Skip to main content

status (Status Update and Recovery)

Inspect conversation history, active processes, and live repository/forge state to deliver an authentic status update. If the agent became stuck, frozen, or dropped the ball in previous turns, diagnose the stall and resume the dropped work immediately.

When this fires

  • “status”, “status update”, “progress update”, “what’s the status”, “where are we”
  • “are you stuck”, “are you frozen”, “did you drop the ball”, “what happened”
  • Any general inquiry about the state of ongoing work or current turn progress

Procedure

1. Inspect the transcript for stalls and dropped balls

Before polling forge or CI state, review recent session turns in the conversation transcript (transcript.jsonl or conversational turn history): 1. Unhandled tool failure: Did a command, API call, or script return an error, auth failure, rate limit, or denial that was never diagnosed or retried? 2. Abandoned or forgotten subagent/task: Was a subagent dispatched or background job started that completed or hung without the conductor processing the result? 3. Unarmed pause: Did the agent end its previous turn without arming a wake mechanism (timer or monitor) while work remained queued? 4. Unfulfilled commitment: Did the agent announce an action (“I’ll start on X”, “I will push after review”) but stop without doing it? 5. Frozen or silent stall: Did execution pause mid-stride without completing the planned task?

If any dropped ball or stall occurred: - State what failed or where execution paused. - Resume immediately: In this very same turn, re-run the failed step, resolve the error, or continue the interrupted workflow. Never make status report-only when work was dropped.

2. Inspect live workspace and git state

Run local git checks:

git status --short
git log -n 3 --oneline
git worktree list

Verify whether any uncommitted edits, unpushed commits, or detached worktrees exist.

3. Inspect live forge and CI state

If working on a PR or issue: - Check if the PR is still open:

gh pr view <N> --json state,title,headRefOid
  • Check CI status:
gh pr checks <N>
  • Parse the latest review comments on HEAD to ensure no unresolved findings remain.

4. Advance work and arm a wake mechanism if waiting

  • If concrete actions are ready (tests to run, review findings to fix, branch to push, PR to merge), execute them in the same turn.
  • If legitimately waiting for external automation (e.g. CI running, remote review in progress), ensure an active wake timer or monitor is armed with schedule.

5. Report concise, dated status

Conclude with: - What was diagnosed and resumed (if a stall occurred). - Current live state (PR#, CI state, review verdict). - Clock time in Pacific Time (TZ=America/Los_Angeles date "+%Y-%m-%d %H:%M %Z"). - Explicit **Stopping Point** declaration stating whether the session is done or ongoing.

  • Do: check the transcript for unhandled errors, forgotten subagents, or unarmed pauses before reporting status.
  • Do: resume stalled or dropped work immediately in the same turn.
  • Do: report fresh live state rather than echoing stale cached facts.
  • Don’t: treat a status request as report-only when the agent previously froze or dropped the ball.
  • Don’t: ask the user for permission to resume work that was already queued or promised.
Back to top