8 changed files with 31 additions and 519 deletions
--- a/.env.example
+++ b/.env.example
@ -32,13 +32,10 @@ FORGE_URL=http://localhost:3000             # [CONFIG] local Forgejo instance
 #   - FORGE_PASS_DEV_QWEN2
 # Name conversion: tr 'a-z-' 'A-Z_' (lowercase→UPPER, hyphens→underscores).
 # The compose generator looks these up via the agent's `forge_user` field in
-# the project TOML. The pre-existing `dev-qwen` llama agent uses
-# FORGE_TOKEN_LLAMA / FORGE_PASS_LLAMA (kept for backwards-compat with the
-# legacy `ENABLE_LLAMA_AGENT=1` single-agent path).
+# the project TOML. Configure local-model agents via [agents.X] sections in
+# projects/*.toml — this is the canonical activation path.
 FORGE_TOKEN=                               # [SECRET] dev-bot API token (default for all agents)
 FORGE_PASS=                                # [SECRET] dev-bot password for git HTTP push (#361)
-FORGE_TOKEN_LLAMA=                         # [SECRET] dev-qwen API token (for agents-llama)
-FORGE_PASS_LLAMA=                          # [SECRET] dev-qwen password for git HTTP push
 FORGE_REVIEW_TOKEN=                        # [SECRET] review-bot API token
 FORGE_REVIEW_PASS=                         # [SECRET] review-bot password for git HTTP push
 FORGE_PLANNER_TOKEN=                       # [SECRET] planner-bot API token
@ -62,10 +59,6 @@ FORGE_BOT_USERNAMES=dev-bot,review-bot,planner-bot,gardener-bot,vault-bot,superv
 # CODEBERG_TOKEN automatically (same for REVIEW_BOT_TOKEN, CODEBERG_REPO,
 # CODEBERG_BOT_USERNAMES). No action needed for existing deployments.
 # Per-agent tokens default to FORGE_TOKEN when unset (single-token setups).
-#
-# Note: `disinto init` auto-generates all bot tokens/passwords when you
-# configure [agents.llama] in a project TOML. The credentials are stored
-# in .env.enc (encrypted) or .env (plaintext fallback).

 # ── Woodpecker CI ─────────────────────────────────────────────────────────
 WOODPECKER_TOKEN=                          # [SECRET] Woodpecker API token
@ -107,13 +100,6 @@ FORWARD_AUTH_SECRET=                      # [SECRET] Shared secret for Caddy ↔
 # 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

--- a/AGENTS.md
+++ b/AGENTS.md
@ -120,8 +120,6 @@ bash dev/phase-test.sh
 | 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.
--- a/bin/disinto
+++ b/bin/disinto
@ -396,15 +396,16 @@ check_pipeline_stall = false
 # for local LLM inference. Each agent gets its own container with isolated
 # credentials and configuration.
 #
-# [agents.llama]
-#   base_url = "http://10.10.10.1:8081"
-#   model = "unsloth/Qwen3.5-35B-A3B"
-#   api_key = "sk-no-key-required"
-#   roles = ["dev"]
-#   forge_user = "dev-qwen"
-#   compact_pct = 60
-#   poll_interval = 60
-
+# Configure via [agents.X] sections:
+#   [agents.dev-qwen]
+#     base_url = "http://10.10.10.1:8081"
+#     model = "unsloth/Qwen3.5-35B-A3B"
+#     api_key = "sk-no-key-required"
+#     roles = ["dev"]
+#     forge_user = "dev-qwen"
+#     compact_pct = 60
+#     poll_interval = 60
+#
 # [mirrors]
 # github   = "git@github.com:user/repo.git"
 # codeberg = "git@codeberg.org:user/repo.git"
@ -882,7 +883,6 @@ p.write_text(text)
    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"
@ -1078,19 +1078,6 @@ 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"

--- a/docker-compose.yml
+++ b/docker-compose.yml
@ -64,134 +64,6 @@ services:
    networks:
      - disinto-net

