What & why

Closes #41923.

When a cron tick is short-circuited before the LLM is ever invoked, the scheduler logs a misleading line claiming the agent ran:

INFO cron.scheduler: Job '…': wakeAgent=false, skipping agent run
INFO cron.scheduler: Job '…': agent returned [SILENT] — skipping delivery   ← wrong, agent never ran

The root cause: four pre-agent skip paths in run_job all returned the same SILENT_MARKER that a genuine agent-produced [SILENT] reply uses, so the delivery loop in _process_job couldn't tell the two apart and always logged agent returned [SILENT].

The skip paths are:

• wakeAgent=false script gate (LLM path) — the case in the issue
• wakeAgent=false gate on a no_agent job
• empty no_agent script stdout
• a prompt that resolves to nothing (script produced no output)

In all four the agent is never invoked.

Change

• Add a distinct AGENT_NOT_INVOKED_MARKER sentinel returned by those four skip paths. It suppresses delivery exactly like SILENT_MARKER (output is still saved for audit, the run still counts as a success), but lets the delivery loop log:

  INFO cron.scheduler: Job '…': agent not invoked — skipping delivery
  

• The genuine "agent ran and replied [SILENT]" path is untouched and keeps its existing agent returned [SILENT] — skipping delivery log line. The two markers are mutually exclusive ([SILENT] is not a substring of [SILENT:agent-not-invoked]), so the substring [SILENT] detection for real agent replies still works.

This keeps the per-cause line already emitted inside run_job (wakeAgent=false, skipping agent run, empty stdout — silent run, etc.) as the authoritative reason, and stops the second line from contradicting it.

How to test

Repro from the issue: a cron job whose pre-run script emits {"wakeAgent": false}. Before this change the scheduler logged agent returned [SILENT] — skipping delivery; now it logs agent not invoked — skipping delivery.

Automated:

pytest tests/cron/ -q

• Updated test_run_job_no_agent_empty_output_is_silent, test_run_job_no_agent_wake_gate_is_silent, and test_wake_false_skips_agent_and_returns_silent to assert the new sentinel.
• Added test_agent_not_invoked_marker_logs_distinct_message asserting the skip path logs agent not invoked and not agent returned, while still suppressing delivery.

All 404 tests in tests/cron/ pass.

Platforms

Logic-only change in cron/scheduler.py; no platform-specific code touched. Tested on macOS.