaffold).\n\n## Notes\n\n- Token accounting from `claude --print`: the stream-json mode emits a final `usage` event. If that event is absent or its format changes, fall back to a coarse request count only — do not block the user on parsing failures.\n- `Retry-After` must be an integer seconds value, not an HTTP-date, for HTMX to handle it cleanly client-side.\n\n## Boundaries for dev-agent\n\n- Do not add a rate-limiting library. A dict + timestamp list is sufficient for three counters.\n- Do not persist counters to disk this chunk. In-memory is the contract.\n- Do not block requests on Anthropic's own rate limiter. That is retried by `claude` itself; this layer is about *cost*, not throttling.\n- Parent vision: #623.\n\n## Affected files\n- `docker/chat/server.{py,go}` — per-user sliding-window request counter and token-cost counter; 429 rejection with Retry-After header\n- `lib/generators.sh` — chat env: CHAT_MAX_REQUESTS_PER_HOUR, CHAT_MAX_REQUESTS_PER_DAY, CHAT_MAX_TOKENS_PER_DAY"
  },
  {
    "action": "edit_body",
    "issue": 712,
    "body": "## Goal\n\nLet `disinto-chat` perform scoped write actions against the factory — specifically: trigger a Woodpecker CI run, create a Forgejo issue, create a Forgejo PR — via explicit backend endpoints. The UI surfaces these as buttons the user clicks from a chat turn that proposes an action. The model never holds API tokens directly.\n\n## Why\n\n- #623 lists these escalations as the difference between \"chat that talks about the project\" and \"chat that moves the project forward\".\n- Routing through explicit backend endpoints (instead of giving the sandboxed claude process API tokens) keeps the trust model tight: the *user* authorises each action, not the model.\n\n## Scope\n\n### Files to touch\n\n- `docker/chat/server.{py,go}` — new authenticated endpoints (reuse #708 / #709 session check):\n  - `POST /chat/action/ci-run` — body `{repo, branch}` → calls Woodpecker API with `WOODPECKER_TOKEN` (already in `.env` from existing factory setup) to trigger a pipeline.\n  - `POST /chat/action/issue-create` — body `{title, body, labels}` → calls Forgejo API `/repos/<owner>/<repo>/issues` with `FORGE_TOKEN`.\n  - `POST /chat/action/pr-create` — body `{head, base, title, body}` → calls `/repos/<owner>/<repo>/pulls`.\n  - All actions record to #710's NDJSON history as `{role: \"action\", ...}` lines.\n- `docker/chat/ui/index.html` — small HTMX pattern: when claude's response contains a marker like `<action type=\"issue-create\">{...}</action>`, render a clickable button below the message; clicking POSTs to `/chat/action/<type>` with the payload.\n- `lib/generators.sh` chat env: pass `WOODPECKER_TOKEN`, `FORGE_TOKEN`, `FORGE_URL`, `FORGE_OWNER`, `FORGE_REPO`.\n\n### Out of scope\n\n- Destructive actions (branch delete, force push, secret rotation) — deliberately excluded.\n- Multi-step workflows / approval chains.\n- Arbitrary code execution in the chat container (that is what the agents exist for).\n\n## Acceptance\n\n- [ ] A chat turn that emits an `<action type=\"issue-create\">{...}</action>` block renders a button; clicking it creates an issue on Forgejo, visible via the API.\n- [ ] CI-trigger action creates a Woodpecker pipeline that can be seen in the CI UI.\n- [ ] PR-create action produces a Forgejo PR with the specified head / base.\n- [ ] All three actions are logged into the #710 history file with role `action` and the response from the API call.\n- [ ] Unauthenticated requests to `/chat/action/*` return 401 (inherits #708 gate).\n\n## Depends on\n\n- #708 (OAuth gate — actions are authorised by the logged-in user).\n- #710 (history — actions need to be logged alongside chat turns).\n\n## Notes\n\n- Forgejo API auth: the factory's `FORGE_TOKEN` is a long-lived admin token. For MVP, reuse it; a follow-up issue can scope it down to per-user Forgejo tokens derived from the OAuth flow.\n- Woodpecker API is at `http://woodpecker:8000/api/...`, reachable via the compose network — no need to go through the edge container.\n- The `<action>` marker is deliberately simple markup the model can emit in its response text. Do not implement tool-calling protocol; do not spin up an MCP server.\n\n## Boundaries for dev-agent\n\n- Do not give the claude subprocess direct API tokens. The chat backend holds them; the model only emits action markers the user clicks.\n- Do not add destructive actions (delete, force-push). Additive only.\n- Do not invent a new markup format beyond `<action type=\"...\">{JSON}</action>`.\n- Parent vision: #623.\n\n## Affected files\n- `docker/chat/server.{py,go}` — new endpoints: POST /chat/action/ci-run, POST /chat/action/issue-create, POST /chat/action/pr-create\n- `docker/chat/ui/index.html` — HTMX pattern for action buttons triggered by <action type=\"...\"> markers\n- `lib/generators.sh` — chat env: WOODPECKER_TOKEN, FORGE_TOKEN, FORGE_URL, FORGE_OWNER, FORGE_REPO"
  },
  {
    "action": "edit_body",
    "issue": 713,
    "body": "## Goal\n\nContingency track: if the subpath routing + Forgejo OAuth combination from #704 and #708 proves unworkable (redirect loops, Forgejo `ROOT_URL` quirks, etc.), provide a documented fallback using per-service subdomains (`forge.<project>.disinto.ai`, `ci.<project>.disinto.ai`, `chat.<project>.disinto.ai`) under the same wildcard cert.\n\n## Why\n\n- #623 Scope Highlights mentions this as the fallback if subpath OAuth fails.\n- Documenting the fallback up front means we can pivot without a days-long investigation when the subpath approach hits a wall.\n- The wildcard cert from #621 already covers `*.disinto.ai` at no extra cost.\n\n## Scope\n\nThis issue is a **plan + small toggle**, not a full implementation. Implementation only happens if #704 or #708 get stuck.\n\n### Files to touch\n\n- `docs/edge-routing-fallback.md` (new) — documents the fallback topology, diffing concretely against #704 / #708:\n  - Caddyfile: four separate host blocks (`<project>.disinto.ai`, `forge.<project>...`, `ci.<project>...`, `chat.<project>...`), each a single `reverse_proxy` to the container.\n  - Forgejo `ROOT_URL` becomes `https://forge.<project>.disinto.ai/` (root path, not subpath).\n  - Woodpecker `WOODPECKER_HOST` becomes `https://ci.<project>.disinto.ai`.\n  - OAuth redirect URIs (chat, woodpecker) become sub-subdomain paths.\n  - DNS: all handled by the existing wildcard; no new records.\n- `lib/generators.sh` — no code change until pivot; document the env vars that would need to change (`EDGE_TUNNEL_FQDN_FORGE`, etc.) in a comment near `generate_compose`.\n- `tools/edge-control/register.sh` (from #621) — leave a TODO comment noting the fallback shape would need an additional subdomain parameter per project.\n\n### Out of scope (unless pivot)\n\n- Actually implementing the fallback — gated on #704 / #708 failing.\n\n## Acceptance\n\n- [ ] `docs/edge-routing-fallback.md` exists and is concrete enough that a follow-up PR to pivot would take under a day.\n- [ ] The doc names exactly which files / lines each pivot would touch (Caddyfile, `lib/generators.sh`, `lib/ci-setup.sh` redirect URI).\n- [ ] A pivot decision criterion is written into the doc: \"pivot if <specific symptom>, not if <symptom with a known fix>\".\n\n## Depends on\n\n- None — can be written in parallel to #704 / #708.\n\n## Notes\n\n- Keep the doc short. This is a pressure-release valve, not a parallel architecture.\n- Whichever chunk is implementing subpaths first should update this doc if they hit a blocker so the pivot decision is informed.\n\n## Boundaries for dev-agent\n\n- This is a documentation chunk. Do not implement the fallback unless someone explicitly says to pivot.\n- Do not make the main chunks \"fallback-ready\" — that is over-engineering for a contingency.\n- Parent vision: #623.\n\n## Affected files\n- `docs/edge-routing-fallback.md` (new) — fallback topology doc with Caddyfile, Forgejo ROOT_URL, Woodpecker HOST, OAuth redirect URI changes\n- `lib/generators.sh` — comment near generate_compose documenting env vars that change on pivot\n- `tools/edge-control/register.sh` — TODO comment noting fallback shape needs additional subdomain parameter"
  }
 ]
--- a/lib/AGENTS.md
+++ b/lib/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Shared Helpers (`lib/`)
 All agents source `lib/env.sh` as their first action. Additional helpers are
@ -6,7 +6,7 @@ sourced as needed.
 | File | What it provides | Sourced by |
 |---|---|---|
-| `lib/env.sh` | Loads `.env`, sets `FACTORY_ROOT`, exports project config (`FORGE_REPO`, `PROJECT_NAME`, etc.), defines `log()`, `forge_api()`, `forge_api_all()` (paginates all pages; accepts optional second TOKEN parameter, defaults to `$FORGE_TOKEN`; handles invalid/empty JSON responses gracefully — returns empty on parse error instead of crashing), `woodpecker_api()`, `wpdb()`, `memory_guard()` (skips agent if RAM < threshold), `load_secret()` (secret-source abstraction — see below). Auto-loads project TOML if `PROJECT_TOML` is set. Exports per-agent tokens (`FORGE_PLANNER_TOKEN`, `FORGE_GARDENER_TOKEN`, `FORGE_VAULT_TOKEN`, `FORGE_SUPERVISOR_TOKEN`, `FORGE_PREDICTOR_TOKEN`) — each falls back to `$FORGE_TOKEN` if not set. **Vault-only token guard (AD-006)**: `unset GITHUB_TOKEN CLAWHUB_TOKEN` so agents never hold external-action tokens — only the runner container receives them. **Container note**: when `DISINTO_CONTAINER=1`, `.env` is NOT re-sourced — compose already injects env vars (including `FORGE_URL=http://forgejo:3000`) and re-sourcing would clobber them. **Save/restore scope (#364)**: only `FORGE_URL` is preserved across `.env` re-sourcing (compose injects `http://forgejo:3000`, `.env` has `http://localhost:3000`). `FORGE_TOKEN` is NOT preserved so refreshed tokens in `.env` take effect immediately. **Per-agent token override (#762)**: agent run scripts export `FORGE_TOKEN_OVERRIDE=<agent-specific-token>` BEFORE sourcing `env.sh`; `env.sh` applies this override at lines 98-100, ensuring the correct identity survives any re-sourcing of `env.sh` by nested shells or `claude -p` invocations. **Required env var**: `FORGE_PASS` — bot password for git HTTP push (Forgejo 11.x rejects API tokens for `git push`, #361). **Hard preconditions (#674)**: `USER` and `HOME` must be exported by the entrypoint before sourcing. When `PROJECT_TOML` is set, `PROJECT_REPO_ROOT`, `PRIMARY_BRANCH`, and `OPS_REPO_ROOT` must also be set (by entrypoint or TOML). **`load_secret NAME [DEFAULT]` (#793)**: backend-agnostic secret resolution. Precedence: (1) `/secrets/<NAME>.env` — Nomad-rendered template, (2) current environment — already set by `.env.enc` / compose, (3) `secrets/<NAME>.enc` — age-encrypted per-key file (decrypted on demand, cached in process env), (4) DEFAULT or empty. Consumers call `$(load_secret GITHUB_TOKEN)` instead of `${GITHUB_TOKEN}` — identical behavior whether secrets come from Docker compose injection or Nomad Vault templates. | Every agent |
+| `lib/env.sh` | Loads `.env`, sets `FACTORY_ROOT`, exports project config (`FORGE_REPO`, `PROJECT_NAME`, etc.), defines `log()`, `forge_api()`, `forge_api_all()` (paginates all pages; accepts optional second TOKEN parameter, defaults to `$FORGE_TOKEN`; handles invalid/empty JSON responses gracefully — returns empty on parse error instead of crashing), `woodpecker_api()`, `wpdb()`, `memory_guard()` (skips agent if RAM < threshold). Auto-loads project TOML if `PROJECT_TOML` is set. Exports per-agent tokens (`FORGE_PLANNER_TOKEN`, `FORGE_GARDENER_TOKEN`, `FORGE_VAULT_TOKEN`, `FORGE_SUPERVISOR_TOKEN`, `FORGE_PREDICTOR_TOKEN`) — each falls back to `$FORGE_TOKEN` if not set. **Vault-only token guard (AD-006)**: `unset GITHUB_TOKEN CLAWHUB_TOKEN` so agents never hold external-action tokens — only the runner container receives them. **Container note**: when `DISINTO_CONTAINER=1`, `.env` is NOT re-sourced — compose already injects env vars (including `FORGE_URL=http://forgejo:3000`) and re-sourcing would clobber them. **Save/restore scope (#364)**: only `FORGE_URL` is preserved across `.env` re-sourcing (compose injects `http://forgejo:3000`, `.env` has `http://localhost:3000`). `FORGE_TOKEN` is NOT preserved so refreshed tokens in `.env` take effect immediately. **Required env var**: `FORGE_PASS` — bot password for git HTTP push (Forgejo 11.x rejects API tokens for `git push`, #361). **Hard preconditions (#674)**: `USER` and `HOME` must be exported by the entrypoint before sourcing. When `PROJECT_TOML` is set, `PROJECT_REPO_ROOT`, `PRIMARY_BRANCH`, and `OPS_REPO_ROOT` must also be set (by entrypoint or TOML). | Every agent |
 | `lib/ci-helpers.sh` | `ci_passed()` — returns 0 if CI state is "success" (or no CI configured). `ci_required_for_pr()` — returns 0 if PR has code files (CI required), 1 if non-code only (CI not required). `is_infra_step()` — returns 0 if a single CI step failure matches infra heuristics (clone/git exit 128, any exit 137, log timeout patterns). `classify_pipeline_failure()` — returns "infra \<reason>" if any failed Woodpecker step matches infra heuristics via `is_infra_step()`, else "code". `ensure_priority_label()` — looks up (or creates) the `priority` label and returns its ID; caches in `_PRIORITY_LABEL_ID`. `ci_commit_status <sha>` — queries Woodpecker directly for CI state, falls back to forge commit status API. `ci_pipeline_number <sha>` — returns the Woodpecker pipeline number for a commit, falls back to parsing forge status `target_url`. `ci_promote <repo_id> <pipeline_num> <environment>` — promotes a pipeline to a named Woodpecker environment (vault-gated deployment: vault approves, vault-fire calls this — vault redesign in progress, see #73-#77). `ci_get_logs <pipeline_number> [--step <name>]` — reads CI logs from Woodpecker SQLite database via `lib/ci-log-reader.py`; outputs last 200 lines to stdout. Requires mounted woodpecker-data volume at /woodpecker-data. | dev-poll, review-poll, review-pr |
 | `lib/ci-debug.sh` | CLI tool for Woodpecker CI: `list`, `status`, `logs`, `failures` subcommands. Not sourced — run directly. | Humans / dev-agent (tool access) |
 | `lib/ci-log-reader.py` | Python tool: reads CI logs from Woodpecker SQLite database. `<pipeline_number> [--step <name>]` — returns last 200 lines from failed steps (or specified step). Used by `ci_get_logs()` in ci-helpers.sh. Requires `WOODPECKER_DATA_DIR` (default: /woodpecker-data). | ci-helpers.sh |
@ -14,7 +14,7 @@ sourced as needed.
 | `lib/parse-deps.sh` | Extracts dependency issue numbers from an issue body (stdin → stdout, one number per line). Matches `## Dependencies` / `## Depends on` / `## Blocked by` sections and inline `depends on #N` / `blocked by #N` patterns. Inline scan skips fenced code blocks to prevent false positives from code examples in issue bodies. Not sourced — executed via `bash lib/parse-deps.sh`. | dev-poll |
 | `lib/formula-session.sh` | `acquire_run_lock()`, `load_formula()`, `load_formula_or_profile()`, `build_context_block()`, `ensure_ops_repo()`, `ops_commit_and_push()`, `build_prompt_footer()`, `build_sdk_prompt_footer()`, `formula_worktree_setup()`, `formula_prepare_profile_context()`, `formula_lessons_block()`, `profile_write_journal()`, `profile_load_lessons()`, `ensure_profile_repo()`, `_profile_has_repo()`, `_count_undigested_journals()`, `_profile_digest_journals()`, `_profile_restore_lessons()`, `_profile_commit_and_push()`, `resolve_agent_identity()`, `build_graph_section()`, `build_scratch_instruction()`, `read_scratch_context()`, `cleanup_stale_crashed_worktrees()` — shared helpers for formula-driven polling-loop agents (lock, .profile repo management, prompt assembly, worktree setup). Memory guard is provided by `memory_guard()` in `lib/env.sh` (not duplicated here). `resolve_agent_identity()` — sets `FORGE_TOKEN`, `AGENT_IDENTITY`, `FORGE_REMOTE` from per-agent token env vars and FORGE_URL remote detection. `build_graph_section()` generates the structural-analysis section (runs `lib/build-graph.py`, formats JSON output) — previously duplicated in planner-run.sh and predictor-run.sh, now shared here. `cleanup_stale_crashed_worktrees()` — thin wrapper around `worktree_cleanup_stale()` from `lib/worktree.sh` (kept for backwards compatibility). **Journal digestion guards (#702)**: `_profile_digest_journals()` respects `PROFILE_DIGEST_TIMEOUT` (default 300s) and `PROFILE_DIGEST_MAX_BATCH` (default 5 journals per run); `_profile_restore_lessons()` restores the previous lessons-learned.md on digest failure. | planner-run.sh, predictor-run.sh, gardener-run.sh, supervisor-run.sh, dev-agent.sh |
 | `lib/guard.sh` | `check_active(agent_name)` — reads `$FACTORY_ROOT/state/.{agent_name}-active`; exits 0 (skip) if the file is absent. Factory is off by default — state files must be created to enable each agent. **Logs a message to stderr** when skipping (`[check_active] SKIP: state file not found`), so agent dropout is visible in loop logs. Sourced by dev-poll.sh, review-poll.sh, predictor-run.sh, supervisor-run.sh. | polling-loop entry points |
-| `lib/mirrors.sh` | `mirror_push()` — pushes `$PRIMARY_BRANCH` + tags to all configured mirror remotes (fire-and-forget background pushes). Reads `MIRROR_NAMES` and `MIRROR_*` vars exported by `load-project.sh` from the `[mirrors]` TOML section. Failures are logged but never block the pipeline. `mirror_pull_register(clone_url, owner, repo_name, [interval])` — registers a Forgejo pull mirror via `POST /repos/migrate` with `mirror: true`. Creates the target repo and queues the first sync automatically. Works against empty Forgejo instances — no pre-existing content required. Used for Nomad migration cutover: point at Codeberg source, wait for sync, then proceed with `disinto init`. See [docs/mirror-bootstrap.md](../docs/mirror-bootstrap.md) for the full cutover path. Sourced by dev-poll.sh — called after every successful merge. | dev-poll.sh |
+| `lib/mirrors.sh` | `mirror_push()` — pushes `$PRIMARY_BRANCH` + tags to all configured mirror remotes (fire-and-forget background pushes). Reads `MIRROR_NAMES` and `MIRROR_*` vars exported by `load-project.sh` from the `[mirrors]` TOML section. Failures are logged but never block the pipeline. Sourced by dev-poll.sh — called after every successful merge. | dev-poll.sh |
 | `lib/build-graph.py` | Python tool: parses VISION.md, prerequisites.md (from ops repo), AGENTS.md, formulas/*.toml, evidence/ (from ops repo), and forge issues/labels into a NetworkX DiGraph. Runs structural analyses (orphaned objectives, stale prerequisites, thin evidence, circular deps) and outputs a JSON report. Used by `review-pr.sh` (per-PR changed-file analysis) and `predictor-run.sh` (full-project analysis) to provide structural context to Claude. | review-pr.sh, predictor-run.sh |
 | `lib/secret-scan.sh` | `scan_for_secrets()` — detects potential secrets (API keys, bearer tokens, private keys, URLs with embedded credentials) in text; returns 1 if secrets found. `redact_secrets()` — replaces detected secret patterns with `[REDACTED]`. | issue-lifecycle.sh |
 | `lib/stack-lock.sh` | File-based lock protocol for singleton project stack access. `stack_lock_acquire(holder, project)` — polls until free, breaks stale heartbeats (>10 min old), claims lock. `stack_lock_release(project)` — deletes lock file. `stack_lock_check(project)` — inspect current lock state. `stack_lock_heartbeat(project)` — update heartbeat timestamp (callers must call every 2 min while holding). Lock files at `~/data/locks/<project>-stack.lock`. | docker/edge/dispatcher.sh, reproduce formula |
@ -22,16 +22,14 @@ sourced as needed.