-  agents-llama:
-    build:
-      context: .
-      dockerfile: docker/agents/Dockerfile
-    image: disinto/agents-llama:latest
-    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=${FORGE_PASS_LLAMA:-}
-      - FORGE_SUPERVISOR_TOKEN=${FORGE_SUPERVISOR_TOKEN:-}
-      - FORGE_PREDICTOR_TOKEN=${FORGE_PREDICTOR_TOKEN:-}
-      - FORGE_ARCHITECT_TOKEN=${FORGE_ARCHITECT_TOKEN:-}
-      - FORGE_VAULT_TOKEN=${FORGE_VAULT_TOKEN:-}
-      - FORGE_PLANNER_TOKEN=${FORGE_PLANNER_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}
-      - 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
-      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
-
  reproduce:
    build:
      context: .
--- 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/lib/AGENTS.md
+++ b/lib/AGENTS.md
@ -30,7 +30,7 @@ sourced as needed.
 | `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/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 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); local-model agents via TOML [agents.X] sections (canonical path, #846); 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/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) |
--- a/lib/forge-setup.sh
+++ b/lib/forge-setup.sh
@ -356,16 +356,6 @@ setup_forge() {
    [predictor-bot]="FORGE_PREDICTOR_PASS"
    [architect-bot]="FORGE_ARCHITECT_PASS"
  )
-  # Llama bot users (local-model agents) — separate from main agents
-  # Each llama agent gets its own Forgejo user, token, and password
-  local -A llama_token_vars=(
-    [dev-qwen]="FORGE_TOKEN_LLAMA"
-    [dev-qwen-nightly]="FORGE_TOKEN_LLAMA_NIGHTLY"
-  )
-  local -A llama_pass_vars=(
-    [dev-qwen]="FORGE_PASS_LLAMA"
-    [dev-qwen-nightly]="FORGE_PASS_LLAMA_NIGHTLY"
-  )

