cancel

Crew Cancel

Mark a crew run as cancelled. Does NOT kill running processes (the orchestrator is per-invocation, not a daemon) — it sets the on-disk status so subsequent runs/lists treat it as cancelled.

RUN_ID="$ARGUMENTS"

If $ARGUMENTS is empty, list active runs (status not in {completed, cancelled*}) and ask which to cancel via AskUserQuestion.

Steps

1. Verify the run exists at .crew/runs/$RUN_ID/state.json.

2. Read current status. If already completed or cancelled*, tell the user "nothing to do" and stop.

3. Confirm with user via AskUserQuestion:

   • Cancel — mark cancelled, leave artifacts for inspection
   • Cancel and delete — mark cancelled, remove the run dir (only the run dir, never anything outside)
   • Never mind — leave alone

4. On Cancel: update state.json:

{
  ...existing fields...,
  "status": "cancelled",
  "cancelled_at": "<ISO timestamp>",
  "cancelled_at_phase": "<previous phase>"
}

1. On Cancel and delete: do the state.json update first, then remove the dir:

rm -rf ".crew/runs/$RUN_ID"

(This is the only place in the plugin where the orchestrator itself does a destructive op — and it's gated by an explicit user confirmation.)

1. Confirm to user:

Cancelled <RUN_ID>. <Artifacts at .crew/runs/<RUN_ID>/ | Artifacts removed.>

Rules

• Confirm before destructive action. Always.
• Never touch anything outside .crew/runs/<run-id>/.
• A cancelled run can be inspected but not resumed (use /crew:do to start fresh).