 | `lib/worktree.sh` | Reusable git worktree management: `worktree_create(path, branch, [base_ref])` — create worktree, checkout base, fetch submodules. `worktree_recover(path, branch, [remote])` — detect existing worktree, reuse if on correct branch (sets `_WORKTREE_REUSED`), otherwise clean and recreate. `worktree_cleanup(path)` — `git worktree remove --force`, clear Claude Code project cache (`~/.claude/projects/` matching path). `worktree_cleanup_stale([max_age_hours])` — scan `/tmp` for orphaned worktrees older than threshold, skip preserved and active tmux worktrees, prune. `worktree_preserve(path, reason)` — mark worktree as preserved for debugging (writes `.worktree-preserved` marker, skipped by stale cleanup). | dev-agent.sh, supervisor-run.sh, planner-run.sh, predictor-run.sh, gardener-run.sh |
 | `lib/pr-lifecycle.sh` | Reusable PR lifecycle library: `pr_create()`, `pr_find_by_branch()`, `pr_poll_ci()`, `pr_poll_review()`, `pr_merge()`, `pr_is_merged()`, `pr_walk_to_merge()`, `build_phase_protocol_prompt()`. Requires `lib/ci-helpers.sh`. | dev-agent.sh (future) |
 | `lib/issue-lifecycle.sh` | Reusable issue lifecycle library: `issue_claim()` (add in-progress, remove backlog), `issue_release()` (remove in-progress, add backlog), `issue_block()` (post diagnostic comment with secret redaction, add blocked label), `issue_close()`, `issue_check_deps()` (parse deps, check transitive closure; sets `_ISSUE_BLOCKED_BY`, `_ISSUE_SUGGESTION`), `issue_suggest_next()` (find next unblocked backlog issue; sets `_ISSUE_NEXT`), `issue_post_refusal()` (structured refusal comment with dedup). Label IDs cached in globals on first lookup. Sources `lib/secret-scan.sh`. | dev-agent.sh (future) |
-| `lib/action-vault.sh` | **Vault PR helper** — create vault action PRs on ops repo via Forgejo API (works from containers without SSH). `vault_request <action_id> <toml_content>` validates TOML (using `validate_vault_action` from `action-vault/vault-env.sh`), creates branch `vault/<action-id>`, writes `vault/actions/<action-id>.toml`, creates PR targeting `main` with title `vault: <action-id>` and body from context field, returns PR number. Idempotent: if PR exists, returns existing number. **Low-tier bypass**: if the action's `blast_radius` classifies as `low` (via `action-vault/classify.sh`), `vault_request` calls `_vault_commit_direct()` which commits directly to ops `main` using `FORGE_ADMIN_TOKEN` — no PR, no approval wait. Returns `0` (not a PR number) for direct commits. Requires `FORGE_TOKEN`, `FORGE_ADMIN_TOKEN` (low-tier only), `FORGE_URL`, `FORGE_REPO`, `FORGE_OPS_REPO`. Uses the calling agent's own token (saves/restores `FORGE_TOKEN` around sourcing `vault-env.sh`), so approval workflow respects individual agent identities. | dev-agent (vault actions), future vault dispatcher |
+| `lib/vault.sh` | **Vault PR helper** — create vault action PRs on ops repo via Forgejo API (works from containers without SSH). `vault_request <action_id> <toml_content>` validates TOML (using `validate_vault_action` from `vault/vault-env.sh`), creates branch `vault/<action-id>`, writes `vault/actions/<action-id>.toml`, creates PR targeting `main` with title `vault: <action-id>` and body from context field, returns PR number. Idempotent: if PR exists, returns existing number. **Low-tier bypass**: if the action's `blast_radius` classifies as `low` (via `vault/classify.sh`), `vault_request` calls `_vault_commit_direct()` which commits directly to ops `main` using `FORGE_ADMIN_TOKEN` — no PR, no approval wait. Returns `0` (not a PR number) for direct commits. Requires `FORGE_TOKEN`, `FORGE_ADMIN_TOKEN` (low-tier only), `FORGE_URL`, `FORGE_REPO`, `FORGE_OPS_REPO`. Uses the calling agent's own token (saves/restores `FORGE_TOKEN` around sourcing `vault-env.sh`), so approval workflow respects individual agent identities. | dev-agent (vault actions), future vault dispatcher |
 | `lib/branch-protection.sh` | Branch protection helpers for Forgejo repos. `setup_vault_branch_protection()` — configures admin-only merge protection on main (require 1 approval, restrict merge to admin role, block direct pushes). `setup_profile_branch_protection()` — same protection for `.profile` repos. `verify_branch_protection()` — checks protection is correctly configured. `remove_branch_protection()` — removes protection (cleanup/testing). Handles race condition after initial push: retries with backoff if Forgejo hasn't processed the branch yet. Requires `FORGE_TOKEN`, `FORGE_URL`, `FORGE_OPS_REPO`. | bin/disinto (hire-an-agent) |
 | `lib/agent-sdk.sh` | `agent_run([--resume SESSION_ID] [--worktree DIR] PROMPT)` — one-shot `claude -p` invocation with session persistence. Saves session ID to `SID_FILE`, reads it back on resume. `agent_recover_session()` — restore previous session ID from `SID_FILE` on startup. **Nudge guard**: skips nudge injection if the worktree is clean and no push is expected, preventing spurious re-invocations. Callers must define `SID_FILE`, `LOGFILE`, and `log()` before sourcing. **Concurrency**: external `flock` on `session.lock` is gated behind `CLAUDE_EXTERNAL_LOCK=1` (default off). When unset, each container's per-session `CLAUDE_CONFIG_DIR` isolation lets Claude Code's native lockfile handle OAuth refresh — no external serialization needed. Set `CLAUDE_EXTERNAL_LOCK=1` to re-enable the old flock wrapper as a rollback mechanism. See [`docs/CLAUDE-AUTH-CONCURRENCY.md`](../docs/CLAUDE-AUTH-CONCURRENCY.md) and AD-002 (#647). | formula-driven agents (dev-agent, planner-run, predictor-run, gardener-run) |
 | `lib/forge-setup.sh` | `setup_forge()` — Forgejo instance provisioning: creates admin user, bot accounts, org, repos (code + ops), configures webhooks, sets repo topics. Extracted from `bin/disinto`. Requires `FORGE_URL`, `FORGE_TOKEN`, `FACTORY_ROOT`. **Password storage (#361)**: after creating each bot account, stores its password in `.env` as `FORGE_<BOT>_PASS` (e.g. `FORGE_PASS`, `FORGE_REVIEW_PASS`, etc.) for use by `forge-push.sh`. | bin/disinto (init) |
 | `lib/forge-push.sh` | `push_to_forge()` — pushes a local clone to the Forgejo remote and verifies the push. `_assert_forge_push_globals()` validates required env vars before use. Requires `FORGE_URL`, `FORGE_PASS`, `FACTORY_ROOT`, `PRIMARY_BRANCH`. **Auth**: uses `FORGE_PASS` (bot password) for git HTTP push — Forgejo 11.x rejects API tokens for `git push` (#361). | bin/disinto (init) |
-| `lib/git-creds.sh` | Shared git credential helper configuration. `configure_git_creds([HOME_DIR] [RUN_AS_CMD])` — writes a static credential helper script and configures git globally to use password-based HTTP auth (Forgejo 11.x rejects API tokens for `git push`, #361). **Retry on cold boot (#741)**: resolves bot username from `FORGE_TOKEN` with 5 retries (exponential backoff 1-5s); fails loudly and returns 1 if Forgejo is unreachable — never falls back to a wrong hardcoded default (exports `BOT_USER` on success). `repair_baked_cred_urls([--as RUN_AS_CMD] DIR ...)` — rewrites any git remote URLs that have credentials baked in to use clean URLs instead; uses `safe.directory` bypass for root-owned repos (#671). Requires `FORGE_PASS`, `FORGE_URL`, `FORGE_TOKEN`. | entrypoints (agents, edge) |
+| `lib/git-creds.sh` | Shared git credential helper configuration. `configure_git_creds([HOME_DIR] [RUN_AS_CMD])` — writes a static credential helper script and configures git globally to use password-based HTTP auth (Forgejo 11.x rejects API tokens for `git push`, #361). `repair_baked_cred_urls([--as RUN_AS_CMD] DIR ...)` — rewrites any git remote URLs that have credentials baked in to use clean URLs instead; uses `safe.directory` bypass for root-owned repos (#671). Requires `FORGE_PASS`, `FORGE_URL`, `FORGE_TOKEN`. | entrypoints (agents, edge) |
 | `lib/ops-setup.sh` | `setup_ops_repo()` — creates ops repo on Forgejo if it doesn't exist, configures bot collaborators, clones/initializes ops repo locally, seeds directory structure (vault, knowledge, evidence, sprints). Evidence subdirectories seeded: engagement/, red-team/, holdout/, evolution/, user-test/. Also seeds sprints/ for architect output. Exports `_ACTUAL_OPS_SLUG`. `migrate_ops_repo(ops_root, [primary_branch])` — idempotent migration helper that seeds missing directories and .gitkeep files on existing ops repos (pre-#407 deployments). | bin/disinto (init) |
 | `lib/ci-setup.sh` | `_install_cron_impl()` — installs crontab entries for bare-metal deployments (compose mode uses polling loop instead). `_create_forgejo_oauth_app()` — generic helper to create an OAuth2 app on Forgejo (shared by Woodpecker and chat). `_create_woodpecker_oauth_impl()` — creates Woodpecker OAuth2 app (thin wrapper). `_create_chat_oauth_impl()` — creates disinto-chat OAuth2 app, writes `CHAT_OAUTH_CLIENT_ID`/`CHAT_OAUTH_CLIENT_SECRET` to `.env` (#708). `_generate_woodpecker_token_impl()` — auto-generates WOODPECKER_TOKEN via OAuth2 flow. `_activate_woodpecker_repo_impl()` — activates repo in Woodpecker. All gated by `_load_ci_context()` which validates required env vars. | bin/disinto (init) |
-| `lib/generators.sh` | Template generation for `disinto init`: `generate_compose()` — docker-compose.yml (uses `codeberg.org/forgejo/forgejo:11.0` tag; adds `security_opt: [apparmor:unconfined]` to all services for rootless container compatibility; Forgejo includes a healthcheck so dependent services use `condition: service_healthy` — fixes cold-start races, #665; adds `chat` service block with isolated `chat-config` named volume and `CHAT_HISTORY_DIR` bind-mount for per-user NDJSON history persistence (#710); injects `FORWARD_AUTH_SECRET` for Caddy↔chat defense-in-depth auth (#709); cost-cap env vars `CHAT_MAX_REQUESTS_PER_HOUR`, `CHAT_MAX_REQUESTS_PER_DAY`, `CHAT_MAX_TOKENS_PER_DAY` (#711); subdomain fallback comment for `EDGE_TUNNEL_FQDN_*` vars (#713); all `depends_on` now use `condition: service_healthy/started` instead of bare service names; all services now include `restart: unless-stopped` including the edge service — #768; agents service now uses `image: ghcr.io/disinto/agents:${DISINTO_IMAGE_TAG:-latest}` instead of `build:` (#429); `WOODPECKER_PLUGINS_PRIVILEGED` env var added to woodpecker service (#779); agents-llama conditional block gated on `ENABLE_LLAMA_AGENT=1` (#769); `agents-llama-all` compose service (profile `agents-llama-all`, all 7 roles: review,dev,gardener,architect,planner,predictor,supervisor) added by #801; agents service gains volume mounts for `./projects`, `./.env`, `./state`), `generate_caddyfile()` — Caddyfile (routes: `/forge/*` → forgejo:3000, `/woodpecker/*` → woodpecker:8000, `/staging/*` → staging:80; `/chat/login` and `/chat/oauth/callback` bypass `forward_auth` so unauthenticated users can reach the OAuth flow; `/chat/*` gated by `forward_auth` on `chat:8080/chat/auth/verify` which stamps `X-Forwarded-User` (#709); root `/` redirects to `/forge/`), `generate_staging_index()` — staging index, `generate_deploy_pipelines()` — Woodpecker deployment pipeline configs. Requires `FACTORY_ROOT`, `PROJECT_NAME`, `PRIMARY_BRANCH`. | bin/disinto (init) |
+| `lib/generators.sh` | Template generation for `disinto init`: `generate_compose()` — docker-compose.yml (uses `codeberg.org/forgejo/forgejo:11.0` tag; adds `security_opt: [apparmor:unconfined]` to all services for rootless container compatibility; Forgejo includes a healthcheck so dependent services use `condition: service_healthy` — fixes cold-start races, #665; adds `chat` service block with isolated `chat-config` named volume; all `depends_on` now use `condition: service_healthy/started` instead of bare service names), `generate_caddyfile()` — Caddyfile (routes: `/forge/*` → forgejo:3000, `/woodpecker/*` → woodpecker:8000, `/staging/*` → staging:80, `/chat/*` → chat:8080; root `/` redirects to `/forge/`), `generate_staging_index()` — staging index, `generate_deploy_pipelines()` — Woodpecker deployment pipeline configs. Requires `FACTORY_ROOT`, `PROJECT_NAME`, `PRIMARY_BRANCH`. | bin/disinto (init) |
 | `lib/sprint-filer.sh` | Post-merge sub-issue filer for sprint PRs. Invoked by the `.woodpecker/ops-filer.yml` pipeline after a sprint PR merges to ops repo `main`. Parses `<!-- filer:begin --> ... <!-- filer:end -->` blocks from sprint PR bodies to extract sub-issue definitions, creates them on the project repo using `FORGE_FILER_TOKEN` (narrow-scope `filer-bot` identity with `issues:write` only), adds `in-progress` label to the parent vision issue, and handles vision lifecycle closure when all sub-issues are closed. Uses `filer_api_all()` for paginated fetches. Idempotent: uses `<!-- decomposed-from: #<vision>, sprint: <slug>, id: <id> -->` markers to skip already-filed issues. Requires `FORGE_FILER_TOKEN`, `FORGE_API`, `FORGE_API_BASE`, `FORGE_OPS_REPO`. | `.woodpecker/ops-filer.yml` (CI pipeline on ops repo) |
 | `lib/hire-agent.sh` | `disinto_hire_an_agent()` — user creation, `.profile` repo setup, formula copying, branch protection, and state marker creation for hiring a new agent. Requires `FORGE_URL`, `FORGE_TOKEN`, `FACTORY_ROOT`, `PROJECT_NAME`. Extracted from `bin/disinto`. | bin/disinto (hire) |
 | `lib/release.sh` | `disinto_release()` — vault TOML creation, branch setup on ops repo, PR creation, and auto-merge request for a versioned release. `_assert_release_globals()` validates required env vars. Requires `FORGE_URL`, `FORGE_TOKEN`, `FORGE_OPS_REPO`, `FACTORY_ROOT`, `PRIMARY_BRANCH`. Extracted from `bin/disinto`. | bin/disinto (release) |
 | `lib/hvault.sh` | HashiCorp Vault helper module. `hvault_kv_get(PATH, [KEY])` — read KV v2 secret, optionally extract one key. `hvault_kv_put(PATH, KEY=VAL ...)` — write KV v2 secret. `hvault_kv_list(PATH)` — list keys at a KV path. `hvault_policy_apply(NAME, FILE)` — idempotent policy upsert. `hvault_jwt_login(ROLE, JWT)` — exchange JWT for short-lived token. `hvault_token_lookup()` — returns TTL/policies/accessor for current token. All functions use `VAULT_ADDR` + `VAULT_TOKEN` from env (fallback: `/etc/vault.d/root.token`), emit structured JSON errors to stderr on failure. Tests: `tests/lib-hvault.bats` (requires `vault server -dev`). | Not sourced at runtime yet — pure scaffolding for Nomad+Vault migration (#799) |
--- a/lib/claude-config.sh
+++ b/lib/claude-config.sh
@ -1,8 +1,10 @@
 #!/usr/bin/env bash
-# lib/claude-config.sh — Shared Claude config directory helpers (#641)
+# lib/claude-config.sh — Shared Claude config directory helpers (#641, #707)
 #
-# Provides setup_claude_config_dir() for creating/migrating CLAUDE_CONFIG_DIR
+# Provides:
-# and _env_set_idempotent() for writing env vars to .env files.
+#   setup_claude_dir <dir> [<auto_yes>] — Create/migrate a Claude config directory
 #   setup_claude_config_dir [auto_yes]   — Wrapper for default CLAUDE_CONFIG_DIR
 #   _env_set_idempotent()                 — Write env vars to .env files
 #
 # Requires: CLAUDE_CONFIG_DIR, CLAUDE_SHARED_DIR (set by lib/env.sh)
@ -21,24 +23,33 @@ _env_set_idempotent() {
  fi
 }
-# Create the shared CLAUDE_CONFIG_DIR, optionally migrating ~/.claude.
+# Create a Claude config directory, optionally migrating ~/.claude.
-# Usage: setup_claude_config_dir [auto_yes]
+# This is the parameterized helper that handles any CLAUDE_CONFIG_DIR path.
-setup_claude_config_dir() {
+# Usage: setup_claude_dir <config_dir> [auto_yes]
-  local auto_yes="${1:-false}"
+#        setup_claude_dir /path/to/config [true]
 #
 # Parameters:
 #   $1 - Path to the Claude config directory to create
 #   $2 - Auto-confirm mode (true/false), defaults to false
 #
 # Returns: 0 on success, 1 on failure
 setup_claude_dir() {
  local config_dir="$1"
  local auto_yes="${2:-false}"
  local home_claude="${HOME}/.claude"
-  # Create the shared config directory (idempotent)
+  # Create the config directory (idempotent)
-  install -d -m 0700 -o "$USER" "$CLAUDE_CONFIG_DIR"
+  install -d -m 0700 -o "$USER" "$config_dir"
-  echo "Claude:  ${CLAUDE_CONFIG_DIR} (ready)"
+  echo "Claude:  ${config_dir} (ready)"
-  # If ~/.claude is already a symlink to CLAUDE_CONFIG_DIR, nothing to do
+  # If ~/.claude is already a symlink to config_dir, nothing to do
  if [ -L "$home_claude" ]; then
    local link_target
    link_target=$(readlink -f "$home_claude")
    local config_real
-    config_real=$(readlink -f "$CLAUDE_CONFIG_DIR")
+    config_real=$(readlink -f "$config_dir")
    if [ "$link_target" = "$config_real" ]; then
-      echo "Claude:  ${home_claude} -> ${CLAUDE_CONFIG_DIR} (symlink OK)"
+      echo "Claude:  ${home_claude} -> ${config_dir} (symlink OK)"
      return 0
    fi
  fi
@ -54,25 +65,25 @@ setup_claude_config_dir() {
    fi
  fi
-  # Check CLAUDE_CONFIG_DIR contents
+  # Check config_dir contents
-  if [ -n "$(ls -A "$CLAUDE_CONFIG_DIR" 2>/dev/null)" ]; then
+  if [ -n "$(ls -A "$config_dir" 2>/dev/null)" ]; then
    config_nonempty=true
  fi
  # Case: both non-empty — abort, operator must reconcile
  if [ "$home_nonempty" = true ] && [ "$config_nonempty" = true ]; then
-    echo "ERROR: both ${home_claude} and ${CLAUDE_CONFIG_DIR} exist and are non-empty" >&2
+    echo "ERROR: both ${home_claude} and ${config_dir} exist and are non-empty" >&2
    echo "  Reconcile manually: merge or remove one, then re-run disinto init" >&2
    return 1
  fi
-  # Case: ~/.claude exists and CLAUDE_CONFIG_DIR is empty — offer migration
+  # Case: ~/.claude exists and config_dir is empty — offer migration
  if [ "$home_nonempty" = true ] && [ "$config_nonempty" = false ]; then
    local do_migrate=false
    if [ "$auto_yes" = true ]; then
      do_migrate=true
    elif [ -t 0 ]; then
-      read -rp "Migrate ${home_claude} to ${CLAUDE_CONFIG_DIR}? [Y/n] " confirm
+      read -rp "Migrate ${home_claude} to ${config_dir}? [Y/n] " confirm
      if [[ ! "$confirm" =~ ^[Nn] ]]; then
        do_migrate=true
      fi
@ -83,11 +94,11 @@ setup_claude_config_dir() {
    fi
    if [ "$do_migrate" = true ]; then
-      # Move contents (not the dir itself) to preserve CLAUDE_CONFIG_DIR ownership
+      # Move contents (not the dir itself) to preserve config_dir ownership
-      cp -a "$home_claude/." "$CLAUDE_CONFIG_DIR/"
+      cp -a "$home_claude/." "$config_dir/"
      rm -rf "$home_claude"
-      ln -sfn "$CLAUDE_CONFIG_DIR" "$home_claude"
+      ln -sfn "$config_dir" "$home_claude"
-      echo "Claude:  migrated ${home_claude} -> ${CLAUDE_CONFIG_DIR}"
+      echo "Claude:  migrated ${home_claude} -> ${config_dir}"
      return 0
    fi
  fi
@ -97,7 +108,15 @@ setup_claude_config_dir() {
    rmdir "$home_claude" 2>/dev/null || true
  fi
  if [ ! -e "$home_claude" ]; then
-    ln -sfn "$CLAUDE_CONFIG_DIR" "$home_claude"
+    ln -sfn "$config_dir" "$home_claude"
-    echo "Claude:  ${home_claude} -> ${CLAUDE_CONFIG_DIR} (symlink created)"
+    echo "Claude:  ${home_claude} -> ${config_dir} (symlink created)"
  fi
 }
 # Create the shared CLAUDE_CONFIG_DIR, optionally migrating ~/.claude.
 # Wrapper around setup_claude_dir for the default config directory.
 # Usage: setup_claude_config_dir [auto_yes]
 setup_claude_config_dir() {
  local auto_yes="${1:-false}"
  setup_claude_dir "$CLAUDE_CONFIG_DIR" "$auto_yes"
 }
--- a/lib/env.sh
+++ b/lib/env.sh
@ -121,10 +121,9 @@ export FORGE_VAULT_TOKEN="${FORGE_VAULT_TOKEN:-${FORGE_TOKEN}}"
 export FORGE_SUPERVISOR_TOKEN="${FORGE_SUPERVISOR_TOKEN:-${FORGE_TOKEN}}"
 export FORGE_PREDICTOR_TOKEN="${FORGE_PREDICTOR_TOKEN:-${FORGE_TOKEN}}"
 export FORGE_ARCHITECT_TOKEN="${FORGE_ARCHITECT_TOKEN:-${FORGE_TOKEN}}"
 export FORGE_FILER_TOKEN="${FORGE_FILER_TOKEN:-${FORGE_TOKEN}}"
 # Bot usernames filter
-export FORGE_BOT_USERNAMES="${FORGE_BOT_USERNAMES:-dev-bot,review-bot,planner-bot,gardener-bot,vault-bot,supervisor-bot,predictor-bot,architect-bot,filer-bot}"
+export FORGE_BOT_USERNAMES="${FORGE_BOT_USERNAMES:-dev-bot,review-bot,planner-bot,gardener-bot,vault-bot,supervisor-bot,predictor-bot,architect-bot}"
 # Project config
 export FORGE_REPO="${FORGE_REPO:-}"
@ -158,8 +157,8 @@ export WOODPECKER_SERVER="${WOODPECKER_SERVER:-http://localhost:8000}"
 export CLAUDE_TIMEOUT="${CLAUDE_TIMEOUT:-7200}"
 # Vault-only token guard (#745): external-action tokens (GITHUB_TOKEN, CLAWHUB_TOKEN)
-# must NEVER be available to agents. They live in secrets/*.enc and are decrypted
+# must NEVER be available to agents. They live in .env.vault.enc and are injected
-# only into the ephemeral runner container at fire time (#777). Unset them here so
+# only into the ephemeral runner container at fire time. Unset them here so
 # even an accidental .env inclusion cannot leak them into agent sessions.
 unset GITHUB_TOKEN 2>/dev/null || true
 unset CLAWHUB_TOKEN 2>/dev/null || true
@ -313,68 +312,6 @@ memory_guard() {
  fi
 }
 # =============================================================================
 # SECRET LOADING ABSTRACTION
 # =============================================================================
 # load_secret NAME [DEFAULT]
 #
 # Resolves a secret value using the following precedence:
 #   1. /secrets/<NAME>.env  — Nomad-rendered template (future)
 #   2. Current environment  — already set by .env.enc, compose, etc.
 #   3. secrets/<NAME>.enc   — age-encrypted per-key file (decrypted on demand)
 #   4. DEFAULT (or empty)
 #
 # Prints the resolved value to stdout.  Caches age-decrypted values in the
 # process environment so subsequent calls are free.
 # =============================================================================
 load_secret() {
  local name="$1"
  local default="${2:-}"
  # 1. Nomad-rendered template (future: Nomad writes /secrets/<NAME>.env)
  local nomad_path="/secrets/${name}.env"
  if [ -f "$nomad_path" ]; then
    # Source into a subshell to extract just the value
    local _nomad_val
    _nomad_val=$(
      set -a
      # shellcheck source=/dev/null
      source "$nomad_path"
      set +a
      printf '%s' "${!name:-}"
    )
    if [ -n "$_nomad_val" ]; then
      export "$name=$_nomad_val"
      printf '%s' "$_nomad_val"
      return 0
    fi
  fi
  # 2. Already in environment (set by .env.enc, compose injection, etc.)
  if [ -n "${!name:-}" ]; then
    printf '%s' "${!name}"
    return 0
  fi
  # 3. Age-encrypted per-key file: secrets/<NAME>.enc (#777)
  local _age_key="${HOME}/.config/sops/age/keys.txt"
  local _enc_path="${FACTORY_ROOT}/secrets/${name}.enc"
  if [ -f "$_enc_path" ] && [ -f "$_age_key" ] && command -v age &>/dev/null; then
    local _dec_val
    if _dec_val=$(age -d -i "$_age_key" "$_enc_path" 2>/dev/null) && [ -n "$_dec_val" ]; then
      export "$name=$_dec_val"
      printf '%s' "$_dec_val"
      return 0
    fi
  fi
  # 4. Default (or empty)
  if [ -n "$default" ]; then
    printf '%s' "$default"
  fi
  return 0
 }
 # Source tea helpers (available when tea binary is installed)
 if command -v tea &>/dev/null; then
  # shellcheck source=tea-helpers.sh
--- a/lib/forge-setup.sh
+++ b/lib/forge-setup.sh
@ -31,9 +31,8 @@ _load_init_context() {
 # Execute a command in the Forgejo container (for admin operations)
 _forgejo_exec() {
  local use_bare="${DISINTO_BARE:-false}"
  local cname="${FORGEJO_CONTAINER_NAME:-disinto-forgejo}"
  if [ "$use_bare" = true ]; then
-    docker exec -u git "$cname" "$@"
+    docker exec -u git disinto-forgejo "$@"
  else
    docker compose -f "${FACTORY_ROOT}/docker-compose.yml" exec -T -u git forgejo "$@"
  fi
@ -95,12 +94,11 @@ setup_forge() {
      # Bare-metal mode: standalone docker run
      mkdir -p "${FORGEJO_DATA_DIR}"
-      local cname="${FORGEJO_CONTAINER_NAME:-disinto-forgejo}"
+      if docker ps -a --format '{{.Names}}' | grep -q '^disinto-forgejo$'; then
-      if docker ps -a --format '{{.Names}}' | grep -q "^${cname}$"; then
+        docker start disinto-forgejo >/dev/null 2>&1 || true
        docker start "$cname" >/dev/null 2>&1 || true
      else
        docker run -d \
-          --name "$cname" \
+          --name disinto-forgejo \
          --restart unless-stopped \
          -p "${forge_port}:3000" \
          -p 2222:22 \
@ -212,8 +210,8 @@ setup_forge() {
  # Create human user (disinto-admin) as site admin if it doesn't exist
  local human_user="disinto-admin"
-  # human_user == admin_user; reuse admin_pass for basic-auth operations
+  local human_pass
-  local human_pass="$admin_pass"
+  human_pass="admin-$(head -c 16 /dev/urandom | base64 | tr -dc 'a-zA-Z0-9' | head -c 20)"
  if ! curl -sf --max-time 5 -H "Authorization: token ${FORGE_TOKEN:-}" "${forge_url}/api/v1/users/${human_user}" >/dev/null 2>&1; then
    echo "Creating human user: ${human_user}"
@ -245,89 +243,63 @@ setup_forge() {
    echo "Human user: ${human_user} (already exists)"
  fi
-  # Preserve admin token if already stored in .env (idempotent re-run)
+  # Delete existing admin token if present (token sha1 is only returned at creation time)
-  local admin_token=""
+  local existing_token_id
-  if _token_exists_in_env "FORGE_ADMIN_TOKEN" "$env_file" && [ "$rotate_tokens" = false ]; then
+  existing_token_id=$(curl -sf \
-    admin_token=$(grep '^FORGE_ADMIN_TOKEN=' "$env_file" | head -1 | cut -d= -f2-)
+    -u "${admin_user}:${admin_pass}" \
-    [ -n "$admin_token" ] && echo "Admin token: preserved (use --rotate-tokens to force)"
+    "${forge_url}/api/v1/users/${admin_user}/tokens" 2>/dev/null \
    | jq -r '.[] | select(.name == "disinto-admin-token") | .id') || existing_token_id=""
  if [ -n "$existing_token_id" ]; then
    curl -sf -X DELETE \
      -u "${admin_user}:${admin_pass}" \
      "${forge_url}/api/v1/users/${admin_user}/tokens/${existing_token_id}" >/dev/null 2>&1 || true
  fi
  # Create admin token (fresh, so sha1 is returned)
  local admin_token
  admin_token=$(curl -sf -X POST \
    -u "${admin_user}:${admin_pass}" \
    -H "Content-Type: application/json" \
    "${forge_url}/api/v1/users/${admin_user}/tokens" \
    -d '{"name":"disinto-admin-token","scopes":["all"]}' 2>/dev/null \
    | jq -r '.sha1 // empty') || admin_token=""
  if [ -z "$admin_token" ]; then
-    # Delete existing admin token if present (token sha1 is only returned at creation time)
+    echo "Error: failed to obtain admin API token" >&2
-    local existing_token_id
+    exit 1
    existing_token_id=$(curl -sf \
      -u "${admin_user}:${admin_pass}" \
      "${forge_url}/api/v1/users/${admin_user}/tokens" 2>/dev/null \
      | jq -r '.[] | select(.name == "disinto-admin-token") | .id') || existing_token_id=""
    if [ -n "$existing_token_id" ]; then
      curl -sf -X DELETE \
        -u "${admin_user}:${admin_pass}" \
        "${forge_url}/api/v1/users/${admin_user}/tokens/${existing_token_id}" >/dev/null 2>&1 || true
    fi
    # Create admin token (fresh, so sha1 is returned)
    admin_token=$(curl -sf -X POST \
      -u "${admin_user}:${admin_pass}" \
      -H "Content-Type: application/json" \
      "${forge_url}/api/v1/users/${admin_user}/tokens" \
      -d '{"name":"disinto-admin-token","scopes":["all"]}' 2>/dev/null \
      | jq -r '.sha1 // empty') || admin_token=""
    if [ -z "$admin_token" ]; then
      echo "Error: failed to obtain admin API token" >&2
      exit 1
    fi
    # Store admin token for idempotent re-runs
    if grep -q '^FORGE_ADMIN_TOKEN=' "$env_file" 2>/dev/null; then
      sed -i "s|^FORGE_ADMIN_TOKEN=.*|FORGE_ADMIN_TOKEN=${admin_token}|" "$env_file"
    else
      printf 'FORGE_ADMIN_TOKEN=%s\n' "$admin_token" >> "$env_file"
    fi
    echo "Admin token: generated and saved (FORGE_ADMIN_TOKEN)"
  fi
-  # Get or create human user token (human_user == admin_user; use admin_pass)
+  # Get or create human user token
  local human_token=""
-  if _token_exists_in_env "HUMAN_TOKEN" "$env_file" && [ "$rotate_tokens" = false ]; then
+  # Delete existing human token if present (token sha1 is only returned at creation time)
-    human_token=$(grep '^HUMAN_TOKEN=' "$env_file" | head -1 | cut -d= -f2-)
+  local existing_human_token_id
-    if [ -n "$human_token" ]; then
+  existing_human_token_id=$(curl -sf \
-      export HUMAN_TOKEN="$human_token"
+    -u "${human_user}:${human_pass}" \
-      echo "  Human token preserved (use --rotate-tokens to force)"
+    "${forge_url}/api/v1/users/${human_user}/tokens" 2>/dev/null \
-    fi
+    | jq -r '.[] | select(.name == "disinto-human-token") | .id') || existing_human_token_id=""
  if [ -n "$existing_human_token_id" ]; then
    curl -sf -X DELETE \
      -u "${human_user}:${human_pass}" \
      "${forge_url}/api/v1/users/${human_user}/tokens/${existing_human_token_id}" >/dev/null 2>&1 || true
  fi
-  if [ -z "$human_token" ]; then
+  # Create human token (fresh, so sha1 is returned)
-    # Delete existing human token if present (token sha1 is only returned at creation time)
+  human_token=$(curl -sf -X POST \
-    local existing_human_token_id
+    -u "${human_user}:${human_pass}" \
-    existing_human_token_id=$(curl -sf \
+    -H "Content-Type: application/json" \
-      -u "${admin_user}:${admin_pass}" \
+    "${forge_url}/api/v1/users/${human_user}/tokens" \
-      "${forge_url}/api/v1/users/${human_user}/tokens" 2>/dev/null \
+    -d '{"name":"disinto-human-token","scopes":["all"]}' 2>/dev/null \
-      | jq -r '.[] | select(.name == "disinto-human-token") | .id') || existing_human_token_id=""
+    | jq -r '.sha1 // empty') || human_token=""
    if [ -n "$existing_human_token_id" ]; then
      curl -sf -X DELETE \
        -u "${admin_user}:${admin_pass}" \
        "${forge_url}/api/v1/users/${human_user}/tokens/${existing_human_token_id}" >/dev/null 2>&1 || true
    fi
-    # Create human token (use admin_pass since human_user == admin_user)
+  if [ -n "$human_token" ]; then
-    human_token=$(curl -sf -X POST \
+    # Store human token in .env
-      -u "${admin_user}:${admin_pass}" \
+    if grep -q '^HUMAN_TOKEN=' "$env_file" 2>/dev/null; then
-      -H "Content-Type: application/json" \
+      sed -i "s|^HUMAN_TOKEN=.*|HUMAN_TOKEN=${human_token}|" "$env_file"
-      "${forge_url}/api/v1/users/${human_user}/tokens" \
+    else
-      -d '{"name":"disinto-human-token","scopes":["all"]}' 2>/dev/null \
+      printf 'HUMAN_TOKEN=%s\n' "$human_token" >> "$env_file"
      | jq -r '.sha1 // empty') || human_token=""
    if [ -n "$human_token" ]; then
      # Store human token in .env
      if grep -q '^HUMAN_TOKEN=' "$env_file" 2>/dev/null; then
        sed -i "s|^HUMAN_TOKEN=.*|HUMAN_TOKEN=${human_token}|" "$env_file"
      else
        printf 'HUMAN_TOKEN=%s\n' "$human_token" >> "$env_file"
      fi
      export HUMAN_TOKEN="$human_token"
      echo "  Human token generated and saved (HUMAN_TOKEN)"
    fi
    export HUMAN_TOKEN="$human_token"
    echo "  Human token saved (HUMAN_TOKEN)"
  fi
  # Create bot users and tokens
@ -747,7 +719,7 @@ setup_forge() {
    fi
    # Add all bot users as collaborators with appropriate permissions
-    # dev-bot: write (PR creation via lib/action-vault.sh)
+    # dev-bot: write (PR creation via lib/vault.sh)
    # review-bot: read (PR review)
    # planner-bot: write (prerequisites.md, memory)
    # gardener-bot: write (backlog grooming)
--- a/lib/formula-session.sh
+++ b/lib/formula-session.sh
@ -819,7 +819,8 @@ build_prompt_footer() {
 Base URL: ${FORGE_API}
 Auth header: -H \"Authorization: token \${FORGE_TOKEN}\"
  Read issue:  curl -sf -H \"Authorization: token \${FORGE_TOKEN}\" '${FORGE_API}/issues/{number}' | jq '.body'
-  List labels: curl -sf -H \"Authorization: token \${FORGE_TOKEN}\" '${FORGE_API}/labels'${extra_api}
+  Create issue: curl -sf -X POST -H \"Authorization: token \${FORGE_TOKEN}\" -H 'Content-Type: application/json' '${FORGE_API}/issues' -d '{\"title\":\"...\",\"body\":\"...\",\"labels\":[LABEL_ID]}'${extra_api}
  List labels: curl -sf -H \"Authorization: token \${FORGE_TOKEN}\" '${FORGE_API}/labels'
 NEVER echo or include the actual token value in output — always reference \${FORGE_TOKEN}.
