fix(cron): one-shot jobs report 'failed' after successful completion

[natehale]

Symptom

hermes cron run <job-id> on a one-shot job (repeat=1) prints Ran now: failed even though:

• Agent completed and produced correct output
• Output file was written to ~/.hermes/cron/output/
• Scheduler logged Job '...' completed successfully

Root cause

mark_job_run() in cron/jobs.py removes one-shot jobs from jobs.json when the repeat limit is reached (line ~1544). It sets last_status="ok" on the in-memory dict first, but that dict is then popped from the list and discarded — the job no longer exists on disk.

_execute_job_now() in tools/cronjob_tools.py then reads the job back via get_job(job_id) to check last_status. For one-shot jobs this returns None, so ok = False and execution_success is incorrectly set to False.

mark_job_run(success=True)
  → sets last_status="ok" in memory
  → completed >= times → jobs.pop(i) → save_jobs  ← job deleted
  → return

_execute_job_now:
  → get_job(job_id) → None  (job removed)
  → ok = ({}).get("last_status") == "ok" → False
  → returns success=False  ← WRONG

Fix

In _execute_job_now, after run_one_job returns True, treat a missing job as one-shot completion rather than failure:

refreshed = get_job(job_id)
if refreshed is None and processed:
    # One-shot auto-removed after success — absence = completion
    return {"claimed": True, "success": True, "error": None}
_r = refreshed or {}

Also uses the _r fallback dict consistently for last_error access (defensive against a None refreshed in the error path).

Scope

• 1 file changed (tools/cronjob_tools.py, 6 lines net)
• No schema changes, no new dependencies
• Does not affect recurring jobs (they still exist after mark_job_run)
• Does not affect delivery semantics

Tests

• 5 new tests in tests/tools/test_oneshot_status_fix.py covering:
  • One-shot removed after success → success=True
  • One-shot agent failure → success=False
  • Recurring success → success=True (unchanged)
  • Processing failure + job gone → success=False (unchanged)
  • Claim failed → claimed=False (unchanged)
• All existing test_cronjob_run_immediate tests pass (5/5)

Edge cases

Scenario	processed	get_job	Result
One-shot success	True	None (removed)	success=True (fix)
One-shot agent failure	True	exists, status=error	success=False
Recurring success	True	exists, status=ok	success=True
Processing failure	False	None	success=False

[natehale]

Closing as duplicate of #57689 — same root cause (mark_job_run removes one-shot jobs before _execute_job_now reads back last_status), and their approach is more complete: capturing structured success/error/delivery_error at the source via instead of inferring from absence.

Our alternative fix (inferring success when get_job returns None and processed=True) is simpler but less robust — it can't surface the actual error or delivery_error. #57689 is the better vehicle for this fix.