wake_lifecycle_emitter: Arm directive never emits — _extract_task_id misses production tool_response shape

[michael-wojcik]

Summary

pact-plugin/hooks/wake_lifecycle_emitter.py::_extract_task_id does not match the actual platform PostToolUse payload shape for TaskCreate. The hook silently no-ops on every first-active-task transition, so the auto-Arm flow shipped in #603 has never fired in production. Manual /PACT:watch-inbox and the SessionStart-resume path still work; only the hook-driven 0→1 Arm is broken.

Empirical Evidence

Captured live PostToolUse stdin in session pact-56ce3a2a on 2026-05-02 by replacing the hook with a logging shim and triggering a real TaskCreate:

{
  "tool_name": "TaskCreate",
  "tool_input": {"subject": "...", "description": "..."},
  "tool_response": {"task": {"id": "5", "subject": "..."}}
}

tool_response.task.id — id is nested under task.

Bug

_extract_task_id (lines ~145-167) probes:

• tool_input.taskId / tool_input.task_id — TaskCreate has neither (id is platform-assigned)
• tool_response.id / tool_response.taskId / tool_response.task_id — production puts it at tool_response.task.id

Result: returns None → _decide_directive exits at the if not _extract_task_id(...): return None guard → main() prints {\"suppressOutput\": true}. Never emits the Arm directive on TaskCreate.

TaskUpdate is unaffected — its production tool_input has taskId directly, so the Teardown path works.

Why Tests Pass

pact-plugin/tests/test_inbox_wake_lifecycle_emitter.py fixture uses:

\"tool_input\": {\"taskId\": \"task-1\"},
\"tool_response\": {\"id\": \"task-1\"},

This shape is incorrect for TaskCreate. Tests are green; production silently fails. See companion issue for the test-fixture-parity problem.

Proposed Fix

Extend _extract_task_id to also probe nested shapes:

tool_response = input_data.get(\"tool_response\") or {}
if isinstance(tool_response, dict):
    # nested-task shape (TaskCreate production)
    task = tool_response.get(\"task\")
    if isinstance(task, dict):
        tid = task.get(\"id\") or task.get(\"taskId\") or task.get(\"task_id\")
        if isinstance(tid, str) and tid:
            return tid
    # flat shape (TaskUpdate production + legacy fixtures)
    tid = tool_response.get(\"id\") or tool_response.get(\"taskId\") or tool_response.get(\"task_id\")
    if isinstance(tid, str) and tid:
        return tid

Add a regression test that pipes the captured production payload (tool_response: {\"task\": {\"id\": \"...\"}}) end-to-end and asserts additionalContext is emitted on the 0→1 transition.

Impact

• PR Add inbox-wake skill (lead-only) + 20/20/60 quiet-window + PostToolUse lifecycle #603 is incomplete as merged. The hook-driven Arm path has never fired since merge.
• Manual /PACT:watch-inbox invocation remains the recovery path.
• Lead sessions that don't manually arm and don't resume into existing tasks have no inbox-wake Monitor — the auto-arm has been silently degraded to no-op.

References

• Surfaced 2026-05-02 in dogfood orchestration session pact-56ce3a2a while running watch-inbox: tighten no-narration discipline to explicitly require truly empty assistant turn on bare Monitor wake #610.
• Companion issue: test-fixture-vs-production-shape parity.