 ## Environment
--- a/lib/generators.sh
+++ b/lib/generators.sh
@ -100,7 +100,9 @@ _generate_local_model_services() {
            cat >> "$temp_file" <<EOF
  agents-${service_name}:
-    image: ghcr.io/disinto/agents:\${DISINTO_IMAGE_TAG:-latest}
+    build:
      context: .
      dockerfile: docker/agents/Dockerfile
    container_name: disinto-agents-${service_name}
    restart: unless-stopped
    security_opt:
@ -109,9 +111,9 @@ _generate_local_model_services() {
      - agents-${service_name}-data:/home/agent/data
      - project-repos:/home/agent/repos
      - \${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:\${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
-      - \${CLAUDE_CONFIG_FILE:-\${HOME}/.claude.json}:/home/agent/.claude.json:ro
+      - \${HOME}/.claude.json:/home/agent/.claude.json:ro
-      - \${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
+      - CLAUDE_BIN_PLACEHOLDER:/usr/local/bin/claude:ro
-      - \${AGENT_SSH_DIR:-\${HOME}/.ssh}:/home/agent/.ssh:ro
+      - \${HOME}/.ssh:/home/agent/.ssh:ro
    environment:
      FORGE_URL: http://forgejo:3000
      FORGE_REPO: ${FORGE_REPO:-disinto-admin/disinto}
@ -140,7 +142,6 @@ _generate_local_model_services() {
      GARDENER_INTERVAL: "${GARDENER_INTERVAL:-21600}"
      ARCHITECT_INTERVAL: "${ARCHITECT_INTERVAL:-21600}"
      PLANNER_INTERVAL: "${PLANNER_INTERVAL:-43200}"
      SUPERVISOR_INTERVAL: "${SUPERVISOR_INTERVAL:-1200}"
    depends_on:
      forgejo:
        condition: service_healthy
@ -232,7 +233,6 @@ for name, config in agents.items():
 # to materialize a working stack on a fresh checkout.
 _generate_compose_impl() {
  local forge_port="${1:-3000}"
  local use_build="${2:-false}"
  local compose_file="${FACTORY_ROOT}/docker-compose.yml"
  # Check if compose file already exists
@ -296,7 +296,6 @@ services:
      WOODPECKER_AGENT_SECRET: ${WOODPECKER_AGENT_SECRET:-}
      WOODPECKER_DATABASE_DRIVER: sqlite3
      WOODPECKER_DATABASE_DATASOURCE: /var/lib/woodpecker/woodpecker.sqlite
      WOODPECKER_PLUGINS_PRIVILEGED: ${WOODPECKER_PLUGINS_PRIVILEGED:-plugins/docker}
      WOODPECKER_ENVIRONMENT: "FORGE_TOKEN:${FORGE_TOKEN}"
    depends_on:
      forgejo:
@ -319,19 +318,15 @@ services:
      WOODPECKER_AGENT_SECRET: ${WOODPECKER_AGENT_SECRET:-}
      WOODPECKER_GRPC_SECURE: "false"
      WOODPECKER_HEALTHCHECK_ADDR: ":3333"
-      WOODPECKER_BACKEND_DOCKER_NETWORK: ${WOODPECKER_CI_NETWORK:-disinto_disinto-net}
+      WOODPECKER_BACKEND_DOCKER_NETWORK: disinto_disinto-net
      WOODPECKER_MAX_WORKFLOWS: 1
    healthcheck:
      test: ["CMD", "wget", "-q", "--spider", "http://localhost:3333/healthz"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 15s
    depends_on:
      - woodpecker
  agents:
-    image: ghcr.io/disinto/agents:${DISINTO_IMAGE_TAG:-latest}
+    build:
      context: .
      dockerfile: docker/agents/Dockerfile
    container_name: disinto-agents
    restart: unless-stopped
    security_opt:
@ -340,14 +335,11 @@ services:
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
-      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
+      - ${HOME}/.claude.json:/home/agent/.claude.json:ro
-      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
+      - CLAUDE_BIN_PLACEHOLDER:/usr/local/bin/claude:ro
-      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
+      - ${HOME}/.ssh:/home/agent/.ssh:ro
-      - ${SOPS_AGE_DIR:-${HOME}/.config/sops/age}:/home/agent/.config/sops/age:ro
+      - ${HOME}/.config/sops/age:/home/agent/.config/sops/age:ro
      - woodpecker-data:/woodpecker-data:ro
      - ./projects:/home/agent/disinto/projects:ro
      - ./.env:/home/agent/disinto/.env:ro
      - ./state:/home/agent/disinto/state
    environment:
      FORGE_URL: http://forgejo:3000
      FORGE_REPO: ${FORGE_REPO:-disinto-admin/disinto}
@ -379,14 +371,8 @@ services:
      PLANNER_INTERVAL: ${PLANNER_INTERVAL:-43200}
    # IMPORTANT: agents get explicit environment variables (forge tokens, CI tokens, config).
    # Vault-only secrets (GITHUB_TOKEN, CLAWHUB_TOKEN, deploy keys) live in
-    # secrets/*.enc and are NEVER injected here — only the runner
+    # .env.vault.enc and are NEVER injected here — only the runner
-    # container receives them at fire time (AD-006, #745, #777).
+    # container receives them at fire time (AD-006, #745).
    healthcheck:
      test: ["CMD", "pgrep", "-f", "entrypoint.sh"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 30s
    depends_on:
      forgejo:
        condition: service_healthy
@ -395,137 +381,10 @@ services:
    networks:
      - disinto-net
 COMPOSEEOF
  # ── Conditional agents-llama block (ENABLE_LLAMA_AGENT=1) ──────────────
  # Local-Qwen dev agent — gated on ENABLE_LLAMA_AGENT so factories without
  # a local llama endpoint don't try to start it.  See docs/agents-llama.md.
  if [ "${ENABLE_LLAMA_AGENT:-0}" = "1" ]; then
    cat >> "$compose_file" <<'LLAMAEOF'
  agents-llama:
    build:
      context: .
      dockerfile: docker/agents/Dockerfile
    container_name: disinto-agents-llama
    restart: unless-stopped
    security_opt:
      - apparmor=unconfined
    volumes:
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
      - ${SOPS_AGE_DIR:-${HOME}/.config/sops/age}:/home/agent/.config/sops/age:ro
      - woodpecker-data:/woodpecker-data:ro
    environment:
      FORGE_URL: http://forgejo:3000
      FORGE_REPO: ${FORGE_REPO:-disinto-admin/disinto}
      FORGE_TOKEN: ${FORGE_TOKEN_LLAMA:-}
      FORGE_PASS: ${FORGE_PASS_LLAMA:-}
      FORGE_BOT_USERNAMES: ${FORGE_BOT_USERNAMES:-}
      WOODPECKER_TOKEN: ${WOODPECKER_TOKEN:-}
      CLAUDE_TIMEOUT: ${CLAUDE_TIMEOUT:-7200}
      CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC: ${CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC:-1}
      CLAUDE_AUTOCOMPACT_PCT_OVERRIDE: "60"
      ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY:-}
      ANTHROPIC_BASE_URL: ${ANTHROPIC_BASE_URL:-}
      FORGE_ADMIN_PASS: ${FORGE_ADMIN_PASS:-}
      DISINTO_CONTAINER: "1"
      PROJECT_NAME: ${PROJECT_NAME:-project}
      PROJECT_REPO_ROOT: /home/agent/repos/${PROJECT_NAME:-project}
      WOODPECKER_DATA_DIR: /woodpecker-data
      WOODPECKER_REPO_ID: "PLACEHOLDER_WP_REPO_ID"
      CLAUDE_CONFIG_DIR: ${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}
      POLL_INTERVAL: ${POLL_INTERVAL:-300}
      AGENT_ROLES: dev
    healthcheck:
      test: ["CMD", "pgrep", "-f", "entrypoint.sh"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 30s
    depends_on:
      forgejo:
        condition: service_healthy
    networks:
      - disinto-net
  agents-llama-all:
    build:
      context: .
      dockerfile: docker/agents/Dockerfile
    container_name: disinto-agents-llama-all
    restart: unless-stopped
    profiles: ["agents-llama-all"]
    security_opt:
      - apparmor=unconfined
    volumes:
      - agent-data:/home/agent/data
      - project-repos:/home/agent/repos
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
      - ${AGENT_SSH_DIR:-${HOME}/.ssh}:/home/agent/.ssh:ro
      - ${SOPS_AGE_DIR:-${HOME}/.config/sops/age}:/home/agent/.config/sops/age:ro
      - woodpecker-data:/woodpecker-data:ro
    environment:
      FORGE_URL: http://forgejo:3000
      FORGE_REPO: ${FORGE_REPO:-disinto-admin/disinto}
      FORGE_TOKEN: ${FORGE_TOKEN_LLAMA:-}
      FORGE_PASS: ${FORGE_PASS_LLAMA:-}
      FORGE_REVIEW_TOKEN: ${FORGE_REVIEW_TOKEN:-}
      FORGE_PLANNER_TOKEN: ${FORGE_PLANNER_TOKEN:-}
      FORGE_GARDENER_TOKEN: ${FORGE_GARDENER_TOKEN:-}
      FORGE_VAULT_TOKEN: ${FORGE_VAULT_TOKEN:-}
      FORGE_SUPERVISOR_TOKEN: ${FORGE_SUPERVISOR_TOKEN:-}
      FORGE_PREDICTOR_TOKEN: ${FORGE_PREDICTOR_TOKEN:-}
      FORGE_ARCHITECT_TOKEN: ${FORGE_ARCHITECT_TOKEN:-}
      FORGE_FILER_TOKEN: ${FORGE_FILER_TOKEN:-}
      FORGE_BOT_USERNAMES: ${FORGE_BOT_USERNAMES:-}
      WOODPECKER_TOKEN: ${WOODPECKER_TOKEN:-}
      CLAUDE_TIMEOUT: ${CLAUDE_TIMEOUT:-7200}
      CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC: ${CLAUDE_CODE_DISABLE_NONESSENTIAL_TRAFFIC:-1}
      CLAUDE_AUTOCOMPACT_PCT_OVERRIDE: "60"
      CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS: "1"
      ANTHROPIC_API_KEY: ${ANTHROPIC_API_KEY:-}
      ANTHROPIC_BASE_URL: ${ANTHROPIC_BASE_URL:-}
      FORGE_ADMIN_PASS: ${FORGE_ADMIN_PASS:-}
      DISINTO_CONTAINER: "1"
      PROJECT_NAME: ${PROJECT_NAME:-project}
      PROJECT_REPO_ROOT: /home/agent/repos/${PROJECT_NAME:-project}
      WOODPECKER_DATA_DIR: /woodpecker-data
      WOODPECKER_REPO_ID: "PLACEHOLDER_WP_REPO_ID"
      CLAUDE_CONFIG_DIR: ${CLAUDE_CONFIG_DIR:-/var/lib/disinto/claude-shared/config}
      POLL_INTERVAL: ${POLL_INTERVAL:-300}
      GARDENER_INTERVAL: ${GARDENER_INTERVAL:-21600}
      ARCHITECT_INTERVAL: ${ARCHITECT_INTERVAL:-21600}
      PLANNER_INTERVAL: ${PLANNER_INTERVAL:-43200}
      SUPERVISOR_INTERVAL: ${SUPERVISOR_INTERVAL:-1200}
      AGENT_ROLES: review,dev,gardener,architect,planner,predictor,supervisor
    healthcheck:
      test: ["CMD", "pgrep", "-f", "entrypoint.sh"]
      interval: 60s
      timeout: 5s
      retries: 3
      start_period: 30s
    depends_on:
      forgejo:
        condition: service_healthy
      woodpecker:
        condition: service_started
    networks:
      - disinto-net
 LLAMAEOF
  fi
  # Resume the rest of the compose file (runner onward)
  cat >> "$compose_file" <<'COMPOSEEOF'
  runner:
-    image: ghcr.io/disinto/agents:${DISINTO_IMAGE_TAG:-latest}
+    build:
      context: .
      dockerfile: docker/agents/Dockerfile
    profiles: ["vault"]
    security_opt:
      - apparmor=unconfined
@ -546,9 +405,8 @@ LLAMAEOF
  # Edge proxy — reverse proxy to Forgejo, Woodpecker, and staging
  # Serves on ports 80/443, routes based on path
  edge:
-    image: ghcr.io/disinto/edge:${DISINTO_IMAGE_TAG:-latest}
+    build: ./docker/edge
    container_name: disinto-edge
    restart: unless-stopped
    security_opt:
      - apparmor=unconfined
    ports:
@ -572,24 +430,13 @@ LLAMAEOF
      - EDGE_TUNNEL_USER=${EDGE_TUNNEL_USER:-tunnel}
      - EDGE_TUNNEL_PORT=${EDGE_TUNNEL_PORT:-}
      - EDGE_TUNNEL_FQDN=${EDGE_TUNNEL_FQDN:-}
      # Subdomain fallback (#713): if subpath routing (#704/#708) fails, add:
      #   EDGE_TUNNEL_FQDN_FORGE, EDGE_TUNNEL_FQDN_CI, EDGE_TUNNEL_FQDN_CHAT
      # See docs/edge-routing-fallback.md for the full pivot plan.
      # Shared secret for Caddy ↔ chat forward_auth (#709)
      - FORWARD_AUTH_SECRET=${FORWARD_AUTH_SECRET:-}
    volumes:
      - ./docker/Caddyfile:/etc/caddy/Caddyfile
      - caddy_data:/data
      - /var/run/docker.sock:/var/run/docker.sock
      - ./secrets/tunnel_key:/run/secrets/tunnel_key:ro
      - ${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}:${CLAUDE_SHARED_DIR:-/var/lib/disinto/claude-shared}
-      - ${CLAUDE_CONFIG_FILE:-${HOME}/.claude.json}:/home/agent/.claude.json:ro
+      - ${HOME}/.claude.json:/home/agent/.claude.json:ro
    healthcheck:
      test: ["CMD", "curl", "-fsS", "http://localhost:2019/config/"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 15s
    depends_on:
      forgejo:
        condition: service_healthy
@ -607,12 +454,6 @@ LLAMAEOF
    command: ["caddy", "file-server", "--root", "/srv/site"]
    security_opt:
      - apparmor=unconfined
    healthcheck:
      test: ["CMD", "wget", "-q", "--spider", "http://localhost:2019/config/"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s
    volumes:
      - ./docker:/srv/site:ro
    networks:
@ -632,9 +473,10 @@ LLAMAEOF
      - disinto-net
    command: ["echo", "staging slot — replace with project image"]
-  # Chat container — Claude chat UI backend (#705)
+  # Chat container — Claude chat UI backend (#705, #707)
  # Internal service only; edge proxy routes to chat:8080
  # Sandbox hardened per #706 — no docker.sock, read-only rootfs, minimal caps
  # Separate identity mount (#707) to avoid OAuth refresh races with factory agents
  chat:
    build:
      context: ./docker/chat
@ -653,11 +495,9 @@ LLAMAEOF
    memswap_limit: 512m
    volumes:
      # Mount claude binary from host (same as agents)
-      - ${CLAUDE_BIN_DIR}:/usr/local/bin/claude:ro
+      - CLAUDE_BIN_PLACEHOLDER:/usr/local/bin/claude:ro
-      # Throwaway named volume for chat config (isolated from host ~/.claude)
+      # Separate Claude identity mount for chat — isolated from factory agents (#707)
-      - chat-config:/var/chat/config
+      - ${CHAT_CLAUDE_DIR:-${HOME}/.claude-chat}:/home/chat/.claude-chat
      # Chat history persistence: per-user NDJSON files on bind-mounted host volume
      - ${CHAT_HISTORY_DIR:-./state/chat-history}:/var/lib/chat/history
    environment:
      CHAT_HOST: "0.0.0.0"
      CHAT_PORT: "8080"
@ -666,18 +506,9 @@ LLAMAEOF
      CHAT_OAUTH_CLIENT_SECRET: ${CHAT_OAUTH_CLIENT_SECRET:-}
      EDGE_TUNNEL_FQDN: ${EDGE_TUNNEL_FQDN:-}
      DISINTO_CHAT_ALLOWED_USERS: ${DISINTO_CHAT_ALLOWED_USERS:-}
-      # Shared secret for Caddy forward_auth verify endpoint (#709)
+      # Point Claude to separate identity directory (#707)
-      FORWARD_AUTH_SECRET: ${FORWARD_AUTH_SECRET:-}
+      CLAUDE_CONFIG_DIR: /home/chat/.claude-chat/config
-      # Cost caps / rate limiting (#711)
+      CLAUDE_CREDENTIALS_DIR: /home/chat/.claude-chat/config/credentials
      CHAT_MAX_REQUESTS_PER_HOUR: ${CHAT_MAX_REQUESTS_PER_HOUR:-60}
      CHAT_MAX_REQUESTS_PER_DAY: ${CHAT_MAX_REQUESTS_PER_DAY:-500}
      CHAT_MAX_TOKENS_PER_DAY: ${CHAT_MAX_TOKENS_PER_DAY:-1000000}
    healthcheck:
      test: ["CMD", "python3", "-c", "import urllib.request; urllib.request.urlopen('http://localhost:8080/health')"]
      interval: 30s
      timeout: 5s
      retries: 3
      start_period: 10s
    networks:
      - disinto-net
@ -687,7 +518,6 @@ volumes:
  agent-data:
  project-repos:
  caddy_data:
  chat-config:
 networks:
  disinto-net:
@ -716,35 +546,20 @@ COMPOSEEOF
  fi
  # Append local-model agent services if any are configured
  # (must run before CLAUDE_BIN_PLACEHOLDER substitution so the placeholder
  # in local-model services is also resolved)
  _generate_local_model_services "$compose_file"
-  # Resolve the Claude CLI binary path and persist as CLAUDE_BIN_DIR in .env.
+  # Patch the Claude CLI binary path — resolve from host PATH at init time.
  # docker-compose.yml references ${CLAUDE_BIN_DIR} so the value must be set.
  local claude_bin
  claude_bin="$(command -v claude 2>/dev/null || true)"
  if [ -n "$claude_bin" ]; then
    # Resolve symlinks to get the real binary path
    claude_bin="$(readlink -f "$claude_bin")"
    sed -i "s|CLAUDE_BIN_PLACEHOLDER|${claude_bin}|g" "$compose_file"
  else
-    echo "Warning: claude CLI not found in PATH — set CLAUDE_BIN_DIR in .env manually" >&2
+    echo "Warning: claude CLI not found in PATH — update docker-compose.yml volumes manually" >&2
-    claude_bin="/usr/local/bin/claude"
+    sed -i "s|CLAUDE_BIN_PLACEHOLDER|/usr/local/bin/claude|g" "$compose_file"
  fi
  # Persist CLAUDE_BIN_DIR into .env so docker-compose can resolve it.
  local env_file="${FACTORY_ROOT}/.env"
  if [ -f "$env_file" ]; then
    if grep -q "^CLAUDE_BIN_DIR=" "$env_file" 2>/dev/null; then
      sed -i "s|^CLAUDE_BIN_DIR=.*|CLAUDE_BIN_DIR=${claude_bin}|" "$env_file"
    else
      printf 'CLAUDE_BIN_DIR=%s\n' "$claude_bin" >> "$env_file"
    fi
  else
    printf 'CLAUDE_BIN_DIR=%s\n' "$claude_bin" > "$env_file"
  fi
  # In build mode, replace image: with build: for locally-built images
  if [ "$use_build" = true ]; then
    sed -i 's|^\(  agents:\)|\1|' "$compose_file"
    sed -i '/^    image: ghcr\.io\/disinto\/agents:/{s|image: ghcr\.io/disinto/agents:.*|build:\n      context: .\n      dockerfile: docker/agents/Dockerfile|}' "$compose_file"
    sed -i '/^    image: ghcr\.io\/disinto\/edge:/{s|image: ghcr\.io/disinto/edge:.*|build: ./docker/edge|}' "$compose_file"
  fi
  echo "Created: ${compose_file}"
@ -763,11 +578,7 @@ _generate_agent_docker_impl() {
  fi
 }
-# Generate docker/Caddyfile for the edge proxy.
+# Generate docker/Caddyfile template for edge proxy.
