stop-hook-needs-agent-stop-intent-detection-not-response-length
The Stop hook script should detect whether the agent actually initiated a stop action, not infer it from response brevity. The current implementation fires on all stop events regardless of cause. The fix is to add a guard in the hook script: if the session is in an active-working state (e.g., TodoWrite tasks in_progress, pending user input), skip the verification gate entirely. Only evaluate completion criteria when the agent has completed its last task and is genuinely finishing.