  local bot_user bot_pass token token_var pass_var

@ -515,159 +505,12 @@ setup_forge() {
    fi
  done

-  # Create llama bot users and tokens (local-model agents)
-  # These are separate from the main agents and get their own credentials
-  echo ""
-  echo "── Setting up llama bot users ────────────────────────────"
-
-  local llama_user llama_pass llama_token llama_token_var llama_pass_var
-  for llama_user in "${!llama_token_vars[@]}"; do
-    llama_token_var="${llama_token_vars[$llama_user]}"
-    llama_pass_var="${llama_pass_vars[$llama_user]}"
-
-    # Check if token already exists in .env
-    local token_exists=false
-    if _token_exists_in_env "$llama_token_var" "$env_file"; then
-      token_exists=true
-    fi
-
-    # Check if password already exists in .env
-    local pass_exists=false
-    if _pass_exists_in_env "$llama_pass_var" "$env_file"; then
-      pass_exists=true
-    fi
-
-    # Check if llama bot user exists on Forgejo
-    local llama_user_exists=false
-    if curl -sf --max-time 5 \
-      -H "Authorization: token ${admin_token}" \
-      "${forge_url}/api/v1/users/${llama_user}" >/dev/null 2>&1; then
-      llama_user_exists=true
-    fi
-
-    # Skip token/password regeneration if both exist in .env and not forcing rotation
-    if [ "$token_exists" = true ] && [ "$pass_exists" = true ] && [ "$rotate_tokens" = false ]; then
-      echo "  ${llama_user} token and password preserved (use --rotate-tokens to force)"
-      # Still export the existing token for use within this run
-      local existing_token existing_pass
-      existing_token=$(grep "^${llama_token_var}=" "$env_file" | head -1 | cut -d= -f2-)
-      existing_pass=$(grep "^${llama_pass_var}=" "$env_file" | head -1 | cut -d= -f2-)
-      export "${llama_token_var}=${existing_token}"
-      export "${llama_pass_var}=${existing_pass}"
-      continue
-    fi
-
-    # Generate new credentials if:
-    # - Token doesn't exist (first run)
-    # - Password doesn't exist (first run)
-    # - --rotate-tokens flag is set (explicit rotation)
-    if [ "$llama_user_exists" = false ]; then
-      # User doesn't exist - create it
-      llama_pass="llama-$(head -c 16 /dev/urandom | base64 | tr -dc 'a-zA-Z0-9' | head -c 20)"
-      echo "Creating llama bot user: ${llama_user}"
-      local create_output
-      if ! create_output=$(_forgejo_exec forgejo admin user create \
-        --username "${llama_user}" \
-        --password "${llama_pass}" \
-        --email "${llama_user}@disinto.local" \
-        --must-change-password=false 2>&1); then
-        echo "Error: failed to create llama bot user '${llama_user}':" >&2
-        echo "  ${create_output}" >&2
-        exit 1
-      fi
-      # Forgejo 11.x ignores --must-change-password=false on create;
-      # explicitly clear the flag so basic-auth token creation works.
-      _forgejo_exec forgejo admin user change-password \
-        --username "${llama_user}" \
-        --password "${llama_pass}" \
-        --must-change-password=false
-
-      # Verify llama bot user was actually created
-      if ! curl -sf --max-time 5 \
-        -H "Authorization: token ${admin_token}" \
-        "${forge_url}/api/v1/users/${llama_user}" >/dev/null 2>&1; then
-        echo "Error: llama bot user '${llama_user}' not found after creation" >&2
-        exit 1
-      fi
-      echo "  ${llama_user} user created"
-    else
-      # User exists - reset password if needed
-      echo "  ${llama_user} user exists"
-      if [ "$rotate_tokens" = true ] || [ "$pass_exists" = false ]; then
-        llama_pass="llama-$(head -c 16 /dev/urandom | base64 | tr -dc 'a-zA-Z0-9' | head -c 20)"
-        _forgejo_exec forgejo admin user change-password \
-          --username "${llama_user}" \
-          --password "${llama_pass}" \
-          --must-change-password=false || {
-          echo "Error: failed to reset password for existing llama bot user '${llama_user}'" >&2
-          exit 1
-        }
-        echo "  ${llama_user} password reset for token generation"
-      else
-        # Password exists, get it from .env
-        llama_pass=$(grep "^${llama_pass_var}=" "$env_file" | head -1 | cut -d= -f2-)
-      fi
-    fi
-
-    # Generate token via API (basic auth as the llama user)
-    # First, delete any existing tokens to avoid name collision
-    local existing_llama_token_ids
-    existing_llama_token_ids=$(curl -sf \
-      -u "${llama_user}:${llama_pass}" \
-      "${forge_url}/api/v1/users/${llama_user}/tokens" 2>/dev/null \
-      | jq -r '.[].id // empty' 2>/dev/null) || existing_llama_token_ids=""
-
-    # Delete any existing tokens for this user
-    if [ -n "$existing_llama_token_ids" ]; then
-      while IFS= read -r tid; do
-        [ -n "$tid" ] && curl -sf -X DELETE \
-          -u "${llama_user}:${llama_pass}" \
-          "${forge_url}/api/v1/users/${llama_user}/tokens/${tid}" >/dev/null 2>&1 || true
-      done <<< "$existing_llama_token_ids"
-    fi
-
-    llama_token=$(curl -sf -X POST \
-      -u "${llama_user}:${llama_pass}" \
-      -H "Content-Type: application/json" \
-      "${forge_url}/api/v1/users/${llama_user}/tokens" \
-      -d "{\"name\":\"disinto-${llama_user}-token\",\"scopes\":[\"all\"]}" 2>/dev/null \
-      | jq -r '.sha1 // empty') || llama_token=""
-
-    if [ -z "$llama_token" ]; then
-      echo "Error: failed to create API token for '${llama_user}'" >&2
-      exit 1
-    fi
-
-    # Store token in .env under the llama-specific variable name
-    if grep -q "^${llama_token_var}=" "$env_file" 2>/dev/null; then
-      sed -i "s|^${llama_token_var}=.*|${llama_token_var}=${llama_token}|" "$env_file"
-    else
-      printf '%s=%s\n' "$llama_token_var" "$llama_token" >> "$env_file"
-    fi
-    export "${llama_token_var}=${llama_token}"
-    echo "  ${llama_user} token generated and saved (${llama_token_var})"
-
-    # Store password in .env for git HTTP push (#361)
-    # Forgejo 11.x API tokens don't work for git push; password auth does.
-    if grep -q "^${llama_pass_var}=" "$env_file" 2>/dev/null; then
-      sed -i "s|^${llama_pass_var}=.*|${llama_pass_var}=${llama_pass}|" "$env_file"
-    else
-      printf '%s=%s\n' "$llama_pass_var" "$llama_pass" >> "$env_file"
-    fi
-    export "${llama_pass_var}=${llama_pass}"
-    echo "  ${llama_user} password saved (${llama_pass_var})"
-  done
-
  # Create .profile repos for all bot users (if they don't already exist)
  # This runs the same logic as hire-an-agent Step 2-3 for idempotent setup
  echo ""
  echo "── Setting up .profile repos ────────────────────────────"