 # **CANONICAL SOURCE**: This generator is the single source of truth for the Caddyfile.
 # Output path: ${FACTORY_ROOT}/docker/Caddyfile (gitignored — generated artifact).
 # The edge compose service mounts this path as /etc/caddy/Caddyfile.
 # On a fresh clone, `disinto init` calls generate_caddyfile before first `disinto up`.
 _generate_caddyfile_impl() {
  local docker_dir="${FACTORY_ROOT}/docker"
  local caddyfile="${docker_dir}/Caddyfile"
@ -803,20 +614,7 @@ _generate_caddyfile_impl() {
    }
    # Chat service — reverse proxy to disinto-chat backend (#705)
    # OAuth routes bypass forward_auth — unauthenticated users need these (#709)
    handle /chat/login {
        reverse_proxy chat:8080
    }
    handle /chat/oauth/callback {
        reverse_proxy chat:8080
    }
    # Defense-in-depth: forward_auth stamps X-Forwarded-User from session (#709)
    handle /chat/* {
        forward_auth chat:8080 {
            uri /chat/auth/verify
            copy_headers X-Forwarded-User
            header_up X-Forward-Auth-Secret {$FORWARD_AUTH_SECRET}
        }
        reverse_proxy chat:8080
    }
 }
--- a/lib/git-creds.sh
+++ b/lib/git-creds.sh
@ -35,35 +35,13 @@ configure_git_creds() {
  forge_host=$(printf '%s' "$FORGE_URL" | sed 's|https\?://||; s|/.*||')
  forge_proto=$(printf '%s' "$FORGE_URL" | sed 's|://.*||')
-  local log_fn="${_GIT_CREDS_LOG_FN:-echo}"
+  # Determine the bot username from FORGE_TOKEN identity (or default to dev-bot)
  # Determine the bot username from FORGE_TOKEN identity with retry/backoff.
  # Never fall back to a hardcoded default — a wrong username paired with the
  # real password produces a cryptic 401 that's much harder to diagnose than
  # a missing credential helper (#741).
  local bot_user=""
  if [ -n "${FORGE_TOKEN:-}" ]; then
-    local attempt
+    bot_user=$(curl -sf -H "Authorization: token ${FORGE_TOKEN}" \
-    for attempt in 1 2 3 4 5; do
+      "${FORGE_URL}/api/v1/user" 2>/dev/null | jq -r '.login // empty') || bot_user=""
      bot_user=$(curl -sf --max-time 5 -H "Authorization: token ${FORGE_TOKEN}" \
        "${FORGE_URL}/api/v1/user" 2>/dev/null | jq -r '.login // empty') || bot_user=""
      if [ -n "$bot_user" ]; then
        break
      fi
      $log_fn "WARNING: Forgejo not reachable (attempt ${attempt}/5) — retrying in ${attempt}s"
      sleep "$attempt"
    done
  fi
-
+  bot_user="${bot_user:-dev-bot}"
  if [ -z "$bot_user" ]; then
    $log_fn "ERROR: Could not determine bot username from FORGE_TOKEN after 5 attempts — credential helper NOT configured"
    $log_fn "ERROR: git push will fail until this is resolved. Restart the container after Forgejo is healthy."
    return 1
  fi
  # Export BOT_USER so downstream functions (e.g. configure_git_identity) can
  # reuse the resolved value without a redundant API call.
  export BOT_USER="$bot_user"
  local helper_path="${home_dir}/.git-credentials-helper"
@ -99,17 +77,6 @@ CREDEOF
  else
    git config --global --add safe.directory '*'
  fi
  # Verify the credential helper actually authenticates (#741).
  # A helper that was written with a valid username but a mismatched password
  # would silently 401 on every push — catch it now.
  if ! curl -sf --max-time 5 -u "${bot_user}:${FORGE_PASS}" \
    "${FORGE_URL}/api/v1/user" >/dev/null 2>&1; then
    $log_fn "ERROR: credential helper verification failed — ${bot_user}:FORGE_PASS rejected by Forgejo"
    rm -f "$helper_path"
    return 1
  fi
  $log_fn "Git credential helper verified: ${bot_user}@${forge_host}"
 }
 # repair_baked_cred_urls [--as RUN_AS_CMD] DIR [DIR ...]
--- a/lib/hvault.sh
+++ b/lib/hvault.sh
@ -1,279 +0,0 @@
 #!/usr/bin/env bash
 # hvault.sh — HashiCorp Vault helper module
 #
 # Typed, audited helpers for Vault KV v2 access so no script re-implements
 # `curl -H "X-Vault-Token: ..."` ad-hoc.
 #
 # Usage: source this file, then call any hvault_* function.
 #
 # Environment:
 #   VAULT_ADDR  — Vault server address (required, no default)
 #   VAULT_TOKEN — auth token (precedence: env > /etc/vault.d/root.token)
 #
 # All functions emit structured JSON errors to stderr on failure.
 set -euo pipefail
 # ── Internal helpers ─────────────────────────────────────────────────────────
 # _hvault_err — emit structured JSON error to stderr
 # Args: func_name, message, [detail]
 _hvault_err() {
  local func="$1" msg="$2" detail="${3:-}"
  jq -n --arg func "$func" --arg msg "$msg" --arg detail "$detail" \
    '{error:true,function:$func,message:$msg,detail:$detail}' >&2
 }
 # _hvault_resolve_token — resolve VAULT_TOKEN from env or token file
 _hvault_resolve_token() {
  if [ -n "${VAULT_TOKEN:-}" ]; then
    return 0
  fi
  local token_file="/etc/vault.d/root.token"
  if [ -f "$token_file" ]; then
    VAULT_TOKEN="$(cat "$token_file")"
    export VAULT_TOKEN
    return 0
  fi
  return 1
 }
 # _hvault_check_prereqs — validate VAULT_ADDR and VAULT_TOKEN are set
 # Args: caller function name
 _hvault_check_prereqs() {
  local caller="$1"
  if [ -z "${VAULT_ADDR:-}" ]; then
    _hvault_err "$caller" "VAULT_ADDR is not set" "export VAULT_ADDR before calling $caller"
    return 1
  fi
  if ! _hvault_resolve_token; then
    _hvault_err "$caller" "VAULT_TOKEN is not set and /etc/vault.d/root.token not found" \
      "export VAULT_TOKEN or write token to /etc/vault.d/root.token"
    return 1
  fi
 }
 # _hvault_request — execute a Vault API request
 # Args: method, path, [data]
 # Outputs: response body to stdout
 # Returns: 0 on 2xx, 1 otherwise (error JSON to stderr)
 _hvault_request() {
  local method="$1" path="$2" data="${3:-}"
  local url="${VAULT_ADDR}/v1/${path}"
  local http_code body
  local tmpfile
  tmpfile="$(mktemp)"
  local curl_args=(
    -s
    -w '%{http_code}'
    -H "X-Vault-Token: ${VAULT_TOKEN}"
    -H "Content-Type: application/json"
    -X "$method"
    -o "$tmpfile"
  )
  if [ -n "$data" ]; then
    curl_args+=(-d "$data")
  fi
  http_code="$(curl "${curl_args[@]}" "$url")" || {
    _hvault_err "_hvault_request" "curl failed" "url=$url"
    rm -f "$tmpfile"
    return 1
  }
  body="$(cat "$tmpfile")"
  rm -f "$tmpfile"
  # Check HTTP status — 2xx is success
  case "$http_code" in
    2[0-9][0-9])
      printf '%s' "$body"
      return 0
      ;;
    *)
      _hvault_err "_hvault_request" "HTTP $http_code" "$body"
      return 1
      ;;
  esac
 }
 # ── Public API ───────────────────────────────────────────────────────────────
 # hvault_kv_get PATH [KEY]
 #   Read a KV v2 secret at PATH, optionally extract a single KEY.
 #   Outputs: JSON value (full data object, or single key value)
 hvault_kv_get() {
  local path="${1:-}"
  local key="${2:-}"
  if [ -z "$path" ]; then
    _hvault_err "hvault_kv_get" "PATH is required" "usage: hvault_kv_get PATH [KEY]"
    return 1
  fi
  _hvault_check_prereqs "hvault_kv_get" || return 1
  local response
  response="$(_hvault_request GET "secret/data/${path}")" || return 1
  if [ -n "$key" ]; then
    printf '%s' "$response" | jq -e -r --arg key "$key" '.data.data[$key]' 2>/dev/null || {
      _hvault_err "hvault_kv_get" "key not found" "key=$key path=$path"
      return 1
    }
  else
    printf '%s' "$response" | jq -e '.data.data' 2>/dev/null || {
      _hvault_err "hvault_kv_get" "failed to parse response" "path=$path"
      return 1
    }
  fi
 }
 # hvault_kv_put PATH KEY=VAL [KEY=VAL ...]
 #   Write a KV v2 secret at PATH. Accepts one or more KEY=VAL pairs.
 hvault_kv_put() {
  local path="${1:-}"
  shift || true
  if [ -z "$path" ] || [ $# -eq 0 ]; then
    _hvault_err "hvault_kv_put" "PATH and at least one KEY=VAL required" \
      "usage: hvault_kv_put PATH KEY=VAL [KEY=VAL ...]"
    return 1
  fi
  _hvault_check_prereqs "hvault_kv_put" || return 1
  # Build JSON payload from KEY=VAL pairs entirely via jq
  local payload='{"data":{}}'
  for kv in "$@"; do
    local k="${kv%%=*}"
    local v="${kv#*=}"
    if [ "$k" = "$kv" ]; then
      _hvault_err "hvault_kv_put" "invalid KEY=VAL pair" "got: $kv"
      return 1
    fi
    payload="$(printf '%s' "$payload" | jq --arg k "$k" --arg v "$v" '.data[$k] = $v')"
  done
  _hvault_request POST "secret/data/${path}" "$payload" >/dev/null
 }
 # hvault_kv_list PATH
 #   List keys at a KV v2 path.
 #   Outputs: JSON array of key names
 hvault_kv_list() {
  local path="${1:-}"
  if [ -z "$path" ]; then
    _hvault_err "hvault_kv_list" "PATH is required" "usage: hvault_kv_list PATH"
    return 1
  fi
  _hvault_check_prereqs "hvault_kv_list" || return 1
  local response
  response="$(_hvault_request LIST "secret/metadata/${path}")" || return 1
  printf '%s' "$response" | jq -e '.data.keys' 2>/dev/null || {
    _hvault_err "hvault_kv_list" "failed to parse response" "path=$path"
    return 1
  }
 }
 # hvault_policy_apply NAME FILE
 #   Idempotent policy upsert — create or update a Vault policy.
 hvault_policy_apply() {
  local name="${1:-}"
  local file="${2:-}"
  if [ -z "$name" ] || [ -z "$file" ]; then
    _hvault_err "hvault_policy_apply" "NAME and FILE are required" \
      "usage: hvault_policy_apply NAME FILE"
    return 1
  fi
  if [ ! -f "$file" ]; then
    _hvault_err "hvault_policy_apply" "policy file not found" "file=$file"
    return 1
  fi
  _hvault_check_prereqs "hvault_policy_apply" || return 1
  local policy_content
  policy_content="$(cat "$file")"
  local payload
  payload="$(jq -n --arg policy "$policy_content" '{"policy": $policy}')"
  _hvault_request PUT "sys/policies/acl/${name}" "$payload" >/dev/null
 }
 # hvault_jwt_login ROLE JWT
 #   Exchange a JWT for a short-lived Vault token.
 #   Outputs: client token string
 hvault_jwt_login() {
  local role="${1:-}"
  local jwt="${2:-}"
  if [ -z "$role" ] || [ -z "$jwt" ]; then
    _hvault_err "hvault_jwt_login" "ROLE and JWT are required" \
      "usage: hvault_jwt_login ROLE JWT"
    return 1
  fi
  # Only need VAULT_ADDR, not VAULT_TOKEN (we're obtaining a token)
  if [ -z "${VAULT_ADDR:-}" ]; then
    _hvault_err "hvault_jwt_login" "VAULT_ADDR is not set"
    return 1
  fi
  local payload
  payload="$(jq -n --arg role "$role" --arg jwt "$jwt" \
    '{"role": $role, "jwt": $jwt}')"
  local response
  # JWT login does not require an existing token — use curl directly
  local tmpfile http_code
  tmpfile="$(mktemp)"
  http_code="$(curl -s -w '%{http_code}' \
    -H "Content-Type: application/json" \
    -X POST \
    -d "$payload" \
    -o "$tmpfile" \
    "${VAULT_ADDR}/v1/auth/jwt/login")" || {
    _hvault_err "hvault_jwt_login" "curl failed"
    rm -f "$tmpfile"
    return 1
  }
  local body
  body="$(cat "$tmpfile")"
  rm -f "$tmpfile"
  case "$http_code" in
    2[0-9][0-9])
      printf '%s' "$body" | jq -e -r '.auth.client_token' 2>/dev/null || {
        _hvault_err "hvault_jwt_login" "failed to extract client_token" "$body"
        return 1
      }
      ;;
    *)
      _hvault_err "hvault_jwt_login" "HTTP $http_code" "$body"
      return 1
      ;;
  esac
 }
 # hvault_token_lookup
 #   Returns TTL, policies, and accessor for the current token.
 #   Outputs: JSON object with ttl, policies, accessor fields
 hvault_token_lookup() {
  _hvault_check_prereqs "hvault_token_lookup" || return 1
  local response
  response="$(_hvault_request GET "auth/token/lookup-self")" || return 1
  printf '%s' "$response" | jq -e '{
    ttl: .data.ttl,
    policies: .data.policies,
    accessor: .data.accessor,
    display_name: .data.display_name
  }' 2>/dev/null || {
    _hvault_err "hvault_token_lookup" "failed to parse token info"
    return 1
  }
 }
--- a/lib/init/nomad/cluster-up.sh
+++ b/lib/init/nomad/cluster-up.sh
@ -1,338 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # lib/init/nomad/cluster-up.sh — Empty Nomad+Vault cluster orchestrator (S0.4)
 #
 # Wires together the S0.1–S0.3 building blocks into one idempotent
 # "bring up a single-node Nomad+Vault cluster" script:
 #
 #   1. install.sh                  (nomad + vault binaries)
 #   2. systemd-nomad.sh            (nomad.service — unit + enable, not started)
 #   3. systemd-vault.sh            (vault.service — unit + vault.hcl + enable)
 #   4. Host-volume dirs            (/srv/disinto/* matching nomad/client.hcl)
 #   5. /etc/nomad.d/*.hcl          (server.hcl + client.hcl from repo)
 #   6. vault-init.sh               (first-run init + unseal + persist keys)
 #   7. systemctl start vault       (auto-unseal via ExecStartPost; poll)
 #   8. systemctl start nomad       (poll until ≥1 ready node)
 #   9. /etc/profile.d/disinto-nomad.sh  (VAULT_ADDR + NOMAD_ADDR for shells)
 #
 # This is the "empty cluster" orchestrator — no jobs deployed. Subsequent
 # Step-1 issues layer job deployment on top of this checkpoint.
 #
 # Idempotency contract:
 #   Running twice back-to-back on a healthy box is a no-op. Each sub-step
 #   is itself idempotent — see install.sh / systemd-*.sh / vault-init.sh
 #   headers for the per-step contract. Fast-paths in steps 7 and 8 skip
 #   the systemctl start when the service is already active + healthy.
 #
 # Usage:
 #   sudo lib/init/nomad/cluster-up.sh            # bring cluster up
 #   sudo lib/init/nomad/cluster-up.sh --dry-run  # print step list, exit 0
 #
 # Environment (override polling for slow boxes):
 #   VAULT_POLL_SECS  max seconds to wait for vault to unseal (default: 30)
 #   NOMAD_POLL_SECS  max seconds to wait for nomad node=ready (default: 60)
 #
 # Exit codes:
 #   0  success (cluster up, or already up)
 #   1  precondition or step failure
 # =============================================================================
 set -euo pipefail
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_ROOT="$(cd "${SCRIPT_DIR}/../../.." && pwd)"
 # Sub-scripts (siblings in this directory).
 INSTALL_SH="${SCRIPT_DIR}/install.sh"
 SYSTEMD_NOMAD_SH="${SCRIPT_DIR}/systemd-nomad.sh"
 SYSTEMD_VAULT_SH="${SCRIPT_DIR}/systemd-vault.sh"
 VAULT_INIT_SH="${SCRIPT_DIR}/vault-init.sh"
 # In-repo Nomad configs copied to /etc/nomad.d/.
 NOMAD_CONFIG_DIR="/etc/nomad.d"
 NOMAD_SERVER_HCL_SRC="${REPO_ROOT}/nomad/server.hcl"
 NOMAD_CLIENT_HCL_SRC="${REPO_ROOT}/nomad/client.hcl"
 # /etc/profile.d entry — makes VAULT_ADDR + NOMAD_ADDR available to
 # interactive shells without requiring the operator to source anything.
 PROFILE_D_FILE="/etc/profile.d/disinto-nomad.sh"
 # Host-volume paths — MUST match the `host_volume "..."` declarations
 # in nomad/client.hcl. Adding a host_volume block there requires adding
 # its path here so the dir exists before nomad starts (otherwise client
 # fingerprinting fails and the node stays in "initializing").
 HOST_VOLUME_DIRS=(
  "/srv/disinto/forgejo-data"
  "/srv/disinto/woodpecker-data"
  "/srv/disinto/agent-data"
  "/srv/disinto/project-repos"
  "/srv/disinto/caddy-data"
  "/srv/disinto/chat-history"
  "/srv/disinto/ops-repo"
 )
 # Default API addresses — matches the listener bindings in
 # nomad/server.hcl and nomad/vault.hcl. If either file ever moves
 # off 127.0.0.1 / default port, update both places together.
 VAULT_ADDR_DEFAULT="http://127.0.0.1:8200"
 NOMAD_ADDR_DEFAULT="http://127.0.0.1:4646"
 VAULT_POLL_SECS="${VAULT_POLL_SECS:-30}"
 NOMAD_POLL_SECS="${NOMAD_POLL_SECS:-60}"
 log() { printf '[cluster-up] %s\n' "$*"; }
 die() { printf '[cluster-up] ERROR: %s\n' "$*" >&2; exit 1; }
 # ── Flag parsing ─────────────────────────────────────────────────────────────
 dry_run=false
 while [ $# -gt 0 ]; do
  case "$1" in
    --dry-run) dry_run=true; shift ;;
    -h|--help)
      cat <<EOF
 Usage: sudo $(basename "$0") [--dry-run]
 Brings up an empty single-node Nomad+Vault cluster (idempotent).
  --dry-run   Print the step list without performing any action.
 EOF
      exit 0
      ;;
    *) die "unknown flag: $1" ;;
  esac
 done
 # ── Dry-run: print step list + exit ──────────────────────────────────────────
 if [ "$dry_run" = true ]; then
  cat <<EOF
 [dry-run] Step 1/9: install nomad + vault binaries
  → sudo ${INSTALL_SH}
 [dry-run] Step 2/9: write + enable nomad.service (NOT started)
  → sudo ${SYSTEMD_NOMAD_SH}
 [dry-run] Step 3/9: write + enable vault.service + vault.hcl (NOT started)
  → sudo ${SYSTEMD_VAULT_SH}
 [dry-run] Step 4/9: create host-volume dirs under /srv/disinto/
 EOF
  for d in "${HOST_VOLUME_DIRS[@]}"; do
    printf '  → install -d -m 0755 %s\n' "$d"
  done
  cat <<EOF
 [dry-run] Step 5/9: install /etc/nomad.d/server.hcl + client.hcl from repo
  → ${NOMAD_SERVER_HCL_SRC} → ${NOMAD_CONFIG_DIR}/server.hcl
  → ${NOMAD_CLIENT_HCL_SRC} → ${NOMAD_CONFIG_DIR}/client.hcl
 [dry-run] Step 6/9: first-run vault init + persist unseal.key + root.token
  → sudo ${VAULT_INIT_SH}
 [dry-run] Step 7/9: systemctl start vault + poll until unsealed (≤${VAULT_POLL_SECS}s)
 [dry-run] Step 8/9: systemctl start nomad + poll until ≥1 node ready (≤${NOMAD_POLL_SECS}s)
 [dry-run] Step 9/9: write ${PROFILE_D_FILE}
  → export VAULT_ADDR=${VAULT_ADDR_DEFAULT}
  → export NOMAD_ADDR=${NOMAD_ADDR_DEFAULT}
 Dry run complete — no changes made.
 EOF
  exit 0
 fi
 # ── Preconditions ────────────────────────────────────────────────────────────
 if [ "$(id -u)" -ne 0 ]; then
  die "must run as root (spawns install/systemd/vault-init sub-scripts)"
 fi
 command -v systemctl >/dev/null 2>&1 \
  || die "systemctl not found (systemd required)"
 for f in "$INSTALL_SH" "$SYSTEMD_NOMAD_SH" "$SYSTEMD_VAULT_SH" "$VAULT_INIT_SH"; do
  [ -x "$f" ] || die "sub-script missing or non-executable: ${f}"
 done
 [ -f "$NOMAD_SERVER_HCL_SRC" ] \
  || die "source config not found: ${NOMAD_SERVER_HCL_SRC}"
 [ -f "$NOMAD_CLIENT_HCL_SRC" ] \
  || die "source config not found: ${NOMAD_CLIENT_HCL_SRC}"
 # ── Helpers ──────────────────────────────────────────────────────────────────
 # install_file_if_differs SRC DST MODE
 #   Copy SRC to DST (root:root with MODE) iff on-disk content differs.
 #   No-op + log otherwise — preserves mtime, avoids spurious reloads.
 install_file_if_differs() {
  local src="$1" dst="$2" mode="$3"
  if [ -f "$dst" ] && cmp -s "$src" "$dst"; then
    log "unchanged: ${dst}"
    return 0
  fi
  log "writing: ${dst}"
  install -m "$mode" -o root -g root "$src" "$dst"
 }
 # vault_status_json — echo `vault status -format=json`, or '' on unreachable.
 #   vault status exit codes: 0 = unsealed, 2 = sealed/uninit, 1 = unreachable.
 #   We treat all of 0/2 as "reachable with state"; 1 yields empty output.
 #   Wrapped in `|| true` so set -e doesn't abort on exit 2 (the expected
 #   sealed-state case during first-boot polling).
 vault_status_json() {
  VAULT_ADDR="$VAULT_ADDR_DEFAULT" vault status -format=json 2>/dev/null || true
 }
 # vault_is_unsealed — true iff vault reachable AND initialized AND unsealed.
 vault_is_unsealed() {
  local out init sealed
  out="$(vault_status_json)"
  [ -n "$out" ] || return 1
  init="$(printf '%s' "$out" | jq -r '.initialized' 2>/dev/null)" || init=""
  sealed="$(printf '%s' "$out" | jq -r '.sealed' 2>/dev/null)" || sealed=""
  [ "$init" = "true" ] && [ "$sealed" = "false" ]
 }
 # nomad_ready_count — echo the number of ready nodes, or 0 on error.
 #   `nomad node status -json` returns a JSON array of nodes, each with a
 #   .Status field ("initializing" | "ready" | "down" | "disconnected").
 nomad_ready_count() {
  local out
  out="$(NOMAD_ADDR="$NOMAD_ADDR_DEFAULT" nomad node status -json 2>/dev/null || true)"
  if [ -z "$out" ]; then
    printf '0'
    return 0
  fi
  printf '%s' "$out" \
    | jq '[.[] | select(.Status == "ready")] | length' 2>/dev/null \
    || printf '0'
 }
 # nomad_has_ready_node — true iff nomad_ready_count ≥ 1. Wrapper exists
 # so poll_until_healthy can call it as a single-arg command name.
 nomad_has_ready_node() { [ "$(nomad_ready_count)" -ge 1 ]; }
 # _die_with_service_status SVC REASON
 #   Log + dump `systemctl status SVC` to stderr + die with REASON. Factored
 #   out so the poll helper doesn't carry three copies of the same dump.
 _die_with_service_status() {
  local svc="$1" reason="$2"
  log "${svc}.service ${reason} — systemctl status follows:"
  systemctl --no-pager --full status "$svc" >&2 || true
  die "${svc}.service ${reason}"
 }
 # poll_until_healthy SVC CHECK_CMD TIMEOUT
 #   Tick once per second for up to TIMEOUT seconds, invoking CHECK_CMD as a
 #   command name (no arguments). Returns 0 on the first successful check.
 #   Fails fast via _die_with_service_status if SVC enters systemd "failed"
 #   state, and dies with a status dump if TIMEOUT elapses before CHECK_CMD
 #   succeeds. Replaces the two in-line ready=1/break/sleep poll loops that
 #   would otherwise each duplicate the same pattern already in vault-init.sh.
 poll_until_healthy() {
  local svc="$1" check="$2" timeout="$3"
  local waited=0
  until [ "$waited" -ge "$timeout" ]; do
    systemctl is-failed --quiet "$svc" \
      && _die_with_service_status "$svc" "entered failed state during startup"
    if "$check"; then
      log "${svc} healthy after ${waited}s"
      return 0
    fi
    waited=$((waited + 1))
    sleep 1
  done
  _die_with_service_status "$svc" "not healthy within ${timeout}s"
 }
 # ── Step 1/9: install.sh (nomad + vault binaries) ────────────────────────────
 log "── Step 1/9: install nomad + vault binaries ──"
 "$INSTALL_SH"
 # ── Step 2/9: systemd-nomad.sh (unit + enable, not started) ──────────────────
 log "── Step 2/9: install nomad.service (enable, not start) ──"
 "$SYSTEMD_NOMAD_SH"
 # ── Step 3/9: systemd-vault.sh (unit + vault.hcl + enable) ───────────────────
 log "── Step 3/9: install vault.service + vault.hcl (enable, not start) ──"
 "$SYSTEMD_VAULT_SH"
 # ── Step 4/9: host-volume dirs matching nomad/client.hcl ─────────────────────
 log "── Step 4/9: host-volume dirs under /srv/disinto/ ──"
 # Parent /srv/disinto/ first (install -d handles missing parents, but being
 # explicit makes the log output read naturally as a top-down creation).
 install -d -m 0755 -o root -g root "/srv/disinto"
 for d in "${HOST_VOLUME_DIRS[@]}"; do
  if [ -d "$d" ]; then
    log "unchanged: ${d}"
  else
    log "creating: ${d}"
    install -d -m 0755 -o root -g root "$d"
  fi
 done
 # ── Step 5/9: /etc/nomad.d/server.hcl + client.hcl ───────────────────────────
 log "── Step 5/9: install /etc/nomad.d/{server,client}.hcl ──"
 # systemd-nomad.sh already created /etc/nomad.d/. Re-assert for clarity +
 # in case someone runs cluster-up.sh with an exotic step ordering later.
 install -d -m 0755 -o root -g root "$NOMAD_CONFIG_DIR"
 install_file_if_differs "$NOMAD_SERVER_HCL_SRC" "${NOMAD_CONFIG_DIR}/server.hcl" 0644
 install_file_if_differs "$NOMAD_CLIENT_HCL_SRC" "${NOMAD_CONFIG_DIR}/client.hcl" 0644
 # ── Step 6/9: vault-init (first-run init + unseal + persist keys) ────────────
 log "── Step 6/9: vault-init (no-op after first run) ──"
 # vault-init.sh spawns a temporary vault server if systemd isn't managing
 # one, runs `operator init`, writes unseal.key + root.token, unseals once,
 # then stops the temp server (EXIT trap). After it returns, port 8200 is
 # free for systemctl-managed vault to take in step 7.
 "$VAULT_INIT_SH"
 # ── Step 7/9: systemctl start vault + poll until unsealed ────────────────────
 log "── Step 7/9: start vault + poll until unsealed ──"
 # Fast-path when vault.service is already active and Vault reports
 # initialized=true,sealed=false — re-runs are a no-op.
 if systemctl is-active --quiet vault && vault_is_unsealed; then
  log "vault already active + unsealed — skip start"
 else
  systemctl start vault
  poll_until_healthy vault vault_is_unsealed "$VAULT_POLL_SECS"
 fi
 # ── Step 8/9: systemctl start nomad + poll until ≥1 node ready ───────────────
 log "── Step 8/9: start nomad + poll until ≥1 node ready ──"
 if systemctl is-active --quiet nomad && nomad_has_ready_node; then
  log "nomad already active + ≥1 node ready — skip start"
 else
  systemctl start nomad
  poll_until_healthy nomad nomad_has_ready_node "$NOMAD_POLL_SECS"
 fi
 # ── Step 9/9: /etc/profile.d/disinto-nomad.sh ────────────────────────────────
 log "── Step 9/9: write ${PROFILE_D_FILE} ──"
 # Shell rc fragments in /etc/profile.d/ are sourced by /etc/profile for
 # every interactive login shell. Setting VAULT_ADDR + NOMAD_ADDR here means
 # the operator can run `vault status` / `nomad node status` straight after
 # `ssh factory-box` without fumbling env vars.
 desired_profile="# /etc/profile.d/disinto-nomad.sh — written by lib/init/nomad/cluster-up.sh
 # Interactive-shell defaults for Vault + Nomad clients on this box.
 export VAULT_ADDR=${VAULT_ADDR_DEFAULT}
 export NOMAD_ADDR=${NOMAD_ADDR_DEFAULT}
 "
 if [ -f "$PROFILE_D_FILE" ] \
   && printf '%s' "$desired_profile" | cmp -s - "$PROFILE_D_FILE"; then
  log "unchanged: ${PROFILE_D_FILE}"
 else
  log "writing: ${PROFILE_D_FILE}"
  # Subshell + EXIT trap: guarantees the tempfile is cleaned up on both
  # success AND set-e-induced failure of `install`. A function-scoped
  # RETURN trap does NOT fire on errexit-abort in bash — the subshell is
  # the reliable cleanup boundary here.
  (
    tmp="$(mktemp)"
    trap 'rm -f "$tmp"' EXIT
    printf '%s' "$desired_profile" > "$tmp"
    install -m 0644 -o root -g root "$tmp" "$PROFILE_D_FILE"
  )
 fi
 log "── done: empty nomad+vault cluster is up ──"
 log "   Vault:  ${VAULT_ADDR_DEFAULT}  (Sealed=false Initialized=true)"
 log "   Nomad:  ${NOMAD_ADDR_DEFAULT}  (≥1 node ready)"
--- a/lib/init/nomad/install.sh
+++ b/lib/init/nomad/install.sh
@ -1,143 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # lib/init/nomad/install.sh — Idempotent apt install of HashiCorp Nomad + Vault
