# Action Agent **Role**: Execute operational tasks described by action formulas — run scripts, call APIs, send messages, collect human approval. Shares the same phase handler as the dev-agent: if an action produces code changes, the orchestrator creates a PR and drives the CI/review loop; otherwise Claude closes the issue directly. **Trigger**: `action-poll.sh` runs every 10 min via cron. Sources `lib/guard.sh` and calls `check_active action` first — skips if `$FACTORY_ROOT/state/.action-active` is absent. Then scans for open issues labeled `action` that have no active tmux session, and spawns `action-agent.sh `. **Key files**: - `action/action-poll.sh` — Cron scheduler: finds open action issues with no active tmux session, spawns action-agent.sh - `action/action-agent.sh` — Orchestrator: fetches issue body + prior comments, **checks all dependencies via `lib/parse-deps.sh` before spawning** (skips silently if any dep is still open), creates tmux session (`action-{project}-{issue_num}`) with interactive `claude`, injects formula prompt with phase protocol, enters `monitor_phase_loop` (shared via `dev/phase-handler.sh`) for CI/review lifecycle or direct completion **Session lifecycle**: 1. `action-poll.sh` finds open `action` issues with no active tmux session. 2. Spawns `action-agent.sh `. 3. Agent creates tmux session `action-{project}-{issue_num}`, injects prompt (formula + prior comments + phase protocol). 4. Agent enters `monitor_phase_loop` (shared with dev-agent via `dev/phase-handler.sh`). 5. **Path A (git output):** Claude pushes branch → `PHASE:awaiting_ci` → handler creates PR, polls CI → injects failures → Claude fixes → push → re-poll → CI passes → `PHASE:awaiting_review` → handler polls reviews → injects REQUEST_CHANGES → Claude fixes → approved → merge → cleanup. 6. **Path B (no git output):** Claude posts results as comment, closes issue → `PHASE:done` → handler cleans up (kill session, docker compose down, remove temp files). 7. For human input: Claude writes `PHASE:escalate`; human responds via vault/forge. **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. **Environment variables consumed**: - `FORGE_TOKEN`, `FORGE_ACTION_TOKEN` (falls back to FORGE_TOKEN), `FORGE_REPO`, `FORGE_API`, `FORGE_URL`, `PROJECT_NAME`, `FORGE_WEB` - `ACTION_IDLE_TIMEOUT` — Max seconds before killing idle session (default 14400 = 4h) - `ACTION_MAX_LIFETIME` — Max total session wall-clock seconds (default 28800 = 8h); caps session independently of idle timeout **FORGE_REMOTE**: `action-agent.sh` auto-detects the git remote for `FORGE_URL` (same logic as dev-agent). Exported as `FORGE_REMOTE`, used for worktree creation and push instructions injected into the Claude prompt.