Merge pull request 'chore: gardener housekeeping' (#767) from chore/gardener-20260415-0806 into main

AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Disinto — Agent Instructions
 
 ## What this repo is

architect/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Architect — Agent Instructions
 
 ## What this agent is
@@ -96,6 +96,7 @@ 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

dev/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Dev Agent
 
 **Role**: Implement issues autonomously — write code, push branches, address

gardener/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # 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`, `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).
 - `PRIMARY_BRANCH`, `CLAUDE_MODEL` (set to sonnet by gardener-run.sh)
 
 **Lifecycle**: gardener-run.sh (invoked by polling loop every 6h, `check_active gardener`) →

gardener/pending-actions.json

@@ -1 +1,26 @@
-[]
+[
+  {
+    "action": "edit_body",
+    "issue": 765,
+    "body": "## Problem\nPlanner phase 5 pushes ops repo changes directly to `main` (`planner/AGENTS.md:37`, `planner/planner-run.sh`). Branch protection blocks this — see #758 for the symptom (PR #30 stuck, ops `main` frozen at v0.2.0 since 2026-04-08).\n\n## Why a new issue\n#758 is assigned to dev-qwen and labelled blocked; this reframes the fix rather than the symptom.\n\n## Proposal\nFold planner into the same flow architect already uses for ops PRs (`pr_create` → `pr_walk_to_merge` against `FORGE_OPS_REPO`). Architect proves merge perms work; review-bot already gates ops PRs and decides auto-approve vs request-changes. No new reviewer, no permission changes.\n\n## Changes\n- `planner/planner-run.sh` phase 5: stop direct push; create branch `planner/run-YYYY-MM-DD`, call `pr_create` then `pr_walk_to_merge`.\n- Planner formula prompt: replace \"push directly\" instructions with phase-protocol terminology used by architect.\n- `planner/AGENTS.md`: update phase 5 description.\n\n## Acceptance\n- Planner run produces a PR on ops repo, walks to merged via review-bot.\n- PR #30 closed (superseded) once new flow lands.\n- ops `main` advances past v0.2.0.\n\n## Acceptance criteria\n- [ ] Planner run produces a PR on ops repo, walks to merged via review-bot\n- [ ] PR #30 closed (superseded) once new flow lands\n- [ ] ops `main` advances past v0.2.0\n- [ ] CI green on the planner changes\n\n## Affected files\n- `planner/planner-run.sh` — replace direct push with `pr_create` + `pr_walk_to_merge`\n- `planner/AGENTS.md` — update phase 5 description"
+  },
+  {
+    "action": "edit_body",
+    "issue": 429,
+    "body": "## Problem\n\nThe generated docker-compose.yml uses `build: context: . dockerfile: docker/agents/Dockerfile` which bakes the disinto code into the image via `COPY . /home/agent/disinto`. This causes:\n\n1. **Read-only code** — runtime state (`state/`), config (`projects/*.toml`), and `.env` are not in the image, but the baked-in directory is read-only. Manual volume mount workarounds break on every compose regeneration.\n2. **No versioning** — every `docker compose build` creates a new image from whatever code is on disk. No way to pin a known-good version or roll back.\n3. **No distribution** — new factory instances must clone the disinto repo and build locally. Cannot just `docker pull` and run.\n4. **Fragile rebuilds** — `docker system prune` removes the locally-built image, requiring a full rebuild that may fail (wrong Dockerfile, missing deps, stale cache).\n\n## Proposed solution: publish versioned images\n\nPublish container images to a registry (e.g. `ghcr.io/disinto/agents:v0.1.0`) on each release. The generated compose uses `image:` instead of `build:`.\n\n### Image structure\n\n```\ndisinto-agents:v0.1.0\n  /home/agent/disinto/       # code (immutable, from COPY at build)\n  /home/agent/data/          # VOLUME — runtime state, logs\n  /home/agent/repos/         # VOLUME — project repos\n```\n\n### Runtime mounts (compose volumes)\n\n```yaml\nagents:\n  image: ghcr.io/disinto/agents:v0.1.0\n  volumes:\n    - agent-data:/home/agent/data           # logs, locks, state\n    - project-repos:/home/agent/repos       # cloned project repos\n    - ./projects:/home/agent/disinto/projects:ro  # project TOMLs\n    - ./.env:/home/agent/disinto/.env:ro    # tokens, config\n    - ./state:/home/agent/disinto/state     # agent activation markers\n    - ~/.claude:/home/agent/.claude         # Claude credentials\n    - ~/.claude.json:/home/agent/.claude.json:ro\n    - <claude-cli>:/usr/local/bin/claude:ro\n```\n\n### What changes\n\n- `bin/disinto init` generates compose with `image: ghcr.io/disinto/agents:<version>` instead of `build:`\n- CI pipeline (Woodpecker) builds + pushes images on tag/release\n- `disinto release` updates the image tag in the compose template\n- Same for edge, reproduce, and any other disinto containers\n- `state/` directory must be a writable mount point, not baked into the image\n\n### Images to publish\n\n| Image | Purpose |\n|-------|----------|\n| `disinto/agents` | Dev, review, gardener, planner, predictor, architect agents |\n| `disinto/reproduce` | Reproduce + triage sidecar (Playwright, Docker CLI) |\n| `disinto/edge` | Caddy reverse proxy + dispatcher |\n\n### Backwards compatibility\n\n- `disinto init --build` flag for dev mode (local build, same as today)\n- Default: `image:` from registry\n- Existing deployments: migration guide to switch from build to image\n\n## Files\n\n- `bin/disinto` — `generate_compose()` to emit `image:` instead of `build:`\n- New: CI pipeline for building + pushing images\n- New: `bin/disinto release` updates image tags\n- `docker/agents/Dockerfile` — declare VOLUME mount points explicitly\n- `docker/reproduce/Dockerfile` — same\n- `docker/edge/Dockerfile` — same\n\n## Acceptance criteria\n- [ ] CI pipeline builds and pushes `disinto/agents` image on tag/release\n- [ ] CI pipeline builds and pushes `disinto/reproduce` image on tag/release\n- [ ] CI pipeline builds and pushes `disinto/edge` image on tag/release\n- [ ] `bin/disinto init` generates compose with `image:` instead of `build:`\n- [ ] `bin/disinto init --build` flag enables local build mode for dev\n- [ ] `docker/agents/Dockerfile` declares VOLUME mount points explicitly\n- [ ] `docker/reproduce/Dockerfile` declares VOLUME mount points\n- [ ] `docker/edge/Dockerfile` declares VOLUME mount points\n\n## Affected files\n- `bin/disinto` — `generate_compose()` to emit `image:` instead of `build:`\n- `docker/agents/Dockerfile` — declare VOLUME mount points\n- `docker/reproduce/Dockerfile` — declare VOLUME mount points\n- `docker/edge/Dockerfile` — declare VOLUME mount points\n- `.woodpecker/` — new CI pipeline for building and pushing images"
+  },
+  {
+    "action": "add_label",
+    "issue": 429,
+    "label": "backlog"
+  },
+  {
+    "action": "create_issue",
+    "title": "fix: vault_request RETURN trap fires prematurely when vault-env.sh is sourced",
+    "body": "## Problem\n\n`vault_request()` in `lib/vault.sh` uses `trap ... RETURN` to clean up its temp TOML file. However, when `vault-env.sh` is sourced inside the function (as part of validation), bash fires RETURN traps for each function call made during the source. This causes the temp file to be deleted before `validate_vault_action` reads it.\n\n## Repro\n\n```bash\nsource lib/env.sh\nsource lib/vault.sh\nsource lib/pr-lifecycle.sh\nvault_request \"test-id\" \"id = \\\"test\\\"\\nformula = \\\"run-rent-a-human\\\"\\ncontext = \\\"test\\\"\\nsecrets = []\"\n# => ERROR: File not found: /tmp/vault-XXXX.toml\n# => ERROR: TOML validation failed\n```\n\n## Root cause\n\n```bash\n# In vault_request:\ntmp_toml=$(mktemp /tmp/vault-XXXXXX.toml)\ntrap 'rm -f \"$tmp_toml\"' RETURN   # <-- fires on source, not just on return\n\n# Later:\nsource \"$vault_env\"   # <-- RETURN trap fires here, deleting tmp_toml\nvalidate_vault_action \"$tmp_toml\"   # <-- file is gone\n```\n\n## Fix\n\nUse `EXIT` trap instead of `RETURN`, or set the trap AFTER sourcing vault-env.sh.\n\n```bash\n# Option A: trap on EXIT instead\ntrap 'rm -f \"$tmp_toml\"' EXIT\n\n# Option B: source first, set trap after  \nsource \"$vault_env\"\ntrap 'rm -f \"$tmp_toml\"' RETURN\n```\n\n## Acceptance criteria\n- [ ] `vault_request` successfully validates TOML without \"File not found\" error\n- [ ] Temp file is still cleaned up after function returns\n- [ ] Existing vault test (if any) passes\n\n## Affected files\n- `lib/vault.sh` — fix `trap ... RETURN` in `vault_request()`",
+    "labels": [
+      "backlog",
+      "bug-report"
+    ]
+  }
+]