 #
 # Part of the Nomad+Vault migration. Installs both the `nomad` binary (S0.2,
 # issue #822) and the `vault` binary (S0.3, issue #823) from the same
 # HashiCorp apt repository. Does NOT configure, start, or enable any systemd
 # unit — lib/init/nomad/systemd-nomad.sh and lib/init/nomad/systemd-vault.sh
 # own that. Does NOT wire this script into `disinto init` — S0.4 owns that.
 #
 # Idempotency contract:
 #   - Running twice back-to-back is a no-op once both target versions are
 #     installed and the apt source is in place.
 #   - Adds the HashiCorp apt keyring only if it is absent.
 #   - Adds the HashiCorp apt sources list only if it is absent.
 #   - Skips `apt-get install` for any package whose installed version already
 #     matches the pin. If both are at pin, exits before touching apt.
 #
 # Configuration:
 #   NOMAD_VERSION  — pinned Nomad version (default: see below). Apt package
 #                    name is versioned as "nomad=<version>-1".
 #   VAULT_VERSION  — pinned Vault version (default: see below). Apt package
 #                    name is versioned as "vault=<version>-1".
 #
 # Usage:
 #   sudo lib/init/nomad/install.sh
 #   sudo NOMAD_VERSION=1.9.5 VAULT_VERSION=1.18.5 lib/init/nomad/install.sh
 #
 # Exit codes:
 #   0  success (installed or already present)
 #   1  precondition failure (not Debian/Ubuntu, missing tools, not root)
 # =============================================================================
 set -euo pipefail
 # Pin to specific 1.x releases. Bump here, not at call sites.
 NOMAD_VERSION="${NOMAD_VERSION:-1.9.5}"
 VAULT_VERSION="${VAULT_VERSION:-1.18.5}"
 HASHICORP_KEYRING="/usr/share/keyrings/hashicorp-archive-keyring.gpg"
 HASHICORP_SOURCES="/etc/apt/sources.list.d/hashicorp.list"
 HASHICORP_GPG_URL="https://apt.releases.hashicorp.com/gpg"
 HASHICORP_REPO_URL="https://apt.releases.hashicorp.com"
 log() { printf '[install] %s\n' "$*"; }
 die() { printf '[install] ERROR: %s\n' "$*" >&2; exit 1; }
 # _installed_version BINARY
 #   Echoes the installed semver for `nomad` or `vault` (e.g. "1.9.5").
 #   Both tools print their version on the first line of `<bin> version` as
 #   "<Name> v<semver>..." — the shared awk extracts $2 with the leading "v"
 #   stripped. Empty string when the binary is absent or output is unexpected.
 _installed_version() {
  local bin="$1"
  command -v "$bin" >/dev/null 2>&1 || { printf ''; return 0; }
  "$bin" version 2>/dev/null \
    | awk 'NR==1 {sub(/^v/, "", $2); print $2; exit}'
 }
 # ── Preconditions ────────────────────────────────────────────────────────────
 if [ "$(id -u)" -ne 0 ]; then
  die "must run as root (needs apt-get + /usr/share/keyrings write access)"
 fi
 for bin in apt-get gpg curl lsb_release; do
  command -v "$bin" >/dev/null 2>&1 \
    || die "required binary not found: ${bin}"
 done
 CODENAME="$(lsb_release -cs)"
 [ -n "$CODENAME" ] || die "lsb_release returned empty codename"
 # ── Fast-path: are both already at desired versions? ─────────────────────────
 nomad_installed="$(_installed_version nomad)"
 vault_installed="$(_installed_version vault)"
 need_pkgs=()
 if [ "$nomad_installed" = "$NOMAD_VERSION" ]; then
  log "nomad ${NOMAD_VERSION} already installed"
 else
  need_pkgs+=("nomad=${NOMAD_VERSION}-1")
 fi
 if [ "$vault_installed" = "$VAULT_VERSION" ]; then
  log "vault ${VAULT_VERSION} already installed"
 else
  need_pkgs+=("vault=${VAULT_VERSION}-1")
 fi
 if [ "${#need_pkgs[@]}" -eq 0 ]; then
  log "nothing to do"
  exit 0
 fi
 # ── Ensure HashiCorp apt keyring ─────────────────────────────────────────────
 if [ ! -f "$HASHICORP_KEYRING" ]; then
  log "adding HashiCorp apt keyring → ${HASHICORP_KEYRING}"
  tmpkey="$(mktemp)"
  trap 'rm -f "$tmpkey"' EXIT
  curl -fsSL "$HASHICORP_GPG_URL" -o "$tmpkey" \
    || die "failed to fetch HashiCorp GPG key from ${HASHICORP_GPG_URL}"
  gpg --dearmor -o "$HASHICORP_KEYRING" < "$tmpkey" \
    || die "failed to dearmor HashiCorp GPG key"
  chmod 0644 "$HASHICORP_KEYRING"
  rm -f "$tmpkey"
  trap - EXIT
 else
  log "HashiCorp apt keyring already present"
 fi
 # ── Ensure HashiCorp apt sources list ────────────────────────────────────────
 desired_source="deb [signed-by=${HASHICORP_KEYRING}] ${HASHICORP_REPO_URL} ${CODENAME} main"
 if [ ! -f "$HASHICORP_SOURCES" ] \
   || ! grep -qxF "$desired_source" "$HASHICORP_SOURCES"; then
  log "writing HashiCorp apt sources list → ${HASHICORP_SOURCES}"
  printf '%s\n' "$desired_source" > "$HASHICORP_SOURCES"
  apt_update_needed=1
 else
  log "HashiCorp apt sources list already present"
  apt_update_needed=0
 fi
 # ── Install the pinned versions ──────────────────────────────────────────────
 if [ "$apt_update_needed" -eq 1 ]; then
  log "running apt-get update"
  DEBIAN_FRONTEND=noninteractive apt-get update -qq \
    || die "apt-get update failed"
 fi
 log "installing ${need_pkgs[*]}"
 DEBIAN_FRONTEND=noninteractive apt-get install -y --no-install-recommends \
  "${need_pkgs[@]}" \
  || die "apt-get install ${need_pkgs[*]} failed"
 # ── Verify ───────────────────────────────────────────────────────────────────
 final_nomad="$(_installed_version nomad)"
 if [ "$final_nomad" != "$NOMAD_VERSION" ]; then
  die "post-install check: expected nomad ${NOMAD_VERSION}, got '${final_nomad}'"
 fi
 final_vault="$(_installed_version vault)"
 if [ "$final_vault" != "$VAULT_VERSION" ]; then
  die "post-install check: expected vault ${VAULT_VERSION}, got '${final_vault}'"
 fi
 log "nomad ${NOMAD_VERSION} + vault ${VAULT_VERSION} installed successfully"
--- a/lib/init/nomad/lib-systemd.sh
+++ b/lib/init/nomad/lib-systemd.sh
@ -1,77 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # lib/init/nomad/lib-systemd.sh — Shared idempotent systemd-unit installer
 #
 # Sourced by lib/init/nomad/systemd-nomad.sh and lib/init/nomad/systemd-vault.sh
 # (and any future sibling) to collapse the "write unit if content differs,
 # daemon-reload, enable (never start)" boilerplate.
 #
 # Install-but-don't-start is the invariant this helper enforces — mid-migration
 # installers land files and enable units; the orchestrator (S0.4) starts them.
 #
 # Public API (sourced into caller scope):
 #
 #   systemd_require_preconditions UNIT_PATH
 #     Asserts the caller is uid 0 and `systemctl` is on $PATH. Calls the
 #     caller's die() with a UNIT_PATH-scoped message on failure.
 #
 #   systemd_install_unit UNIT_PATH UNIT_NAME UNIT_CONTENT
 #     Writes UNIT_CONTENT to UNIT_PATH (0644 root:root) only if on-disk
 #     content differs. If written, runs `systemctl daemon-reload`. Then
 #     enables UNIT_NAME (no-op if already enabled). Never starts the unit.
 #
 # Caller contract:
 #   - Callers MUST define `log()` and `die()` before sourcing this file (we
 #     call log() for status chatter and rely on the caller's error-handling
 #     stance; `set -e` propagates install/cmp/systemctl failures).
 # =============================================================================
 # systemd_require_preconditions UNIT_PATH
 systemd_require_preconditions() {
  local unit_path="$1"
  if [ "$(id -u)" -ne 0 ]; then
    die "must run as root (needs write access to ${unit_path})"
  fi
  command -v systemctl >/dev/null 2>&1 \
    || die "systemctl not found (systemd is required)"
 }
 # systemd_install_unit UNIT_PATH UNIT_NAME UNIT_CONTENT
 systemd_install_unit() {
  local unit_path="$1"
  local unit_name="$2"
  local unit_content="$3"
  local needs_reload=0
  if [ ! -f "$unit_path" ] \
     || ! printf '%s\n' "$unit_content" | cmp -s - "$unit_path"; then
    log "writing unit → ${unit_path}"
    # Subshell-scoped EXIT trap guarantees the temp file is removed on
    # both success AND set-e-induced failure of `install`. A function-
    # scoped RETURN trap does NOT fire on errexit-abort (bash only runs
    # RETURN on normal function exit), so the subshell is the reliable
    # cleanup boundary. It's also isolated from the caller's EXIT trap.
    (
      local tmp
      tmp="$(mktemp)"
      trap 'rm -f "$tmp"' EXIT
      printf '%s\n' "$unit_content" > "$tmp"
      install -m 0644 -o root -g root "$tmp" "$unit_path"
    )
    needs_reload=1
  else
    log "unit file already up to date"
  fi
  if [ "$needs_reload" -eq 1 ]; then
    log "systemctl daemon-reload"
    systemctl daemon-reload
  fi
  if systemctl is-enabled --quiet "$unit_name" 2>/dev/null; then
    log "${unit_name} already enabled"
  else
    log "systemctl enable ${unit_name}"
    systemctl enable "$unit_name" >/dev/null
  fi
 }
--- a/lib/init/nomad/systemd-nomad.sh
+++ b/lib/init/nomad/systemd-nomad.sh
@ -1,102 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # lib/init/nomad/systemd-nomad.sh — Idempotent systemd unit installer for Nomad
 #
 # Part of the Nomad+Vault migration (S0.2, issue #822). Writes
 # /etc/systemd/system/nomad.service pointing at /etc/nomad.d/ and runs
 # `systemctl enable nomad` WITHOUT starting the service — we don't launch
 # the cluster until S0.4 wires everything together.
 #
 # Idempotency contract:
 #   - Existing unit file is NOT rewritten when on-disk content already
 #     matches the desired content (avoids spurious `daemon-reload`).
 #   - `systemctl enable` on an already-enabled unit is a no-op.
 #   - This script is safe to run unconditionally before every factory boot.
 #
 # Preconditions:
 #   - nomad binary installed (see lib/init/nomad/install.sh)
 #   - /etc/nomad.d/ will hold server.hcl / client.hcl (placed by S0.4)
 #
 # Usage:
 #   sudo lib/init/nomad/systemd-nomad.sh
 #
 # Exit codes:
 #   0  success (unit installed + enabled, or already so)
 #   1  precondition failure (not root, no systemctl, no nomad binary)
 # =============================================================================
 set -euo pipefail
 UNIT_PATH="/etc/systemd/system/nomad.service"
 NOMAD_CONFIG_DIR="/etc/nomad.d"
 NOMAD_DATA_DIR="/var/lib/nomad"
 log() { printf '[systemd-nomad] %s\n' "$*"; }
 die() { printf '[systemd-nomad] ERROR: %s\n' "$*" >&2; exit 1; }
 # shellcheck source=lib-systemd.sh
 . "$(dirname "${BASH_SOURCE[0]}")/lib-systemd.sh"
 # ── Preconditions ────────────────────────────────────────────────────────────
 systemd_require_preconditions "$UNIT_PATH"
 NOMAD_BIN="$(command -v nomad 2>/dev/null || true)"
 [ -n "$NOMAD_BIN" ] \
  || die "nomad binary not found — run lib/init/nomad/install.sh first"
 # ── Desired unit content ─────────────────────────────────────────────────────
 # Upstream-recommended baseline (https://developer.hashicorp.com/nomad/docs/install/production/deployment-guide)
 # trimmed for a single-node combined server+client dev box.
 #   - Wants=/After= network-online: nomad must have networking up.
 #   - User/Group=root: the Docker driver needs root to talk to dockerd.
 #   - LimitNOFILE/LimitNPROC=infinity: avoid Nomad's startup warning.
 #   - KillSignal=SIGINT: triggers Nomad's graceful shutdown path.
 #   - Restart=on-failure with a bounded burst to avoid crash-loops eating the
 #     journal when /etc/nomad.d/ is mis-configured.
 read -r -d '' DESIRED_UNIT <<EOF || true
 [Unit]
 Description=Nomad
 Documentation=https://developer.hashicorp.com/nomad/docs
 Wants=network-online.target
 After=network-online.target
 # When Docker is present, ensure dockerd is up before nomad starts — the
 # Docker task driver needs the daemon socket available at startup.
 Wants=docker.service
 After=docker.service
 [Service]
 Type=notify
 User=root
 Group=root
 ExecReload=/bin/kill -HUP \$MAINPID
 ExecStart=${NOMAD_BIN} agent -config=${NOMAD_CONFIG_DIR}
 KillMode=process
 KillSignal=SIGINT
 LimitNOFILE=infinity
 LimitNPROC=infinity
 Restart=on-failure
 RestartSec=2
 StartLimitBurst=3
 StartLimitIntervalSec=10
 TasksMax=infinity
 OOMScoreAdjust=-1000
 [Install]
 WantedBy=multi-user.target
 EOF
 # ── Ensure config + data dirs exist ──────────────────────────────────────────
 # We do not populate /etc/nomad.d/ here (that's S0.4). We do create the
 # directory so `nomad agent -config=/etc/nomad.d` doesn't error if the unit
 # is started before hcl files are dropped in.
 for d in "$NOMAD_CONFIG_DIR" "$NOMAD_DATA_DIR"; do
  if [ ! -d "$d" ]; then
    log "creating ${d}"
    install -d -m 0755 "$d"
  fi
 done
 # ── Install + reload + enable (shared with systemd-vault.sh via lib-systemd) ─
 systemd_install_unit "$UNIT_PATH" "nomad.service" "$DESIRED_UNIT"
 log "done — unit installed and enabled (NOT started; S0.4 brings the cluster up)"
--- a/lib/init/nomad/systemd-vault.sh
+++ b/lib/init/nomad/systemd-vault.sh
@ -1,151 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # lib/init/nomad/systemd-vault.sh — Idempotent systemd unit installer for Vault
 #
 # Part of the Nomad+Vault migration (S0.3, issue #823). Lands three things:
 #   1. /etc/vault.d/               (0755 root:root)
 #   2. /etc/vault.d/vault.hcl      (copy of nomad/vault.hcl, 0644 root:root)
 #   3. /var/lib/vault/data/        (0700 root:root, Vault file-storage backend)
 #   4. /etc/systemd/system/vault.service  (0644 root:root)
 #
 # Then `systemctl enable vault` WITHOUT starting the service. Bootstrap
 # order is:
 #   lib/init/nomad/install.sh         (nomad + vault binaries)
 #   lib/init/nomad/systemd-vault.sh   (this script — unit + config + dirs)
 #   lib/init/nomad/vault-init.sh      (init + write unseal.key + unseal once)
 #   systemctl start vault             (ExecStartPost auto-unseals from file)
 #
 # The systemd unit's ExecStartPost reads /etc/vault.d/unseal.key and calls
 # `vault operator unseal`. That file is written by vault-init.sh on first
 # run; until it exists, `systemctl start vault` will leave Vault sealed
 # (ExecStartPost fails, unit goes into failed state — intentional, visible).
 #
 # Seal model:
 #   The single unseal key lives at /etc/vault.d/unseal.key (0400 root).
 #   Seal-key theft == vault theft. Factory-dev-box-acceptable tradeoff —
 #   we avoid running a second Vault to auto-unseal the first.
 #
 # Idempotency contract:
 #   - Unit file NOT rewritten when on-disk content already matches desired.
 #   - vault.hcl NOT rewritten when on-disk content matches the repo copy.
 #   - `systemctl enable` on an already-enabled unit is a no-op.
 #   - Safe to run unconditionally before every factory boot.
 #
 # Preconditions:
 #   - vault binary installed (lib/init/nomad/install.sh)
 #   - nomad/vault.hcl present in the repo (relative to this script)
 #
 # Usage:
 #   sudo lib/init/nomad/systemd-vault.sh
 #
 # Exit codes:
 #   0  success (unit+config installed + enabled, or already so)
 #   1  precondition failure (not root, no systemctl, no vault binary,
 #      missing source config)
 # =============================================================================
 set -euo pipefail
 UNIT_PATH="/etc/systemd/system/vault.service"
 VAULT_CONFIG_DIR="/etc/vault.d"
 VAULT_CONFIG_FILE="${VAULT_CONFIG_DIR}/vault.hcl"
 VAULT_DATA_DIR="/var/lib/vault/data"
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 REPO_ROOT="$(cd "${SCRIPT_DIR}/../../.." && pwd)"
 VAULT_HCL_SRC="${REPO_ROOT}/nomad/vault.hcl"
 log() { printf '[systemd-vault] %s\n' "$*"; }
 die() { printf '[systemd-vault] ERROR: %s\n' "$*" >&2; exit 1; }
 # shellcheck source=lib-systemd.sh
 . "${SCRIPT_DIR}/lib-systemd.sh"
 # ── Preconditions ────────────────────────────────────────────────────────────
 systemd_require_preconditions "$UNIT_PATH"
 VAULT_BIN="$(command -v vault 2>/dev/null || true)"
 [ -n "$VAULT_BIN" ] \
  || die "vault binary not found — run lib/init/nomad/install.sh first"
 [ -f "$VAULT_HCL_SRC" ] \
  || die "source config not found: ${VAULT_HCL_SRC}"
 # ── Desired unit content ─────────────────────────────────────────────────────
 # Adapted from HashiCorp's recommended vault.service template
 # (https://developer.hashicorp.com/vault/tutorials/getting-started-deploy/deploy)
 # for a single-node factory dev box:
 #   - User=root keeps the seal-key read path simple (unseal.key is 0400 root).
 #   - CAP_IPC_LOCK lets mlock() succeed so disable_mlock=false is honoured.
 #     Harmless when running as root; required if this is ever flipped to a
 #     dedicated `vault` user.
 #   - ExecStartPost auto-unseals on every boot using the persisted key.
 #     This is the dev-persisted-seal tradeoff — seal-key theft == vault
 #     theft, but no second Vault to babysit.
 #   - ConditionFileNotEmpty guards against starting without config — makes
 #     a missing vault.hcl visible in systemctl status, not a crash loop.
 #   - Type=notify so systemd waits for Vault's listener-ready notification
 #     before running ExecStartPost (ExecStartPost also has `sleep 2` as a
 #     belt-and-braces guard against Type=notify edge cases).
 #   - \$MAINPID is escaped so bash doesn't expand it inside this heredoc.
 #   - \$(cat ...) is escaped so the subshell runs at unit-execution time
 #     (inside bash -c), not at heredoc-expansion time here.
 read -r -d '' DESIRED_UNIT <<EOF || true
 [Unit]
 Description=HashiCorp Vault
 Documentation=https://developer.hashicorp.com/vault/docs
 Requires=network-online.target
 After=network-online.target
 ConditionFileNotEmpty=${VAULT_CONFIG_FILE}
 StartLimitIntervalSec=60
 StartLimitBurst=3
 [Service]
 Type=notify
 User=root
 Group=root
 Environment=VAULT_ADDR=http://127.0.0.1:8200
 SecureBits=keep-caps
 CapabilityBoundingSet=CAP_IPC_LOCK
 AmbientCapabilities=CAP_IPC_LOCK
 ExecStart=${VAULT_BIN} server -config=${VAULT_CONFIG_FILE}
 ExecStartPost=/bin/bash -c 'sleep 2 && ${VAULT_BIN} operator unseal \$(cat ${VAULT_CONFIG_DIR}/unseal.key)'
 ExecReload=/bin/kill --signal HUP \$MAINPID
 KillMode=process
 KillSignal=SIGINT
 Restart=on-failure
 RestartSec=5
 TimeoutStopSec=30
 LimitNOFILE=65536
 LimitMEMLOCK=infinity
 [Install]
 WantedBy=multi-user.target
 EOF
 # ── Ensure config + data dirs exist ──────────────────────────────────────────
 # /etc/vault.d is 0755 — vault.hcl is world-readable (no secrets in it);
 # the real secrets (unseal.key, root.token) get their own 0400 mode.
 # /var/lib/vault/data is 0700 — vault's on-disk state (encrypted-at-rest
 # by Vault itself, but an extra layer of "don't rely on that").
 if [ ! -d "$VAULT_CONFIG_DIR" ]; then
  log "creating ${VAULT_CONFIG_DIR}"
  install -d -m 0755 -o root -g root "$VAULT_CONFIG_DIR"
 fi
 if [ ! -d "$VAULT_DATA_DIR" ]; then
  log "creating ${VAULT_DATA_DIR}"
  install -d -m 0700 -o root -g root "$VAULT_DATA_DIR"
 fi
 # ── Install vault.hcl only if content differs ────────────────────────────────
 if [ ! -f "$VAULT_CONFIG_FILE" ] \
   || ! cmp -s "$VAULT_HCL_SRC" "$VAULT_CONFIG_FILE"; then
  log "writing config → ${VAULT_CONFIG_FILE}"
  install -m 0644 -o root -g root "$VAULT_HCL_SRC" "$VAULT_CONFIG_FILE"
 else
  log "config already up to date"
 fi
 # ── Install + reload + enable (shared with systemd-nomad.sh via lib-systemd) ─
 systemd_install_unit "$UNIT_PATH" "vault.service" "$DESIRED_UNIT"
 log "done — unit+config installed and enabled (NOT started; vault-init.sh next)"
--- a/lib/init/nomad/vault-init.sh
+++ b/lib/init/nomad/vault-init.sh
@ -1,206 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # lib/init/nomad/vault-init.sh — Idempotent Vault first-run initializer
 #
 # Part of the Nomad+Vault migration (S0.3, issue #823). Initializes Vault
 # in dev-persisted-seal mode (single unseal key on disk) and unseals once.
 # On re-run, becomes a no-op — never re-initializes or rotates the key.
 #
 # What it does (first run):
 #   1. Ensures Vault is reachable at ${VAULT_ADDR} — spawns a temporary
 #      `vault server -config=/etc/vault.d/vault.hcl` if not already up.
 #   2. Runs `vault operator init -key-shares=1 -key-threshold=1` and
 #      captures the resulting unseal key + root token.
 #   3. Writes /etc/vault.d/unseal.key   (0400 root, no trailing newline).
 #   4. Writes /etc/vault.d/root.token   (0400 root, no trailing newline).
 #   5. Unseals Vault once in the current process.
 #   6. Shuts down the temporary server if we started one (so a subsequent
 #      `systemctl start vault` doesn't conflict on port 8200).
 #
 # Idempotency contract:
 #   - /etc/vault.d/unseal.key exists AND `vault status` reports
 #     initialized=true → exit 0, no mutation, no re-init.
 #   - Initialized-but-unseal.key-missing is a hard failure (can't recover
 #     the key without the existing storage; user must restore from backup).
 #
 # Bootstrap order:
 #   lib/init/nomad/install.sh          (installs vault binary)
 #   lib/init/nomad/systemd-vault.sh    (lands unit + config + dirs; enables)
 #   lib/init/nomad/vault-init.sh       (this script — init + unseal once)
 #   systemctl start vault              (ExecStartPost auto-unseals henceforth)
 #
 # Seal model:
 #   Single unseal key persisted on disk at /etc/vault.d/unseal.key. Seal-key
 #   theft == vault theft. Factory-dev-box-acceptable tradeoff — we avoid
 #   running a second Vault to auto-unseal the first.
 #
 # Environment:
 #   VAULT_ADDR  — Vault API address (default: http://127.0.0.1:8200).
 #
 # Usage:
 #   sudo lib/init/nomad/vault-init.sh
 #
 # Exit codes:
 #   0  success (initialized + unsealed + keys persisted; or already done)
 #   1  precondition / operational failure
 # =============================================================================
 set -euo pipefail
 VAULT_CONFIG_FILE="/etc/vault.d/vault.hcl"
 UNSEAL_KEY_FILE="/etc/vault.d/unseal.key"
 ROOT_TOKEN_FILE="/etc/vault.d/root.token"
 VAULT_ADDR="${VAULT_ADDR:-http://127.0.0.1:8200}"
 export VAULT_ADDR
 # Track whether we spawned a temporary vault (for cleanup).
 spawned_pid=""
 spawned_log=""
 log() { printf '[vault-init] %s\n' "$*"; }
 die() { printf '[vault-init] ERROR: %s\n' "$*" >&2; exit 1; }
 # ── Cleanup: stop the temporary server (if we started one) on any exit ───────
 # EXIT trap fires on success AND failure AND signals — so we never leak a
 # background vault process holding port 8200 after this script returns.
 cleanup() {
  if [ -n "$spawned_pid" ] && kill -0 "$spawned_pid" 2>/dev/null; then
    log "stopping temporary vault (pid=${spawned_pid})"
    kill "$spawned_pid" 2>/dev/null || true
    wait "$spawned_pid" 2>/dev/null || true
  fi
  if [ -n "$spawned_log" ] && [ -f "$spawned_log" ]; then
    rm -f "$spawned_log"
  fi
 }
 trap cleanup EXIT
 # ── Preconditions ────────────────────────────────────────────────────────────
 if [ "$(id -u)" -ne 0 ]; then
  die "must run as root (needs to write 0400 files under /etc/vault.d)"
 fi
 for bin in vault jq; do
  command -v "$bin" >/dev/null 2>&1 \
    || die "required binary not found: ${bin}"
 done
 [ -f "$VAULT_CONFIG_FILE" ] \
  || die "config not found: ${VAULT_CONFIG_FILE} — run systemd-vault.sh first"
 # ── Helpers ──────────────────────────────────────────────────────────────────
 # vault_reachable — true iff `vault status` can reach the server.
 #   Exit codes from `vault status`:
 #     0 = reachable, initialized, unsealed
 #     2 = reachable, sealed (or uninitialized)
 #     1 = unreachable / other error
 #   We treat 0 and 2 as "reachable". `|| status=$?` avoids set -e tripping
 #   on the expected sealed-is-also-fine case.
 vault_reachable() {
  local status=0
  vault status -format=json >/dev/null 2>&1 || status=$?
  [ "$status" -eq 0 ] || [ "$status" -eq 2 ]
 }
 # vault_initialized — echoes "true" / "false" / "" (empty on parse failure
 # or unreachable vault). Always returns 0 so that `x="$(vault_initialized)"`
 # is safe under `set -euo pipefail`.
 #
 # Key subtlety: `vault status` exits 2 when Vault is sealed OR uninitialized
 # — the exact state we need to *observe* on first run. Without the
 # `|| true` guard, pipefail + set -e inside a standalone assignment would
 # propagate that exit 2 to the outer script and abort before we ever call
 # `vault operator init`. We capture `vault status`'s output to a variable
 # first (pipefail-safe), then feed it to jq separately.
 vault_initialized() {
  local out=""
  out="$(vault status -format=json 2>/dev/null || true)"
  [ -n "$out" ] || { printf ''; return 0; }
  printf '%s' "$out" | jq -r '.initialized' 2>/dev/null || printf ''
 }
 # write_secret_file PATH CONTENT
 #   Write CONTENT to PATH atomically with 0400 root:root and no trailing
 #   newline. mktemp+install keeps perms tight for the whole lifetime of
 #   the file on disk — no 0644-then-chmod window.
 write_secret_file() {
  local path="$1" content="$2"
  local tmp
  tmp="$(mktemp)"
  printf '%s' "$content" > "$tmp"
  install -m 0400 -o root -g root "$tmp" "$path"
  rm -f "$tmp"
 }
 # ── Ensure vault is reachable ────────────────────────────────────────────────
 if ! vault_reachable; then
  log "vault not reachable at ${VAULT_ADDR} — starting temporary server"
  spawned_log="$(mktemp)"
  vault server -config="$VAULT_CONFIG_FILE" >"$spawned_log" 2>&1 &
  spawned_pid=$!