  local -a bot_users=(dev-bot review-bot planner-bot gardener-bot vault-bot supervisor-bot predictor-bot architect-bot)
-  # Add llama bot users to .profile repo creation
-  for llama_user in "${!llama_token_vars[@]}"; do
-    bot_users+=("$llama_user")
-  done
  local bot_user

  for bot_user in "${bot_users[@]}"; do
@ -775,15 +618,6 @@ setup_forge() {
        -d "{\"permission\":\"${bot_perm}\"}" >/dev/null 2>&1 || true
    done

-    # Add llama bot users as write collaborators for local-model agents
-    for llama_user in "${!llama_token_vars[@]}"; do
-      curl -sf -X PUT \
-        -H "Authorization: token ${admin_token:-${FORGE_TOKEN}}" \
-        -H "Content-Type: application/json" \
-        "${forge_url}/api/v1/repos/${repo_slug}/collaborators/${llama_user}" \
-        -d '{"permission":"write"}' >/dev/null 2>&1 || true
-    done
-
    # Add disinto-admin as admin collaborator
    curl -sf -X PUT \
      -H "Authorization: token ${admin_token:-${FORGE_TOKEN}}" \
--- a/lib/generators.sh
+++ b/lib/generators.sh
@ -68,6 +68,10 @@ _get_primary_woodpecker_repo_id() {

 # Parse project TOML for local-model agents and emit compose services.
 # Writes service definitions to stdout; caller handles insertion into compose file.
+#
+# This is the canonical activation path for llama-backed agents.
+# The legacy ENABLE_LLAMA_AGENT env flag is removed — its presence in .env
+# triggers a migration error at disinto up time.
 _generate_local_model_services() {
  local compose_file="$1"
  local projects_dir="${FACTORY_ROOT}/projects"
@ -248,6 +252,10 @@ for name, config in agents.items():
 # **CANONICAL SOURCE**: This generator is the single source of truth for docker-compose.yml.
 # The tracked docker-compose.yml file has been removed. Operators must run 'bin/disinto init'
 # to materialize a working stack on a fresh checkout.
+#
+# Local-model agents are configured via [agents.X] TOML sections in projects/*.toml.
+# The legacy ENABLE_LLAMA_AGENT flag is removed — its presence in .env triggers
+# a migration error at compose generation time.
 _generate_compose_impl() {
  local forge_port="${1:-3000}"
  local use_build="${2:-false}"
@ -415,130 +423,6 @@ services:

 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'

@ -712,6 +596,16 @@ networks:
    driver: bridge
 COMPOSEEOF

+  # ── Check for legacy ENABLE_LLAMA_AGENT flag ─────────────────────────
+  # The TOML [agents.X] path is canonical. If ENABLE_LLAMA_AGENT=1 is found
+  # in .env, fail with a migration message.
+  local env_file="${FACTORY_ROOT}/.env"
+  if [ -f "$env_file" ] && grep -q '^ENABLE_LLAMA_AGENT=1$' "$env_file" 2>/dev/null; then
+    echo "Error: ENABLE_LLAMA_AGENT=1 is no longer supported" >&2
+    echo "  Use the [agents.X] TOML path in projects/*.toml instead." >&2
+    exit 1
+  fi
+
  # Patch PROJECT_REPO_ROOT — interpolate PROJECT_NAME at generation time
  # (Docker Compose cannot resolve it; it's a shell variable, not a .env var)
  sed -i "s|\${PROJECT_NAME:-project}|${PROJECT_NAME}|g" "$compose_file"