lib/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # 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 `$FORGE_TOKEN`; handles invalid/empty JSON responses gracefully — returns empty on parse error instead of crashing), `woodpecker_api()`, `wpdb()`, `memory_guard()` (skips agent if RAM < threshold). Auto-loads project TOML if `PROJECT_TOML` is set. Exports per-agent tokens (`FORGE_PLANNER_TOKEN`, `FORGE_GARDENER_TOKEN`, `FORGE_VAULT_TOKEN`, `FORGE_SUPERVISOR_TOKEN`, `FORGE_PREDICTOR_TOKEN`) — each falls back to `$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/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. **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). | 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 |

planner/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Planner Agent
 
 **Role**: Strategic planning using a Prerequisite Tree (Theory of Constraints),

predictor/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Predictor Agent
 
 **Role**: Abstract adversary (the "goblin"). Runs a 2-step formula

review/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Review Agent
 
 **Role**: AI-powered PR review — post structured findings and formal

supervisor/AGENTS.md

@@ -1,4 +1,4 @@
-<!-- last-reviewed: cdb0408466d875cdd893c712e3b3f0d0b909d4ef -->
+<!-- last-reviewed: 10c7a88416b14e849f80ad3fe7ea8e51d26177e8 -->
 # Supervisor Agent
 
 **Role**: Health monitoring and auto-remediation, executed as a formula-driven