  # Poll for readiness. Vault's API listener comes up before notify-ready
  # in Type=notify mode, but well inside a few seconds even on cold boots.
  ready=0
  for _ in 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15; do
    if vault_reachable; then
      ready=1
      break
    fi
    sleep 1
  done
  if [ "$ready" -ne 1 ]; then
    log "vault did not become reachable within 15s — server log follows:"
    if [ -f "$spawned_log" ]; then
      sed 's/^/[vault-server] /' "$spawned_log" >&2 || true
    fi
    die "failed to start temporary vault server"
  fi
  log "temporary vault ready (pid=${spawned_pid})"
 fi
 # ── Idempotency gate ─────────────────────────────────────────────────────────
 initialized="$(vault_initialized)"
 if [ "$initialized" = "true" ] && [ -f "$UNSEAL_KEY_FILE" ]; then
  log "vault already initialized and unseal.key present — no-op"
  exit 0
 fi
 if [ "$initialized" = "true" ] && [ ! -f "$UNSEAL_KEY_FILE" ]; then
  die "vault is initialized but ${UNSEAL_KEY_FILE} is missing — cannot recover the unseal key; restore from backup or wipe ${VAULT_CONFIG_FILE%/*}/data and re-run"
 fi
 if [ "$initialized" != "false" ]; then
  die "unexpected initialized state: '${initialized}' (expected 'true' or 'false')"
 fi
 # ── Initialize ───────────────────────────────────────────────────────────────
 log "initializing vault (key-shares=1, key-threshold=1)"
 init_json="$(vault operator init \
  -key-shares=1 \
  -key-threshold=1 \
  -format=json)" \
  || die "vault operator init failed"
 unseal_key="$(printf '%s' "$init_json" | jq -er '.unseal_keys_b64[0]')" \
  || die "failed to extract unseal key from init response"
 root_token="$(printf '%s' "$init_json" | jq -er '.root_token')" \
  || die "failed to extract root token from init response"
 # Best-effort scrub of init_json from the env (the captured key+token still
 # sit in the local vars above — there's no clean way to wipe bash memory).
 unset init_json
 # ── Persist keys ─────────────────────────────────────────────────────────────
 log "writing ${UNSEAL_KEY_FILE} (0400 root)"
 write_secret_file "$UNSEAL_KEY_FILE" "$unseal_key"
 log "writing ${ROOT_TOKEN_FILE} (0400 root)"
 write_secret_file "$ROOT_TOKEN_FILE" "$root_token"
 # ── Unseal in the current process ────────────────────────────────────────────
 log "unsealing vault"
 vault operator unseal "$unseal_key" >/dev/null \
  || die "vault operator unseal failed"
 log "done — vault initialized + unsealed + keys persisted"
--- a/lib/mirrors.sh
+++ b/lib/mirrors.sh
@ -1,10 +1,8 @@
 #!/usr/bin/env bash
-# mirrors.sh — Mirror helpers: push to remotes + register pull mirrors via API.
+# mirrors.sh — Push primary branch + tags to configured mirror remotes.
 #
 # Usage: source lib/mirrors.sh; mirror_push
 #        source lib/mirrors.sh; mirror_pull_register <clone_url> <owner> <repo_name> [interval]
 # Requires: PROJECT_REPO_ROOT, PRIMARY_BRANCH, MIRROR_* vars from load-project.sh
 #           FORGE_API_BASE, FORGE_TOKEN for pull-mirror registration
 # shellcheck disable=SC2154  # globals set by load-project.sh / calling script
@ -39,73 +37,3 @@ mirror_push() {
    log "mirror: pushed to ${name} (pid $!)"
  done
 }
 # ---------------------------------------------------------------------------
 # mirror_pull_register — register a Forgejo pull mirror via the /repos/migrate API.
 #
 # Creates a new repo as a pull mirror of an external source.  Works against
 # empty target repos (the repo is created by the API call itself).
 #
 # Usage:
 #   mirror_pull_register <clone_url> <owner> <repo_name> [interval]
 #
 # Args:
 #   clone_url  — HTTPS URL of the source repo (e.g. https://codeberg.org/johba/disinto.git)
 #   owner      — Forgejo org or user that will own the mirror repo
 #   repo_name  — name of the new mirror repo on Forgejo
 #   interval   — sync interval (default: "8h0m0s"; Forgejo duration format)
 #
 # Requires:
 #   FORGE_API_BASE, FORGE_TOKEN (from env.sh)
 #
 # Returns 0 on success, 1 on failure.  Prints the new repo JSON to stdout.
 # ---------------------------------------------------------------------------
 mirror_pull_register() {
  local clone_url="$1"
  local owner="$2"
  local repo_name="$3"
  local interval="${4:-8h0m0s}"
  if [ -z "${FORGE_API_BASE:-}" ] || [ -z "${FORGE_TOKEN:-}" ]; then
    echo "ERROR: FORGE_API_BASE and FORGE_TOKEN must be set" >&2
    return 1
  fi
  if [ -z "$clone_url" ] || [ -z "$owner" ] || [ -z "$repo_name" ]; then
    echo "Usage: mirror_pull_register <clone_url> <owner> <repo_name> [interval]" >&2
    return 1
  fi
  local payload
  payload=$(jq -n \
    --arg clone_addr "$clone_url" \
    --arg repo_name  "$repo_name" \
    --arg repo_owner "$owner" \
    --arg interval   "$interval" \
    '{
      clone_addr:      $clone_addr,
      repo_name:       $repo_name,
      repo_owner:      $repo_owner,
      mirror:          true,
      mirror_interval: $interval,
      service:         "git"
    }')
  local http_code body
  body=$(curl -s -w "\n%{http_code}" -X POST \
    -H "Authorization: token ${FORGE_TOKEN}" \
    -H "Content-Type: application/json" \
    "${FORGE_API_BASE}/repos/migrate" \
    -d "$payload")
  http_code=$(printf '%s' "$body" | tail -n1)
  body=$(printf '%s' "$body" | sed '$d')
  if [ "$http_code" -ge 200 ] && [ "$http_code" -lt 300 ]; then
    printf '%s\n' "$body"
    return 0
  else
    echo "ERROR: mirror_pull_register failed (HTTP ${http_code}): ${body}" >&2
    return 1
  fi
 }
--- a/lib/release.sh
+++ b/lib/release.sh
@ -18,8 +18,8 @@
 # =============================================================================
 set -euo pipefail
-# Source action-vault.sh for _vault_log helper
+# Source vault.sh for _vault_log helper
-source "${FACTORY_ROOT}/lib/action-vault.sh"
+source "${FACTORY_ROOT}/lib/vault.sh"
 # Assert required globals are set before using this module.
 _assert_release_globals() {
--- a/lib/sprint-filer.sh
+++ b/lib/sprint-filer.sh
@ -1,585 +0,0 @@
 #!/usr/bin/env bash
 # =============================================================================
 # sprint-filer.sh — Parse merged sprint PRs and file sub-issues via filer-bot
 #
 # Invoked by the ops-filer Woodpecker pipeline after a sprint PR merges on the
 # ops repo main branch.  Parses each sprints/*.md file for a structured
 # ## Sub-issues block (filer:begin/end markers), then creates idempotent
 # Forgejo issues on the project repo using FORGE_FILER_TOKEN.
 #
 # Permission model (#764):
 #   filer-bot has issues:write on the project repo.
 #   architect-bot is read-only on the project repo.
 #
 # Usage:
 #   sprint-filer.sh <sprint-file.md>          — file sub-issues from one sprint
 #   sprint-filer.sh --all <sprints-dir>       — scan all sprint files in dir
 #
 # Environment:
 #   FORGE_FILER_TOKEN   — filer-bot API token (issues:write on project repo)
 #   FORGE_API           — project repo API base (e.g. http://forgejo:3000/api/v1/repos/org/repo)
 #   FORGE_API_BASE      — API base URL (e.g. http://forgejo:3000/api/v1)
 # =============================================================================
 set -euo pipefail
 SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
 # Source env.sh only if not already loaded (allows standalone + sourced use)
 if [ -z "${FACTORY_ROOT:-}" ]; then
  FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
  # shellcheck source=env.sh
  source "$SCRIPT_DIR/env.sh"
 fi
 # ── Logging ──────────────────────────────────────────────────────────────
 LOG_AGENT="${LOG_AGENT:-filer}"
 filer_log() {
  printf '[%s] %s: %s\n' "$(date -u '+%Y-%m-%dT%H:%M:%SZ')" "$LOG_AGENT" "$*" >&2
 }
 # ── Validate required environment ────────────────────────────────────────
 : "${FORGE_FILER_TOKEN:?sprint-filer.sh requires FORGE_FILER_TOKEN}"
 : "${FORGE_API:?sprint-filer.sh requires FORGE_API}"
 # ── Paginated Forgejo API fetch ──────────────────────────────────────────
 # Reuses forge_api_all from lib/env.sh with FORGE_FILER_TOKEN.
 # Args: api_path (e.g. /issues?state=all&type=issues)
 # Output: merged JSON array to stdout
 filer_api_all() { forge_api_all "$1" "$FORGE_FILER_TOKEN"; }
 # ── Parse sub-issues block from a sprint markdown file ───────────────────
 # Extracts the YAML-in-markdown between <!-- filer:begin --> and <!-- filer:end -->
 # Args: sprint_file_path
 # Output: the raw sub-issues block (YAML lines) to stdout
 # Returns: 0 if block found, 1 if not found or malformed
 parse_subissues_block() {
  local sprint_file="$1"
  if [ ! -f "$sprint_file" ]; then
    filer_log "ERROR: sprint file not found: ${sprint_file}"
    return 1
  fi
  local in_block=false
  local block=""
  local found=false
  while IFS= read -r line; do
    if [[ "$line" == *"<!-- filer:begin -->"* ]]; then
      in_block=true
      found=true
      continue
    fi
    if [[ "$line" == *"<!-- filer:end -->"* ]]; then
      in_block=false
      continue
    fi
    if [ "$in_block" = true ]; then
      block+="${line}"$'\n'
    fi
  done < "$sprint_file"
  if [ "$found" = false ]; then
    filer_log "No filer:begin/end block found in ${sprint_file}"
    return 1
  fi
  if [ "$in_block" = true ]; then
    filer_log "ERROR: malformed sub-issues block in ${sprint_file} — filer:begin without filer:end"
    return 1
  fi
  if [ -z "$block" ]; then
    filer_log "WARNING: empty sub-issues block in ${sprint_file}"
    return 1
  fi
  printf '%s' "$block"
 }
 # ── Extract vision issue number from sprint file ─────────────────────────
 # Looks for "#N" references specifically in the "## Vision issues" section
 # to avoid picking up cross-links or related-issue mentions earlier in the file.
 # Falls back to first #N in the file if no "## Vision issues" section found.
 # Args: sprint_file_path
 # Output: first vision issue number found
 extract_vision_issue() {
  local sprint_file="$1"
  # Try to extract from "## Vision issues" section first
  local in_section=false
  local result=""
  while IFS= read -r line; do
    if [[ "$line" =~ ^##[[:space:]]+Vision[[:space:]]+issues ]]; then
      in_section=true
      continue
    fi
    # Stop at next heading
    if [ "$in_section" = true ] && [[ "$line" =~ ^## ]]; then
      break
    fi
    if [ "$in_section" = true ]; then
      result=$(printf '%s' "$line" | grep -oE '#[0-9]+' | head -1 | tr -d '#')
      if [ -n "$result" ]; then
        printf '%s' "$result"
        return 0
      fi
    fi
  done < "$sprint_file"
  # Fallback: first #N in the entire file
  grep -oE '#[0-9]+' "$sprint_file" | head -1 | tr -d '#'
 }
 # ── Extract sprint slug from file path ───────────────────────────────────
 # Args: sprint_file_path
 # Output: slug (filename without .md)
 extract_sprint_slug() {
  local sprint_file="$1"
  basename "$sprint_file" .md
 }
 # ── Parse individual sub-issue entries from the block ────────────────────
 # The block is a simple YAML-like format:
 #   - id: foo
 #     title: "..."
 #     labels: [backlog, priority]
 #     depends_on: [bar]
 #     body: |
 #       multi-line body
 #
 # Args: raw_block (via stdin)
 # Output: JSON array of sub-issue objects
 parse_subissue_entries() {
  local block
  block=$(cat)
  # Use awk to parse the YAML-like structure into JSON
  printf '%s' "$block" | awk '
  BEGIN {
    printf "["
    first = 1
    inbody = 0
    id = ""; title = ""; labels = ""; depends = ""; body = ""
  }
  function flush_entry() {
    if (id == "") return
    if (!first) printf ","
    first = 0
    # Escape JSON special characters in body
    gsub(/\\/, "\\\\", body)
    gsub(/"/, "\\\"", body)
    gsub(/\t/, "\\t", body)
    # Replace newlines with \n for JSON
    gsub(/\n/, "\\n", body)
    # Remove trailing \n
    sub(/\\n$/, "", body)
    # Clean up title (remove surrounding quotes)
    gsub(/^"/, "", title)
    gsub(/"$/, "", title)
    printf "{\"id\":\"%s\",\"title\":\"%s\",\"labels\":%s,\"depends_on\":%s,\"body\":\"%s\"}", id, title, labels, depends, body
    id = ""; title = ""; labels = "[]"; depends = "[]"; body = ""
    inbody = 0
  }
  /^- id:/ {
    flush_entry()
    sub(/^- id: */, "")
    id = $0
    labels = "[]"
    depends = "[]"
    next
  }
  /^  title:/ {
    sub(/^  title: */, "")
    title = $0
    # Remove surrounding quotes
    gsub(/^"/, "", title)
    gsub(/"$/, "", title)
    next
  }
  /^  labels:/ {
    sub(/^  labels: */, "")
    # Convert [a, b] to JSON array ["a","b"]
    gsub(/\[/, "", $0)
    gsub(/\]/, "", $0)
    n = split($0, arr, /, */)
    labels = "["
    for (i = 1; i <= n; i++) {
      gsub(/^ */, "", arr[i])
      gsub(/ *$/, "", arr[i])
      if (arr[i] != "") {
        if (i > 1) labels = labels ","
        labels = labels "\"" arr[i] "\""
      }
    }
    labels = labels "]"
    next
  }
  /^  depends_on:/ {
    sub(/^  depends_on: */, "")
    gsub(/\[/, "", $0)
    gsub(/\]/, "", $0)
    n = split($0, arr, /, */)
    depends = "["
    for (i = 1; i <= n; i++) {
      gsub(/^ */, "", arr[i])
      gsub(/ *$/, "", arr[i])
      if (arr[i] != "") {
        if (i > 1) depends = depends ","
        depends = depends "\"" arr[i] "\""
      }
    }
    depends = depends "]"
    next
  }
  /^  body: *\|/ {
    inbody = 1
    body = ""
    next
  }
  inbody && /^    / {
    sub(/^    /, "")
    body = body $0 "\n"
    next
  }
  inbody && !/^    / && !/^$/ {
    inbody = 0
    # This line starts a new field or entry — re-process it
    # (awk does not support re-scanning, so handle common cases)
    if ($0 ~ /^- id:/) {
      flush_entry()
      sub(/^- id: */, "")
      id = $0
      labels = "[]"
      depends = "[]"
    }
  }
  END {
    flush_entry()
    printf "]"
  }
  '
 }
 # ── Check if sub-issue already exists (idempotency) ─────────────────────
 # Searches for the decomposed-from marker in existing issues.
 # Args: vision_issue_number sprint_slug subissue_id
 # Returns: 0 if already exists, 1 if not
 subissue_exists() {
  local vision_issue="$1"
  local sprint_slug="$2"
  local subissue_id="$3"
  local marker="<!-- decomposed-from: #${vision_issue}, sprint: ${sprint_slug}, id: ${subissue_id} -->"
  # Search all issues (paginated) for the exact marker
  local issues_json
  issues_json=$(filer_api_all "/issues?state=all&type=issues")
  if printf '%s' "$issues_json" | jq -e --arg marker "$marker" \
    '[.[] | select(.body // "" | contains($marker))] | length > 0' >/dev/null 2>&1; then
    return 0  # Already exists
  fi
  return 1  # Does not exist
 }
 # ── Resolve label names to IDs ───────────────────────────────────────────
 # Args: label_names_json (JSON array of strings)
 # Output: JSON array of label IDs
 resolve_label_ids() {
  local label_names_json="$1"
  # Fetch all labels from project repo
  local all_labels
  all_labels=$(curl -sf -H "Authorization: token ${FORGE_FILER_TOKEN}" \
    "${FORGE_API}/labels" 2>/dev/null) || all_labels="[]"
  # Map names to IDs
  printf '%s' "$label_names_json" | jq -r '.[]' | while IFS= read -r label_name; do
    [ -z "$label_name" ] && continue
    printf '%s' "$all_labels" | jq -r --arg name "$label_name" \
      '.[] | select(.name == $name) | .id' 2>/dev/null
  done | jq -Rs 'split("\n") | map(select(. != "") | tonumber)'
 }
 # ── Add in-progress label to vision issue ────────────────────────────────
 # Args: vision_issue_number
 add_inprogress_label() {
  local issue_num="$1"
  local labels_json
  labels_json=$(curl -sf -H "Authorization: token ${FORGE_FILER_TOKEN}" \
    "${FORGE_API}/labels" 2>/dev/null) || return 1
  local label_id
  label_id=$(printf '%s' "$labels_json" | jq -r '.[] | select(.name == "in-progress") | .id' 2>/dev/null) || true
  if [ -z "$label_id" ]; then
    filer_log "WARNING: in-progress label not found"
    return 1
  fi
  if curl -sf -X POST \
    -H "Authorization: token ${FORGE_FILER_TOKEN}" \
    -H "Content-Type: application/json" \
    "${FORGE_API}/issues/${issue_num}/labels" \
    -d "{\"labels\": [${label_id}]}" >/dev/null 2>&1; then
    filer_log "Added in-progress label to vision issue #${issue_num}"
    return 0
  else
    filer_log "WARNING: failed to add in-progress label to vision issue #${issue_num}"
    return 1
  fi
 }
 # ── File sub-issues from a sprint file ───────────────────────────────────
 # This is the main entry point. Parses the sprint file, extracts sub-issues,
 # and creates them idempotently via the Forgejo API.
 # Args: sprint_file_path
 # Returns: 0 on success, 1 on any error (fail-fast)
 file_subissues() {
  local sprint_file="$1"
  filer_log "Processing sprint file: ${sprint_file}"
  # Extract metadata
  local vision_issue sprint_slug
  vision_issue=$(extract_vision_issue "$sprint_file")
  sprint_slug=$(extract_sprint_slug "$sprint_file")
  if [ -z "$vision_issue" ]; then
    filer_log "ERROR: could not extract vision issue number from ${sprint_file}"
    return 1
  fi
  filer_log "Vision issue: #${vision_issue}, sprint slug: ${sprint_slug}"
  # Parse the sub-issues block
  local raw_block
  raw_block=$(parse_subissues_block "$sprint_file") || return 1
  # Parse individual entries
  local entries_json
  entries_json=$(printf '%s' "$raw_block" | parse_subissue_entries)
  # Validate parsing produced valid JSON
  if ! printf '%s' "$entries_json" | jq empty 2>/dev/null; then
    filer_log "ERROR: failed to parse sub-issues block as valid JSON in ${sprint_file}"
    return 1
  fi
  local entry_count
  entry_count=$(printf '%s' "$entries_json" | jq 'length')
  if [ "$entry_count" -eq 0 ]; then
    filer_log "WARNING: no sub-issue entries found in ${sprint_file}"
    return 1
  fi
  filer_log "Found ${entry_count} sub-issue(s) to file"
  # File each sub-issue (fail-fast on first error)
  local filed_count=0
  local i=0
  while [ "$i" -lt "$entry_count" ]; do
    local entry
    entry=$(printf '%s' "$entries_json" | jq ".[$i]")
    local subissue_id subissue_title subissue_body labels_json
    subissue_id=$(printf '%s' "$entry" | jq -r '.id')
    subissue_title=$(printf '%s' "$entry" | jq -r '.title')
    subissue_body=$(printf '%s' "$entry" | jq -r '.body')
    labels_json=$(printf '%s' "$entry" | jq -c '.labels')
    if [ -z "$subissue_id" ] || [ "$subissue_id" = "null" ]; then
      filer_log "ERROR: sub-issue entry at index ${i} has no id — aborting"
      return 1
    fi
    if [ -z "$subissue_title" ] || [ "$subissue_title" = "null" ]; then
      filer_log "ERROR: sub-issue '${subissue_id}' has no title — aborting"
      return 1
    fi
    # Idempotency check
    if subissue_exists "$vision_issue" "$sprint_slug" "$subissue_id"; then
      filer_log "Sub-issue '${subissue_id}' already exists — skipping"
      i=$((i + 1))
      continue
    fi
    # Append decomposed-from marker to body
    local marker="<!-- decomposed-from: #${vision_issue}, sprint: ${sprint_slug}, id: ${subissue_id} -->"
    local full_body="${subissue_body}
 ${marker}"
    # Resolve label names to IDs
    local label_ids
    label_ids=$(resolve_label_ids "$labels_json")
    # Build issue payload using jq for safe JSON construction
    local payload
    payload=$(jq -n \
      --arg title "$subissue_title" \
      --arg body "$full_body" \
      --argjson labels "$label_ids" \
      '{title: $title, body: $body, labels: $labels}')
    # Create the issue
    local response
    response=$(curl -sf -X POST \
      -H "Authorization: token ${FORGE_FILER_TOKEN}" \
      -H "Content-Type: application/json" \
      "${FORGE_API}/issues" \
      -d "$payload" 2>/dev/null) || {
      filer_log "ERROR: failed to create sub-issue '${subissue_id}' — aborting (${filed_count}/${entry_count} filed so far)"
      return 1
    }
    local new_issue_num
    new_issue_num=$(printf '%s' "$response" | jq -r '.number // empty')
    filer_log "Filed sub-issue '${subissue_id}' as #${new_issue_num}: ${subissue_title}"
    filed_count=$((filed_count + 1))
    i=$((i + 1))
  done
  # Add in-progress label to the vision issue
  add_inprogress_label "$vision_issue" || true
  filer_log "Successfully filed ${filed_count}/${entry_count} sub-issue(s) for sprint ${sprint_slug}"
  return 0
 }
 # ── Vision lifecycle: close completed vision issues ──────────────────────
 # Checks open vision issues and closes any whose sub-issues are all closed.
 # Uses the decomposed-from marker to find sub-issues.
 check_and_close_completed_visions() {
  filer_log "Checking for vision issues with all sub-issues complete..."
  local vision_issues_json
  vision_issues_json=$(filer_api_all "/issues?labels=vision&state=open")
  if [ "$vision_issues_json" = "[]" ] || [ "$vision_issues_json" = "null" ]; then
    filer_log "No open vision issues found"
    return 0
  fi
  local all_issues
  all_issues=$(filer_api_all "/issues?state=all&type=issues")
  local vision_nums
  vision_nums=$(printf '%s' "$vision_issues_json" | jq -r '.[].number' 2>/dev/null) || return 0
  local closed_count=0
  while IFS= read -r vid; do
    [ -z "$vid" ] && continue
    # Find sub-issues with decomposed-from marker for this vision
    local sub_issues
    sub_issues=$(printf '%s' "$all_issues" | jq --arg vid "$vid" \
      '[.[] | select(.body // "" | contains("<!-- decomposed-from: #" + $vid))]')
    local sub_count
    sub_count=$(printf '%s' "$sub_issues" | jq 'length')
    # No sub-issues means not ready to close
    [ "$sub_count" -eq 0 ] && continue
    # Check if all are closed
    local open_count
    open_count=$(printf '%s' "$sub_issues" | jq '[.[] | select(.state != "closed")] | length')
    if [ "$open_count" -gt 0 ]; then
      continue
    fi
    # All sub-issues closed — close the vision issue
    filer_log "All ${sub_count} sub-issues for vision #${vid} are closed — closing vision"
    local comment_body
    comment_body="## Vision Issue Completed
 All sub-issues have been implemented and merged. This vision issue is now closed.
 ---
 *Automated closure by filer-bot · $(date -u '+%Y-%m-%d %H:%M UTC')*"
    local comment_payload
    comment_payload=$(jq -n --arg body "$comment_body" '{body: $body}')
    curl -sf -X POST \
      -H "Authorization: token ${FORGE_FILER_TOKEN}" \
      -H "Content-Type: application/json" \
      "${FORGE_API}/issues/${vid}/comments" \
      -d "$comment_payload" >/dev/null 2>&1 || true
    curl -sf -X PATCH \
      -H "Authorization: token ${FORGE_FILER_TOKEN}" \
      -H "Content-Type: application/json" \
      "${FORGE_API}/issues/${vid}" \
      -d '{"state":"closed"}' >/dev/null 2>&1 || true
    closed_count=$((closed_count + 1))
  done <<< "$vision_nums"
  if [ "$closed_count" -gt 0 ]; then
    filer_log "Closed ${closed_count} vision issue(s)"
  fi
 }
 # ── Main ─────────────────────────────────────────────────────────────────
 main() {
  if [ "${1:-}" = "--all" ]; then
    local sprints_dir="${2:?Usage: sprint-filer.sh --all <sprints-dir>}"
    local exit_code=0
    for sprint_file in "${sprints_dir}"/*.md; do
      [ -f "$sprint_file" ] || continue
      # Only process files with filer:begin markers
      if ! grep -q '<!-- filer:begin -->' "$sprint_file"; then
        continue
      fi
      if ! file_subissues "$sprint_file"; then
        filer_log "ERROR: failed to process ${sprint_file}"
        exit_code=1
      fi
    done
    # Run vision lifecycle check after filing
    check_and_close_completed_visions || true
    return "$exit_code"
  elif [ -n "${1:-}" ]; then
    file_subissues "$1"
    # Run vision lifecycle check after filing
    check_and_close_completed_visions || true
  else
    echo "Usage: sprint-filer.sh <sprint-file.md>" >&2
    echo "       sprint-filer.sh --all <sprints-dir>" >&2
    return 1
  fi
 }
 # Run main only when executed directly (not when sourced for testing)
 if [[ "${BASH_SOURCE[0]}" == "${0}" ]]; then
  main "$@"
 fi
--- a/lib/action-vault.sh
+++ b/lib/action-vault.sh
@ -1,9 +1,9 @@
 #!/usr/bin/env bash
-# action-vault.sh — Helper for agents to create vault PRs on ops repo
+# vault.sh — Helper for agents to create vault PRs on ops repo
 #
 # Source after lib/env.sh:
 #   source "$(dirname "$0")/../lib/env.sh"
-#   source "$(dirname "$0")/lib/action-vault.sh"
+#   source "$(dirname "$0")/lib/vault.sh"
 #
 # Required globals: FORGE_TOKEN, FORGE_URL, FORGE_REPO, FORGE_OPS_REPO
 # Optional: OPS_REPO_ROOT (local path for ops repo)
@ -12,7 +12,7 @@
 #   vault_request <action_id> <toml_content>  — Create vault PR, return PR number
 #
 # The function:
-# 1. Validates TOML content using validate_vault_action() from action-vault/vault-env.sh
+# 1. Validates TOML content using validate_vault_action() from vault/vault-env.sh
 # 2. Creates a branch on the ops repo: vault/<action-id>
 # 3. Writes TOML to vault/actions/<action-id>.toml on that branch
 # 4. Creates PR targeting main with title "vault: <action-id>"
@ -133,7 +133,7 @@ vault_request() {
  printf '%s' "$toml_content" > "$tmp_toml"
  # Source vault-env.sh for validate_vault_action
-  local vault_env="${FACTORY_ROOT:-$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)}/action-vault/vault-env.sh"
+  local vault_env="${FACTORY_ROOT:-$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)}/vault/vault-env.sh"
  if [ ! -f "$vault_env" ]; then
    echo "ERROR: vault-env.sh not found at $vault_env" >&2
    return 1
@ -161,7 +161,7 @@ vault_request() {
  ops_api="$(_vault_ops_api)"
  # Classify the action to determine if PR bypass is allowed
-  local classify_script="${FACTORY_ROOT:-$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)}/action-vault/classify.sh"
+  local classify_script="${FACTORY_ROOT:-$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)}/vault/classify.sh"
  local vault_tier
  vault_tier=$("$classify_script" "${VAULT_ACTION_FORMULA:-}" "${VAULT_BLAST_RADIUS_OVERRIDE:-}") || {
    # Classification failed, default to high tier (require PR)
--- a/nomad/client.hcl
+++ b/nomad/client.hcl
@ -1,88 +0,0 @@
 # =============================================================================
 # nomad/client.hcl — Docker driver + host_volume declarations
 #
 # Part of the Nomad+Vault migration (S0.2, issue #822). Deployed to
 # /etc/nomad.d/client.hcl on the factory dev box alongside server.hcl.
 #
 # This file owns: Docker driver plugin config + host_volume pre-wiring.
 # server.hcl owns: agent role, bind, ports, data_dir.
 #
 # NOTE: Nomad merges every *.hcl under -config=/etc/nomad.d, so declaring
 # a second `client { ... }` block here augments (not replaces) the one in
 # server.hcl. On a single-node setup this file could be inlined into
 # server.hcl — the split is for readability, not semantics.
 #
 # host_volume declarations let Nomad jobspecs mount factory state by name
 # (volume = "forgejo-data", etc.) without coupling host paths into jobspec
 # HCL. Host paths under /srv/disinto/* are created out-of-band by the
 # orchestrator (S0.4) before any job references them.
 # =============================================================================
 client {
  # forgejo git server data (repos, avatars, attachments).
  host_volume "forgejo-data" {
    path      = "/srv/disinto/forgejo-data"
    read_only = false
  }
  # woodpecker CI data (pipeline artifacts, sqlite db).
  host_volume "woodpecker-data" {
    path      = "/srv/disinto/woodpecker-data"
    read_only = false
  }
  # agent runtime data (claude config, logs, phase files).
  host_volume "agent-data" {
    path      = "/srv/disinto/agent-data"
    read_only = false
  }
  # per-project git clones and worktrees.
  host_volume "project-repos" {
    path      = "/srv/disinto/project-repos"
    read_only = false
  }
  # caddy config + ACME state.
  host_volume "caddy-data" {
    path      = "/srv/disinto/caddy-data"
    read_only = false
  }
  # disinto chat transcripts + attachments.
  host_volume "chat-history" {
    path      = "/srv/disinto/chat-history"
    read_only = false
  }
  # ops repo clone (vault actions, sprint artifacts, knowledge).
  host_volume "ops-repo" {
    path      = "/srv/disinto/ops-repo"
    read_only = false
  }
 }
 # Docker task driver. `volumes.enabled = true` is required so jobspecs
 # can mount host_volume declarations defined above. `allow_privileged`
 # stays false — no factory workload needs privileged containers today,
 # and flipping it is an audit-worthy change.
 plugin "docker" {
  config {
    allow_privileged = false
    volumes {
      enabled = true
    }
    # Leave images behind when jobs stop, so short job churn doesn't thrash
    # the image cache. Factory disk is not constrained; `docker system prune`
    # is the escape hatch.
    gc {
      image       = false
      container   = true
      dangling_containers {
        enabled = true
      }
    }
  }
 }
--- a/nomad/server.hcl
+++ b/nomad/server.hcl
@ -1,53 +0,0 @@
 # =============================================================================
 # nomad/server.hcl — Single-node combined server+client configuration
 #
 # Part of the Nomad+Vault migration (S0.2, issue #822). Deployed to
 # /etc/nomad.d/server.hcl on the factory dev box alongside client.hcl.
 #
 # This file owns: agent role, ports, bind, data directory.
 # client.hcl owns: Docker driver plugin config + host_volume declarations.
 #
 # NOTE: On single-node setups these two files could be merged into one
 # (Nomad auto-merges every *.hcl under -config=/etc/nomad.d). The split is
 # purely for readability — role/bind/port vs. plugin/volume wiring.
 #
 # This is a factory dev-box baseline — TLS, ACLs, gossip encryption, and
 # consul/vault integration are deliberately absent and land in later steps.
 # =============================================================================
 data_dir  = "/var/lib/nomad"
 bind_addr = "127.0.0.1"
 log_level = "INFO"
 # All Nomad agent traffic stays on localhost — the factory box does not
 # federate with peers. Ports are the Nomad defaults, pinned here so that
 # future changes to these numbers are a visible diff.
 ports {
  http = 4646
  rpc  = 4647
  serf = 4648
 }
 # Single-node combined mode: this agent is both the only server and the
 # only client. bootstrap_expect=1 makes the server quorum-of-one.
 server {
  enabled          = true
  bootstrap_expect = 1
 }
 client {
  enabled = true
 }
 # Advertise localhost to self to avoid surprises if the default IP
 # autodetection picks a transient interface (e.g. docker0, wg0).
 advertise {
  http = "127.0.0.1"
  rpc  = "127.0.0.1"
  serf = "127.0.0.1"
 }
 # UI on by default — same bind as http, no TLS (localhost only).
 ui {
  enabled = true
 }
--- a/nomad/vault.hcl
+++ b/nomad/vault.hcl
@ -1,41 +0,0 @@
 # =============================================================================
 # nomad/vault.hcl — Single-node Vault configuration (dev-persisted seal)
 #
 # Part of the Nomad+Vault migration (S0.3, issue #823). Deployed to
 # /etc/vault.d/vault.hcl on the factory dev box.
 #
 # Seal model: the single unseal key lives on disk at /etc/vault.d/unseal.key
 # (0400 root) and is read by systemd ExecStartPost on every boot. This is
 # the factory-dev-box-acceptable tradeoff — seal-key theft equals vault
 # theft, but we avoid running a second Vault to auto-unseal the first.
 #
 # This is a factory dev-box baseline — TLS, HA, Raft storage, and audit
 # devices are deliberately absent. Storage is the `file` backend (single
 # node only). Listener is localhost-only, so no external TLS is needed.
 # =============================================================================
 # File storage backend — single-node only, no HA, no raft. State lives in
 # /var/lib/vault/data which is created (root:root 0700) by
 # lib/init/nomad/systemd-vault.sh before the unit starts.
 storage "file" {
  path = "/var/lib/vault/data"
 }
 # Localhost-only listener. TLS is disabled because all callers are on the
 # same box — flipping this to tls_disable=false is an audit-worthy change
 # paired with cert provisioning.
 listener "tcp" {
  address     = "127.0.0.1:8200"
  tls_disable = true
 }
 # mlock prevents Vault's in-memory secrets from being swapped to disk. We
 # keep it enabled; the systemd unit grants CAP_IPC_LOCK so mlock() succeeds.
 disable_mlock = false
 # Advertised API address — used by Vault clients on this host. Matches
 # the listener above.
 api_addr = "http://127.0.0.1:8200"
 # UI on by default — same bind as listener, no TLS (localhost only).
 ui = true
--- a/planner/AGENTS.md
+++ b/planner/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Planner Agent
 **Role**: Strategic planning using a Prerequisite Tree (Theory of Constraints),
@ -34,9 +34,7 @@ will then sections) and marks the prerequisite as blocked-on-vault in the tree.
 Deduplication: checks pending/ + approved/ + fired/ before creating.
 Phase 4 (journal-and-memory): write updated prerequisite tree + daily journal
 entry (committed to ops repo) and update `$OPS_REPO_ROOT/knowledge/planner-memory.md`.
-Phase 5 (commit-ops): commit all ops repo changes to a `planner/run-YYYY-MM-DD`
+Phase 5 (commit-ops): commit all ops repo changes, push directly.
 branch, then create a PR and walk it to merge via review-bot (`pr_create` →
 `pr_walk_to_merge`), mirroring the architect's ops flow. No direct push to main.
 AGENTS.md maintenance is handled by the Gardener.
 **Artifacts use `$OPS_REPO_ROOT`**: All planner artifacts (journal,
@ -57,7 +55,7 @@ nervous system component, not work.
  creates tmux session, injects formula prompt, monitors phase file, handles crash recovery, cleans up
 - `formulas/run-planner.toml` — Execution spec: six steps (preflight,
  prediction-triage, update-prerequisite-tree, file-at-constraints,
-  journal-and-memory, commit-ops-changes) with `needs` dependencies. Claude
+  journal-and-memory, commit-and-pr) with `needs` dependencies. Claude
  executes all steps in a single interactive session with tool access
 - `formulas/groom-backlog.toml` — Grooming formula for backlog triage and
  grooming. (Note: the planner no longer dispatches breakdown mode — complex
--- a/planner/planner-run.sh
+++ b/planner/planner-run.sh
@ -10,9 +10,7 @@
 #   2. Load formula (formulas/run-planner.toml)
 #   3. Context: VISION.md, AGENTS.md, ops:RESOURCES.md, structural graph,
 #      planner memory, journal entries
-#   4. Create ops branch planner/run-YYYY-MM-DD for changes
+#   4. agent_run(worktree, prompt) → Claude plans, may push knowledge updates
 #   5. agent_run(worktree, prompt) → Claude plans, commits to ops branch
 #   6. If ops branch has commits: pr_create → pr_walk_to_merge (review-bot)
 #
 # Usage:
 #   planner-run.sh [projects/disinto.toml]   # project config (default: disinto)
@ -24,11 +22,10 @@ FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
 # Accept project config from argument; default to disinto (planner is disinto infrastructure)
 export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 # Set override BEFORE sourcing env.sh so it survives any later re-source of
 # env.sh from nested shells / claude -p tools (#762, #747)
 export FORGE_TOKEN_OVERRIDE="${FORGE_PLANNER_TOKEN:-}"
 # shellcheck source=../lib/env.sh
 source "$FACTORY_ROOT/lib/env.sh"
 # Use planner-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_PLANNER_TOKEN:-${FORGE_TOKEN}}"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
@ -37,10 +34,6 @@ source "$FACTORY_ROOT/lib/worktree.sh"
 source "$FACTORY_ROOT/lib/guard.sh"
 # shellcheck source=../lib/agent-sdk.sh
 source "$FACTORY_ROOT/lib/agent-sdk.sh"
 # shellcheck source=../lib/ci-helpers.sh
 source "$FACTORY_ROOT/lib/ci-helpers.sh"
 # shellcheck source=../lib/pr-lifecycle.sh
 source "$FACTORY_ROOT/lib/pr-lifecycle.sh"
 LOG_FILE="${DISINTO_LOG_DIR}/planner/planner.log"
 # shellcheck disable=SC2034  # consumed by agent-sdk.sh
@ -152,69 +145,12 @@ ${PROMPT_FOOTER}"
 # ── Create worktree ──────────────────────────────────────────────────────
 formula_worktree_setup "$WORKTREE"
 # ── Prepare ops branch for PR-based merge (#765) ────────────────────────
 PLANNER_OPS_BRANCH="planner/run-$(date -u +%Y-%m-%d)"
 (
  cd "$OPS_REPO_ROOT"
  git fetch origin "${PRIMARY_BRANCH}" --quiet 2>/dev/null || true
  git checkout "${PRIMARY_BRANCH}" --quiet 2>/dev/null || true
  git pull --ff-only origin "${PRIMARY_BRANCH}" --quiet 2>/dev/null || true
  # Create (or reset to) a fresh branch from PRIMARY_BRANCH
  git checkout -B "$PLANNER_OPS_BRANCH" "origin/${PRIMARY_BRANCH}" --quiet 2>/dev/null || \
    git checkout -b "$PLANNER_OPS_BRANCH" --quiet 2>/dev/null || true
 )
 log "ops branch: ${PLANNER_OPS_BRANCH}"
 # ── Run agent ─────────────────────────────────────────────────────────────
 export CLAUDE_MODEL="opus"
 agent_run --worktree "$WORKTREE" "$PROMPT"
 log "agent_run complete"
 # ── PR lifecycle: create PR on ops repo and walk to merge (#765) ─────────
 OPS_FORGE_API="${FORGE_API_BASE}/repos/${FORGE_OPS_REPO}"
 ops_has_commits=false
 if ! git -C "$OPS_REPO_ROOT" diff --quiet "origin/${PRIMARY_BRANCH}..${PLANNER_OPS_BRANCH}" 2>/dev/null; then
  ops_has_commits=true
 fi
 if [ "$ops_has_commits" = "true" ]; then
  log "ops branch has commits — creating PR"
  # Push the branch to the ops remote
  git -C "$OPS_REPO_ROOT" push origin "$PLANNER_OPS_BRANCH" --quiet 2>/dev/null || \
    git -C "$OPS_REPO_ROOT" push --force-with-lease origin "$PLANNER_OPS_BRANCH" 2>/dev/null
  # Temporarily point FORGE_API at the ops repo for pr-lifecycle functions
  ORIG_FORGE_API="$FORGE_API"
  export FORGE_API="$OPS_FORGE_API"
  # Ops repo typically has no Woodpecker CI — skip CI polling
  ORIG_WOODPECKER_REPO_ID="${WOODPECKER_REPO_ID:-2}"
  export WOODPECKER_REPO_ID="0"
  PR_NUM=$(pr_create "$PLANNER_OPS_BRANCH" \
    "chore: planner run $(date -u +%Y-%m-%d)" \
    "Automated planner run — updates prerequisite tree, memory, and vault items." \
    "${PRIMARY_BRANCH}" \
    "$OPS_FORGE_API") || true
  if [ -n "$PR_NUM" ]; then
    log "ops PR #${PR_NUM} created — walking to merge"
    SESSION_ID=$(cat "$SID_FILE" 2>/dev/null || echo "planner-$$")
    pr_walk_to_merge "$PR_NUM" "$SESSION_ID" "$OPS_REPO_ROOT" 1 2 || {
      log "ops PR #${PR_NUM} walk finished: ${_PR_WALK_EXIT_REASON:-unknown}"
    }
    log "ops PR #${PR_NUM} result: ${_PR_WALK_EXIT_REASON:-unknown}"
  else
    log "WARNING: failed to create ops PR for branch ${PLANNER_OPS_BRANCH}"
  fi
  # Restore original FORGE_API
  export FORGE_API="$ORIG_FORGE_API"
  export WOODPECKER_REPO_ID="$ORIG_WOODPECKER_REPO_ID"
 else
  log "no ops changes — skipping PR creation"
 fi
 # Persist watermarks so next run can skip if nothing changed
 mkdir -p "$FACTORY_ROOT/state"
 echo "$CURRENT_SHA" > "$LAST_SHA_FILE"
--- a/predictor/AGENTS.md
+++ b/predictor/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Predictor Agent
 **Role**: Abstract adversary (the "goblin"). Runs a 2-step formula
--- a/predictor/predictor-run.sh
+++ b/predictor/predictor-run.sh
@ -23,11 +23,10 @@ FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
 # Accept project config from argument; default to disinto
 export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 # Set override BEFORE sourcing env.sh so it survives any later re-source of
 # env.sh from nested shells / claude -p tools (#762, #747)
 export FORGE_TOKEN_OVERRIDE="${FORGE_PREDICTOR_TOKEN:-}"
 # shellcheck source=../lib/env.sh
 source "$FACTORY_ROOT/lib/env.sh"
 # Use predictor-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_PREDICTOR_TOKEN:-${FORGE_TOKEN}}"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
--- a/review/AGENTS.md
+++ b/review/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Review Agent
 **Role**: AI-powered PR review — post structured findings and formal
--- a/site/collect-engagement.sh
+++ b/site/collect-engagement.sh
@ -59,21 +59,6 @@ fi
 mkdir -p "$EVIDENCE_DIR"
 # Verify input is Caddy JSON format (not Combined Log Format or other)
 first_line=$(grep -m1 '.' "$CADDY_LOG" || true)
 if [ -z "$first_line" ]; then
  log "WARN: Caddy access log is empty at ${CADDY_LOG}"
  echo "WARN: Caddy access log is empty — nothing to parse." >&2
  exit 0
 fi
 if ! printf '%s\n' "$first_line" | jq empty 2>/dev/null; then
  preview="${first_line:0:200}"
  log "ERROR: Input file is not Caddy JSON format (expected structured JSON access log). Got: ${preview}"
  echo "ERROR: Input file is not Caddy JSON format (expected structured JSON access log)." >&2
  echo "Got: ${preview}" >&2
  exit 1
 fi
 # ── Parse access log ────────────────────────────────────────────────────────
 log "Parsing ${CADDY_LOG} for entries since $(date -u -d "@${CUTOFF_TS}" +%Y-%m-%dT%H:%M:%SZ 2>/dev/null || echo "${CUTOFF_TS}")"
--- a/supervisor/AGENTS.md
+++ b/supervisor/AGENTS.md
@ -1,4 +1,4 @@
-<!-- last-reviewed: c363ee0aea2ae447daab28c2c850d6abefc8c6b5 -->
+<!-- last-reviewed: 3e65878093bbbcea6dfe4db341f82dc89d4e0ac0 -->
 # Supervisor Agent
 **Role**: Health monitoring and auto-remediation, executed as a formula-driven
@ -7,11 +7,13 @@ then runs an interactive Claude session (sonnet) that assesses health, auto-fixe
 issues, and writes a daily journal. When blocked on external
 resources or human decisions, files vault items instead of escalating directly.
-**Trigger**: `supervisor-run.sh` is invoked by two polling loops:
+**Trigger**: `supervisor-run.sh` is invoked by the polling loop in `docker/edge/entrypoint-edge.sh`
- **Agents container** (`docker/agents/entrypoint.sh`): every `SUPERVISOR_INTERVAL` seconds (default 1200 = 20 min). Controlled by the `supervisor` role in `AGENT_ROLES` (included in the default seven-role set since P1/#801). Logs to `supervisor.log` in the agents container.
+every 20 minutes (line 50-53). Sources `lib/guard.sh` and calls `check_active supervisor` first
- **Edge container** (`docker/edge/entrypoint-edge.sh`): separate loop in the edge container (line 169-172). Runs independently of the agents container's polling schedule.
+— skips if `$FACTORY_ROOT/state/.supervisor-active` is absent. Then runs `claude -p` via
-
+`agent-sdk.sh`, injects `formulas/run-supervisor.toml` with pre-collected metrics as context,
-Both invoke the same `supervisor-run.sh`. Sources `lib/guard.sh` and calls `check_active supervisor` first — skips if `$FACTORY_ROOT/state/.supervisor-active` is absent. Then runs `claude -p` via `agent-sdk.sh`, injects `formulas/run-supervisor.toml` with pre-collected metrics as context, and cleans up on completion or timeout.
+and cleans up on completion or timeout (20 min max session). Note: the supervisor runs in the
 **edge container** (`entrypoint-edge.sh`), not the agent container — this distinction matters
 for operators debugging the factory.
 **Key files**:
 - `supervisor/supervisor-run.sh` — Polling loop participant + orchestrator: lock, memory guard,
@ -37,7 +39,6 @@ P3 (degraded PRs, circular deps, stale deps), P4 (housekeeping).
 **Environment variables consumed**:
 - `FORGE_TOKEN`, `FORGE_SUPERVISOR_TOKEN` (falls back to FORGE_TOKEN), `FORGE_REPO`, `FORGE_API`, `PROJECT_NAME`, `PROJECT_REPO_ROOT`, `OPS_REPO_ROOT`
 - `PRIMARY_BRANCH`, `CLAUDE_MODEL` (set to sonnet by supervisor-run.sh)
 - `SUPERVISOR_INTERVAL` — polling interval in seconds for agents container (default 1200 = 20 min)
 - `WOODPECKER_TOKEN`, `WOODPECKER_SERVER`, `WOODPECKER_DB_PASSWORD`, `WOODPECKER_DB_USER`, `WOODPECKER_DB_HOST`, `WOODPECKER_DB_NAME` — CI database queries
 **Degraded mode (Issue #544)**: When `OPS_REPO_ROOT` is not set or the directory doesn't exist, the supervisor runs in degraded mode:
--- a/supervisor/supervisor-run.sh
+++ b/supervisor/supervisor-run.sh
@ -25,11 +25,10 @@ FACTORY_ROOT="$(dirname "$SCRIPT_DIR")"
 # Accept project config from argument; default to disinto
 export PROJECT_TOML="${1:-$FACTORY_ROOT/projects/disinto.toml}"
 # Set override BEFORE sourcing env.sh so it survives any later re-source of
 # env.sh from nested shells / claude -p tools (#762, #747)
 export FORGE_TOKEN_OVERRIDE="${FORGE_SUPERVISOR_TOKEN:-}"
 # shellcheck source=../lib/env.sh
 source "$FACTORY_ROOT/lib/env.sh"
 # Use supervisor-bot's own Forgejo identity (#747)
 FORGE_TOKEN="${FORGE_SUPERVISOR_TOKEN:-${FORGE_TOKEN}}"
 # shellcheck source=../lib/formula-session.sh
 source "$FACTORY_ROOT/lib/formula-session.sh"
 # shellcheck source=../lib/worktree.sh
--- a/tests/lib-hvault.bats
+++ b/tests/lib-hvault.bats
@ -1,215 +0,0 @@
 #!/usr/bin/env bats
 # tests/lib-hvault.bats — Unit tests for lib/hvault.sh
 #
 # Runs against a dev-mode Vault server (single binary, no LXC needed).
 # CI launches vault server -dev inline before running these tests.
 VAULT_BIN="${VAULT_BIN:-vault}"
 setup_file() {
  export TEST_DIR
  TEST_DIR="$(cd "$(dirname "$BATS_TEST_FILENAME")/.." && pwd)"
  # Start dev-mode vault on a random port
  export VAULT_DEV_PORT
  VAULT_DEV_PORT="$(shuf -i 18200-18299 -n 1)"
  export VAULT_ADDR="http://127.0.0.1:${VAULT_DEV_PORT}"
  "$VAULT_BIN" server -dev \
    -dev-listen-address="127.0.0.1:${VAULT_DEV_PORT}" \
    -dev-root-token-id="test-root-token" \
    -dev-no-store-token \
    &>"${BATS_FILE_TMPDIR}/vault.log" &
  export VAULT_PID=$!
  export VAULT_TOKEN="test-root-token"
  # Wait for vault to be ready (up to 10s)
  local i=0
  while ! curl -sf "${VAULT_ADDR}/v1/sys/health" >/dev/null 2>&1; do
    sleep 0.5
    i=$((i + 1))
    if [ "$i" -ge 20 ]; then
      echo "Vault failed to start. Log:" >&2
      cat "${BATS_FILE_TMPDIR}/vault.log" >&2
      return 1
    fi
  done
 }
 teardown_file() {
  if [ -n "${VAULT_PID:-}" ]; then
    kill "$VAULT_PID" 2>/dev/null || true
    wait "$VAULT_PID" 2>/dev/null || true
  fi
 }
 setup() {
  # Source the module under test
  source "${TEST_DIR}/lib/hvault.sh"
  export VAULT_ADDR VAULT_TOKEN
 }
 # ── hvault_kv_put + hvault_kv_get ────────────────────────────────────────────
@test "hvault_kv_put writes and hvault_kv_get reads a secret" {
  run hvault_kv_put "test/myapp" "username=admin" "password=s3cret"
  [ "$status" -eq 0 ]
  run hvault_kv_get "test/myapp"
  [ "$status" -eq 0 ]
  echo "$output" | jq -e '.username == "admin"'
  echo "$output" | jq -e '.password == "s3cret"'
 }
@test "hvault_kv_get extracts a single key" {
  hvault_kv_put "test/single" "foo=bar" "baz=qux"
  run hvault_kv_get "test/single" "foo"
  [ "$status" -eq 0 ]
  [ "$output" = "bar" ]
 }
@test "hvault_kv_get fails for missing key" {
  hvault_kv_put "test/keymiss" "exists=yes"
  run hvault_kv_get "test/keymiss" "nope"
  [ "$status" -ne 0 ]
 }
@test "hvault_kv_get fails for missing path" {
  run hvault_kv_get "test/does-not-exist-$(date +%s)"
  [ "$status" -ne 0 ]
 }
@test "hvault_kv_put fails without KEY=VAL" {
  run hvault_kv_put "test/bad"
  [ "$status" -ne 0 ]
  echo "$output" | grep -q '"error":true' || echo "$stderr" | grep -q '"error":true'
 }
@test "hvault_kv_put rejects malformed pair (no =)" {
  run hvault_kv_put "test/bad2" "noequals"
  [ "$status" -ne 0 ]
 }
@test "hvault_kv_get fails without PATH" {
  run hvault_kv_get
  [ "$status" -ne 0 ]
 }
 # ── hvault_kv_list ───────────────────────────────────────────────────────────
@test "hvault_kv_list lists keys at a path" {
  hvault_kv_put "test/listdir/a" "k=1"
  hvault_kv_put "test/listdir/b" "k=2"
  run hvault_kv_list "test/listdir"
  [ "$status" -eq 0 ]
  echo "$output" | jq -e '. | length >= 2'
  echo "$output" | jq -e 'index("a")'
  echo "$output" | jq -e 'index("b")'
 }
@test "hvault_kv_list fails on nonexistent path" {
  run hvault_kv_list "test/no-such-path-$(date +%s)"
  [ "$status" -ne 0 ]
 }
@test "hvault_kv_list fails without PATH" {
  run hvault_kv_list
  [ "$status" -ne 0 ]
 }
 # ── hvault_policy_apply ──────────────────────────────────────────────────────
@test "hvault_policy_apply creates a policy" {
  local pfile="${BATS_TEST_TMPDIR}/test-policy.hcl"
  cat > "$pfile" <<'HCL'
 path "secret/data/test/*" {
  capabilities = ["read"]
 }
 HCL
  run hvault_policy_apply "test-reader" "$pfile"
  [ "$status" -eq 0 ]
  # Verify the policy exists via Vault API
  run curl -sf -H "X-Vault-Token: ${VAULT_TOKEN}" \
    "${VAULT_ADDR}/v1/sys/policies/acl/test-reader"
  [ "$status" -eq 0 ]
  echo "$output" | jq -e '.data.policy' | grep -q "secret/data/test"
 }
@test "hvault_policy_apply is idempotent" {
  local pfile="${BATS_TEST_TMPDIR}/idem-policy.hcl"
  printf 'path "secret/*" { capabilities = ["list"] }\n' > "$pfile"
  run hvault_policy_apply "idem-policy" "$pfile"
  [ "$status" -eq 0 ]
  # Apply again — should succeed
  run hvault_policy_apply "idem-policy" "$pfile"
  [ "$status" -eq 0 ]
 }
@test "hvault_policy_apply fails with missing file" {
  run hvault_policy_apply "bad-policy" "/nonexistent/policy.hcl"
  [ "$status" -ne 0 ]
 }
@test "hvault_policy_apply fails without args" {
  run hvault_policy_apply
  [ "$status" -ne 0 ]
 }
 # ── hvault_token_lookup ──────────────────────────────────────────────────────
@test "hvault_token_lookup returns token info" {
  run hvault_token_lookup
  [ "$status" -eq 0 ]
  echo "$output" | jq -e '.policies'
  echo "$output" | jq -e '.accessor'
  echo "$output" | jq -e 'has("ttl")'
 }
@test "hvault_token_lookup fails without VAULT_TOKEN" {
  unset VAULT_TOKEN
  run hvault_token_lookup
  [ "$status" -ne 0 ]
 }
@test "hvault_token_lookup fails without VAULT_ADDR" {
  unset VAULT_ADDR
  run hvault_token_lookup
  [ "$status" -ne 0 ]
 }
 # ── hvault_jwt_login ─────────────────────────────────────────────────────────
@test "hvault_jwt_login fails without VAULT_ADDR" {
  unset VAULT_ADDR
  run hvault_jwt_login "myrole" "fakejwt"
  [ "$status" -ne 0 ]
 }
@test "hvault_jwt_login fails without args" {
  run hvault_jwt_login
  [ "$status" -ne 0 ]
 }
@test "hvault_jwt_login returns error for unconfigured jwt auth" {
  # JWT auth backend is not enabled in dev mode by default — expect failure
  run hvault_jwt_login "myrole" "eyJhbGciOiJSUzI1NiJ9.fake.sig"
  [ "$status" -ne 0 ]
 }
 # ── Env / prereq errors ─────────────────────────────────────────────────────
@test "all functions fail with structured JSON error when VAULT_ADDR unset" {
  unset VAULT_ADDR
  for fn in hvault_kv_get hvault_kv_put hvault_kv_list hvault_policy_apply hvault_token_lookup; do
    run $fn "dummy" "dummy"
    [ "$status" -ne 0 ]
  done
 }
--- a/tests/smoke-init.sh
+++ b/tests/smoke-init.sh
@ -29,8 +29,7 @@ cleanup() {
  pkill -f "mock-forgejo.py" 2>/dev/null || true
  rm -rf "$MOCK_BIN" /tmp/smoke-test-repo \
         "${FACTORY_ROOT}/projects/smoke-repo.toml" \
-         /tmp/smoke-claude-shared /tmp/smoke-home-claude \
+         /tmp/smoke-claude-shared /tmp/smoke-home-claude
         /tmp/smoke-env-before-rerun /tmp/smoke-env-before-dryrun
  # Restore .env only if we created the backup
  if [ -f "${FACTORY_ROOT}/.env.smoke-backup" ]; then
    mv "${FACTORY_ROOT}/.env.smoke-backup" "${FACTORY_ROOT}/.env"
@ -179,30 +178,8 @@ else
  fail "disinto init exited non-zero"
 fi
-# ── Dry-run test: must not modify state ────────────────────────────────────
+# ── Idempotency test: run init again ───────────────────────────────────────
 echo "=== Dry-run test ==="
 cp "${FACTORY_ROOT}/.env" /tmp/smoke-env-before-dryrun
 if bash "${FACTORY_ROOT}/bin/disinto" init \
  "${TEST_SLUG}" \
  --bare --yes --dry-run \
  --forge-url "$FORGE_URL" \
  --repo-root "/tmp/smoke-test-repo" 2>&1 | grep -q "Dry run complete"; then
  pass "disinto init --dry-run exited successfully"
 else
  fail "disinto init --dry-run did not complete"
 fi
 # Verify --dry-run did not modify .env
 if diff -q /tmp/smoke-env-before-dryrun "${FACTORY_ROOT}/.env" >/dev/null 2>&1; then
  pass "dry-run: .env unchanged"
 else
  fail "dry-run: .env was modified (should be read-only)"
 fi
 rm -f /tmp/smoke-env-before-dryrun
 # ── Idempotency test: run init again, verify .env is stable ────────────────
 echo "=== Idempotency test: running disinto init again ==="
 cp "${FACTORY_ROOT}/.env" /tmp/smoke-env-before-rerun
 if bash "${FACTORY_ROOT}/bin/disinto" init \
  "${TEST_SLUG}" \
  --bare --yes \
@ -213,29 +190,6 @@ else
  fail "disinto init (re-run) exited non-zero"
 fi
 # Verify .env is stable across re-runs (no token churn)
 if diff -q /tmp/smoke-env-before-rerun "${FACTORY_ROOT}/.env" >/dev/null 2>&1; then
  pass "idempotency: .env unchanged on re-run"
 else
  fail "idempotency: .env changed on re-run (token churn detected)"
  diff /tmp/smoke-env-before-rerun "${FACTORY_ROOT}/.env" >&2 || true
 fi
 rm -f /tmp/smoke-env-before-rerun
 # Verify FORGE_ADMIN_TOKEN is stored in .env
 if grep -q '^FORGE_ADMIN_TOKEN=' "${FACTORY_ROOT}/.env"; then
  pass ".env contains FORGE_ADMIN_TOKEN"
 else
  fail ".env missing FORGE_ADMIN_TOKEN"
 fi
 # Verify HUMAN_TOKEN is stored in .env
 if grep -q '^HUMAN_TOKEN=' "${FACTORY_ROOT}/.env"; then
  pass ".env contains HUMAN_TOKEN"
 else
  fail ".env missing HUMAN_TOKEN"
 fi
 # ── 4. Verify Forgejo state ─────────────────────────────────────────────────
 echo "=== 4/6 Verifying Forgejo state ==="
--- a/tests/smoke-load-secret.sh
+++ b/tests/smoke-load-secret.sh
@ -1,162 +0,0 @@
 #!/usr/bin/env bash
 # tests/smoke-load-secret.sh — Unit tests for load_secret() precedence chain
 #
 # Covers the 4 precedence cases:
 #   1. /secrets/<NAME>.env  (Nomad template)
 #   2. Current environment
 #   3. secrets/<NAME>.enc   (age-encrypted per-key file)
 #   4. Default / empty fallback
 #
 # Required tools: bash, age (for case 3)
 set -euo pipefail
 FACTORY_ROOT="$(cd "$(dirname "$0")/.." && pwd)"
 fail() { printf 'FAIL: %s\n' "$*" >&2; FAILED=1; }
 pass() { printf 'PASS: %s\n' "$*"; }
 FAILED=0
 # Set up a temp workspace and fake HOME so age key paths work
 test_dir=$(mktemp -d)
 fake_home=$(mktemp -d)
 trap 'rm -rf "$test_dir" "$fake_home"' EXIT
 # Minimal env for sourcing env.sh's load_secret function without the full boot
 # We source the function definition directly to isolate the unit under test.
 # shellcheck disable=SC2034
 export USER="${USER:-test}"
 export HOME="$fake_home"
 # Source env.sh to get load_secret (and FACTORY_ROOT)
 source "${FACTORY_ROOT}/lib/env.sh"
 # ── Case 4: Default / empty fallback ────────────────────────────────────────
 echo "=== 1/5 Case 4: default fallback ==="
 unset TEST_SECRET_FALLBACK 2>/dev/null || true
 val=$(load_secret TEST_SECRET_FALLBACK "my-default")
 if [ "$val" = "my-default" ]; then
  pass "load_secret returns default when nothing is set"
 else
  fail "Expected 'my-default', got '${val}'"
 fi
 val=$(load_secret TEST_SECRET_FALLBACK)
 if [ -z "$val" ]; then
  pass "load_secret returns empty when no default and nothing set"
 else
  fail "Expected empty, got '${val}'"
 fi
 # ── Case 2: Environment variable already set ────────────────────────────────
 echo "=== 2/5 Case 2: environment variable ==="
 export TEST_SECRET_ENV="from-environment"
 val=$(load_secret TEST_SECRET_ENV "ignored-default")
 if [ "$val" = "from-environment" ]; then
  pass "load_secret returns env value over default"
 else
  fail "Expected 'from-environment', got '${val}'"
 fi
 unset TEST_SECRET_ENV
 # ── Case 3: Age-encrypted per-key file ──────────────────────────────────────
 echo "=== 3/5 Case 3: age-encrypted secret ==="
 if command -v age &>/dev/null && command -v age-keygen &>/dev/null; then
  # Generate a test age key
  age_key_dir="${fake_home}/.config/sops/age"
  mkdir -p "$age_key_dir"
  age-keygen -o "${age_key_dir}/keys.txt" 2>/dev/null
  pub_key=$(age-keygen -y "${age_key_dir}/keys.txt")
  # Create encrypted secret
  secrets_dir="${FACTORY_ROOT}/secrets"
  mkdir -p "$secrets_dir"
  printf 'age-test-value' | age -r "$pub_key" -o "${secrets_dir}/TEST_SECRET_AGE.enc"
  unset TEST_SECRET_AGE 2>/dev/null || true
  val=$(load_secret TEST_SECRET_AGE "fallback")
  if [ "$val" = "age-test-value" ]; then
    pass "load_secret decrypts age-encrypted secret"
  else
    fail "Expected 'age-test-value', got '${val}'"
  fi
  # Verify caching: call load_secret directly (not in subshell) so export propagates
  unset TEST_SECRET_AGE 2>/dev/null || true
  load_secret TEST_SECRET_AGE >/dev/null
  if [ "${TEST_SECRET_AGE:-}" = "age-test-value" ]; then
    pass "load_secret caches decrypted value in environment (direct call)"
  else
    fail "Decrypted value not cached in environment"
  fi
  # Clean up test secret
  rm -f "${secrets_dir}/TEST_SECRET_AGE.enc"
  rmdir "$secrets_dir" 2>/dev/null || true
  unset TEST_SECRET_AGE
 else
  echo "SKIP: age/age-keygen not found — skipping age decryption test"
 fi
 # ── Case 1: Nomad template path ────────────────────────────────────────────
 echo "=== 4/5 Case 1: Nomad template (/secrets/<NAME>.env) ==="
 nomad_dir="/secrets"
 if [ -w "$(dirname "$nomad_dir")" ] 2>/dev/null || [ -w "$nomad_dir" ] 2>/dev/null; then
  mkdir -p "$nomad_dir"
  printf 'TEST_SECRET_NOMAD=from-nomad-template\n' > "${nomad_dir}/TEST_SECRET_NOMAD.env"
  # Even with env set, Nomad path takes precedence
  export TEST_SECRET_NOMAD="from-env-should-lose"
  val=$(load_secret TEST_SECRET_NOMAD "default")
  if [ "$val" = "from-nomad-template" ]; then
    pass "load_secret prefers Nomad template over env"
  else
    fail "Expected 'from-nomad-template', got '${val}'"
  fi
  rm -f "${nomad_dir}/TEST_SECRET_NOMAD.env"
  rmdir "$nomad_dir" 2>/dev/null || true
  unset TEST_SECRET_NOMAD
 else
  echo "SKIP: /secrets not writable — skipping Nomad template test (needs root or container)"
 fi
 # ── Precedence: env beats age ────────────────────────────────────────────
 echo "=== 5/5 Precedence: env beats age-encrypted ==="
 if command -v age &>/dev/null && command -v age-keygen &>/dev/null; then
  age_key_dir="${fake_home}/.config/sops/age"
  mkdir -p "$age_key_dir"
  [ -f "${age_key_dir}/keys.txt" ] || age-keygen -o "${age_key_dir}/keys.txt" 2>/dev/null
  pub_key=$(age-keygen -y "${age_key_dir}/keys.txt")
  secrets_dir="${FACTORY_ROOT}/secrets"
  mkdir -p "$secrets_dir"
  printf 'age-value-should-lose' | age -r "$pub_key" -o "${secrets_dir}/TEST_SECRET_PREC.enc"
  export TEST_SECRET_PREC="env-value-wins"
  val=$(load_secret TEST_SECRET_PREC "default")
  if [ "$val" = "env-value-wins" ]; then
    pass "load_secret prefers env over age-encrypted file"
  else
    fail "Expected 'env-value-wins', got '${val}'"
  fi
  rm -f "${secrets_dir}/TEST_SECRET_PREC.enc"
  rmdir "$secrets_dir" 2>/dev/null || true
  unset TEST_SECRET_PREC
 else
  echo "SKIP: age not found — skipping precedence test"
 fi
 # ── Summary ───────────────────────────────────────────────────────────────
 echo ""
 if [ "$FAILED" -ne 0 ]; then
  echo "=== SMOKE-LOAD-SECRET TEST FAILED ==="
  exit 1
 fi
 echo "=== SMOKE-LOAD-SECRET TEST PASSED ==="
--- a/tools/edge-control/README.md
+++ b/tools/edge-control/README.md
@ -83,12 +83,9 @@ curl -sL https://raw.githubusercontent.com/disinto-admin/disinto/fix/issue-621/t
   - Permissions: `root:disinto-register 0750`
 3. **Installs Caddy**:
   - Backs up any pre-existing `/etc/caddy/Caddyfile` to `/etc/caddy/Caddyfile.pre-disinto`
   - Download Caddy with Gandi DNS plugin
   - Enable admin API on `127.0.0.1:2019`
   - Configure wildcard cert for `*.disinto.ai` via DNS-01
   - Creates `/etc/caddy/extra.d/` for operator-owned site blocks
   - Emitted Caddyfile ends with `import /etc/caddy/extra.d/*.caddy`
 4. **Sets up SSH**:
   - Creates `disinto-register` authorized_keys with forced command
@ -98,27 +95,6 @@ curl -sL https://raw.githubusercontent.com/disinto-admin/disinto/fix/issue-621/t
   - `/opt/disinto-edge/register.sh` — forced command handler
   - `/opt/disinto-edge/lib/*.sh` — helper libraries
 ## Operator-Owned Site Blocks
 Edge-control owns the top-level `/etc/caddy/Caddyfile` and dynamic `<project>.<DOMAIN_SUFFIX>` routes injected via the Caddy admin API. Operators own everything under `/etc/caddy/extra.d/`.
 To serve non-tunnel content (apex domain, www redirect, static sites), drop `.caddy` files into `/etc/caddy/extra.d/`:
 ```bash
 # Example: /etc/caddy/extra.d/landing.caddy
 disinto.ai {
  root * /home/debian/disinto-site
  file_server
 }
 # Example: /etc/caddy/extra.d/www-redirect.caddy
 www.disinto.ai {
  redir https://disinto.ai{uri} permanent
 }
 ```
 These files survive across `install.sh` re-runs. The `--extra-caddyfile <path>` flag overrides the default import glob (`/etc/caddy/extra.d/*.caddy`) if needed.
 ## Usage
 ### Register a Tunnel (from dev box)
--- a/tools/edge-control/install.sh
+++ b/tools/edge-control/install.sh
@ -43,21 +43,18 @@ INSTALL_DIR="/opt/disinto-edge"
 REGISTRY_DIR="/var/lib/disinto"
 CADDY_VERSION="2.8.4"
 DOMAIN_SUFFIX="disinto.ai"
 EXTRA_CADDYFILE="/etc/caddy/extra.d/*.caddy"
 usage() {
  cat <<EOF
 Usage: $0 [options]
 Options:
-  --gandi-token <token>       Gandi API token for wildcard cert (required)
+  --gandi-token <token>   Gandi API token for wildcard cert (required)
-  --install-dir <dir>         Install directory (default: /opt/disinto-edge)
+  --install-dir <dir>     Install directory (default: /opt/disinto-edge)
-  --registry-dir <dir>        Registry directory (default: /var/lib/disinto)
+  --registry-dir <dir>    Registry directory (default: /var/lib/disinto)
-  --caddy-version <ver>       Caddy version to install (default: ${CADDY_VERSION})
+  --caddy-version <ver>   Caddy version to install (default: ${CADDY_VERSION})
-  --domain-suffix <suffix>    Domain suffix for tunnels (default: disinto.ai)
+  --domain-suffix <suffix> Domain suffix for tunnels (default: disinto.ai)
-  --extra-caddyfile <path>    Import path for operator-owned Caddy config
+  -h, --help              Show this help
                              (default: /etc/caddy/extra.d/*.caddy)
  -h, --help                  Show this help
 Example:
  $0 --gandi-token YOUR_GANDI_API_TOKEN
@ -87,10 +84,6 @@ while [[ $# -gt 0 ]]; do
      DOMAIN_SUFFIX="$2"
      shift 2
      ;;
    --extra-caddyfile)
      EXTRA_CADDYFILE="$2"
      shift 2
      ;;
    -h|--help)
      usage
      ;;
@ -232,29 +225,8 @@ EOF
 chmod 600 "$GANDI_ENV"
 # Create Caddyfile with admin API and wildcard cert
 # Note: Caddy auto-generates server names (srv0, srv1, …). lib/caddy.sh
 # discovers the server name dynamically via _discover_server_name() so we
 # don't need to name the server here.
 CADDYFILE="/etc/caddy/Caddyfile"
-
+cat > "$CADDYFILE" <<EOF
 # Back up existing Caddyfile before overwriting
 if [ -f "$CADDYFILE" ] && [ ! -f "${CADDYFILE}.pre-disinto" ]; then
  cp "$CADDYFILE" "${CADDYFILE}.pre-disinto"
  log_info "Backed up existing Caddyfile to ${CADDYFILE}.pre-disinto"
 fi
 # Create extra.d directory for operator-owned site blocks
 EXTRA_DIR="/etc/caddy/extra.d"
 mkdir -p "$EXTRA_DIR"
 chmod 0755 "$EXTRA_DIR"
 if getent group caddy >/dev/null 2>&1; then
  chown root:caddy "$EXTRA_DIR"
 else
  log_warn "Group 'caddy' does not exist; extra.d owned by root:root"
 fi
 log_info "Created ${EXTRA_DIR} for operator-owned Caddy config"
 cat > "$CADDYFILE" <<CADDYEOF
 # Caddy configuration for edge control plane
 # Admin API enabled on 127.0.0.1:2019
@ -268,10 +240,7 @@ cat > "$CADDYFILE" <<CADDYEOF
    dns gandi {env.GANDI_API_KEY}
  }
 }
-
+EOF
 # Operator-owned site blocks (apex, www, static content, etc.)
 import ${EXTRA_CADDYFILE}
 CADDYEOF
 # Start Caddy
 systemctl restart caddy 2>/dev/null || {
@ -390,7 +359,6 @@ echo "Configuration:"
 echo "  Install directory: ${INSTALL_DIR}"
 echo "  Registry: ${REGISTRY_FILE}"
 echo "  Caddy admin API: http://127.0.0.1:2019"
 echo "  Operator site blocks: ${EXTRA_DIR}/ (import ${EXTRA_CADDYFILE})"
 echo ""
 echo "Users:"
 echo "  disinto-register - SSH forced command (runs ${INSTALL_DIR}/register.sh)"
--- a/tools/edge-control/lib/caddy.sh
+++ b/tools/edge-control/lib/caddy.sh
@ -19,24 +19,6 @@ CADDY_ADMIN_URL="${CADDY_ADMIN_URL:-http://127.0.0.1:2019}"
 # Domain suffix for projects
 DOMAIN_SUFFIX="${DOMAIN_SUFFIX:-disinto.ai}"
 # Discover the Caddy server name that listens on :80/:443
 # Usage: _discover_server_name
 _discover_server_name() {
  local server_name
  server_name=$(curl -sS "${CADDY_ADMIN_URL}/config/apps/http/servers" \
    | jq -r 'to_entries | map(select(.value.listen[]? | test(":(80|443)$"))) | .[0].key // empty') || {
    echo "Error: could not query Caddy admin API for servers" >&2
    return 1
  }
  if [ -z "$server_name" ]; then
    echo "Error: could not find a Caddy server listening on :80/:443" >&2
    return 1
  fi
  echo "$server_name"
 }
 # Add a route for a project
 # Usage: add_route <project> <port>
 add_route() {
@ -44,9 +26,6 @@ add_route() {
  local port="$2"
  local fqdn="${project}.${DOMAIN_SUFFIX}"
  local server_name
  server_name=$(_discover_server_name) || return 1
  # Build the route configuration (partial config)
  local route_config
  route_config=$(cat <<EOF
@ -79,21 +58,16 @@ add_route() {
 EOF
 )
-  # Append route via admin API, checking HTTP status
+  # Append route using POST /config/apps/http/servers/edge/routes
-  local response status body
+  local response
-  response=$(curl -sS -w '\n%{http_code}' -X POST \
+  response=$(curl -s -X POST \
-    "${CADDY_ADMIN_URL}/config/apps/http/servers/${server_name}/routes" \
+    "${CADDY_ADMIN_URL}/config/apps/http/servers/edge/routes" \
    -H "Content-Type: application/json" \
-    -d "$route_config") || {
+    -d "$route_config" 2>&1) || {
    echo "Error: failed to add route for ${fqdn}" >&2
    echo "Response: ${response}" >&2
    return 1
  }
  status=$(echo "$response" | tail -n1)
  body=$(echo "$response" | sed '$d')
  if [ "$status" -ge 400 ]; then
    echo "Error: Caddy admin API returned ${status}: ${body}" >&2
    return 1
  fi
  echo "Added route: ${fqdn} → 127.0.0.1:${port}" >&2
 }
@ -104,45 +78,31 @@ remove_route() {
  local project="$1"
  local fqdn="${project}.${DOMAIN_SUFFIX}"
-  local server_name
+  # First, get current routes
-  server_name=$(_discover_server_name) || return 1
+  local routes_json
-
+  routes_json=$(curl -s "${CADDY_ADMIN_URL}/config/apps/http/servers/edge/routes" 2>&1) || {
  # First, get current routes, checking HTTP status
  local response status body
  response=$(curl -sS -w '\n%{http_code}' \
    "${CADDY_ADMIN_URL}/config/apps/http/servers/${server_name}/routes") || {
    echo "Error: failed to get current routes" >&2
    return 1
  }
  status=$(echo "$response" | tail -n1)
  body=$(echo "$response" | sed '$d')
  if [ "$status" -ge 400 ]; then
    echo "Error: Caddy admin API returned ${status}: ${body}" >&2
    return 1
  fi
  # Find the route index that matches our fqdn using jq
  local route_index
-  route_index=$(echo "$body" | jq -r "to_entries[] | select(.value.match[]?.host[]? == \"${fqdn}\") | .key" 2>/dev/null | head -1)
+  route_index=$(echo "$routes_json" | jq -r "to_entries[] | select(.value.match[]?.host[]? == \"${fqdn}\") | .key" 2>/dev/null | head -1)
  if [ -z "$route_index" ] || [ "$route_index" = "null" ]; then
    echo "Warning: route for ${fqdn} not found" >&2
    return 0
  fi
-  # Delete the route at the found index, checking HTTP status
+  # Delete the route at the found index
-  response=$(curl -sS -w '\n%{http_code}' -X DELETE \
+  local response
-    "${CADDY_ADMIN_URL}/config/apps/http/servers/${server_name}/routes/${route_index}" \
+  response=$(curl -s -X DELETE \
-    -H "Content-Type: application/json") || {
+    "${CADDY_ADMIN_URL}/config/apps/http/servers/edge/routes/${route_index}" \
    -H "Content-Type: application/json" 2>&1) || {
    echo "Error: failed to remove route for ${fqdn}" >&2
    echo "Response: ${response}" >&2
    return 1
  }
  status=$(echo "$response" | tail -n1)
  body=$(echo "$response" | sed '$d')
  if [ "$status" -ge 400 ]; then
    echo "Error: Caddy admin API returned ${status}: ${body}" >&2
    return 1
  fi
  echo "Removed route: ${fqdn}" >&2
 }
@ -150,18 +110,13 @@ remove_route() {
 # Reload Caddy to apply configuration changes
 # Usage: reload_caddy
 reload_caddy() {
-  local response status body
+  local response
-  response=$(curl -sS -w '\n%{http_code}' -X POST \
+  response=$(curl -s -X POST \
-    "${CADDY_ADMIN_URL}/reload") || {
+    "${CADDY_ADMIN_URL}/reload" 2>&1) || {
    echo "Error: failed to reload Caddy" >&2
    echo "Response: ${response}" >&2
    return 1
  }
  status=$(echo "$response" | tail -n1)
  body=$(echo "$response" | sed '$d')
  if [ "$status" -ge 400 ]; then
    echo "Error: Caddy reload returned ${status}: ${body}" >&2
    return 1
  fi
  echo "Caddy reloaded" >&2
 }
--- a/tools/edge-control/register.sh
+++ b/tools/edge-control/register.sh
@ -39,11 +39,6 @@ EOF
  exit 1
 }
 # TODO(#713): Subdomain fallback — if subpath routing (#704/#708) fails, this
 # function would need to register additional routes for forge.<project>,
 # ci.<project>, chat.<project> subdomains (or accept a --subdomain parameter).
 # See docs/edge-routing-fallback.md for the full pivot plan.
 # Register a new tunnel
 # Usage: do_register <project> <pubkey>
 do_register() {
--- a/action-vault/SCHEMA.md
+++ b/action-vault/SCHEMA.md
@ -50,7 +50,7 @@ blast_radius = "low"       # optional: overrides policy.toml tier ("low"|"medium
 ## Secret Names
-Secret names must have a corresponding `secrets/<NAME>.enc` file (age-encrypted). The vault validates that requested secrets exist in the allowlist before execution.
+Secret names must be defined in `.env.vault.enc` on the ops repo. The vault validates that requested secrets exist in the allowlist before execution.
 Common secret names:
 - `CLAWHUB_TOKEN` - Token for ClawHub skill publishing
--- a/action-vault/classify.sh
+++ b/action-vault/classify.sh
--- a/action-vault/examples/promote.toml
+++ b/action-vault/examples/promote.toml
--- a/action-vault/examples/publish.toml
+++ b/action-vault/examples/publish.toml
--- a/action-vault/examples/release.toml
+++ b/action-vault/examples/release.toml
--- a/action-vault/examples/webhook-call.toml
+++ b/action-vault/examples/webhook-call.toml
--- a/action-vault/policy.toml
+++ b/action-vault/policy.toml
--- a/action-vault/validate.sh
+++ b/action-vault/validate.sh
--- a/action-vault/vault-env.sh
+++ b/action-vault/vault-env.sh
@ -28,7 +28,7 @@ fi
 # VAULT ACTION VALIDATION
 # =============================================================================
-# Allowed secret names - must match files in secrets/<NAME>.enc
+# Allowed secret names - must match keys in .env.vault.enc
 VAULT_ALLOWED_SECRETS="CLAWHUB_TOKEN GITHUB_TOKEN CODEBERG_TOKEN DEPLOY_KEY NPM_TOKEN DOCKER_HUB_TOKEN"
 # Allowed mount aliases — well-known file-based credential directories