fix(cron): clean up output dir when auto-removing repeat-limited job

[nftpoetrist]

What does this PR do?

mark_job_run() removes a job from jobs.json when its repeat limit is reached, but the corresponding OUTPUT_DIR/<job_id> directory (written by save_job_output() on every run) was never deleted. Each completed one-shot job leaves behind an orphaned output directory that accumulates indefinitely.

remove_job() gained the identical cleanup in PR #21882 (merged today). This PR extends that fix to mark_job_run(), the only other code path that deletes a job — parity with remove_job() lines 700–703.

One-line fix in mark_job_run(), scoped to the repeat-limit branch. No behavior change for recurring jobs or jobs without a repeat limit.

Related Issue

Fixes #22065

Type of Change

• 🐛 Bug fix
• ✨ New feature
• 🔒 Security fix
• 📝 Documentation update
• ✅ Tests
• ♻️ Refactor
• 🎯 New skill

Changes Made

• cron/jobs.py: add shutil.rmtree(OUTPUT_DIR / job_id) guard in the repeat-limit auto-delete branch of mark_job_run() (+4 lines)
• tests/cron/test_jobs.py: two new tests in TestMarkJobRun (+33 lines)
  • test_repeat_limit_cleans_up_output_dir — verifies the dir is removed after auto-delete
  • test_repeat_limit_no_output_dir_is_safe — verifies no crash when no output was saved

How to Test

python3.11 -m pytest tests/cron/test_jobs.py::TestMarkJobRun -v --override-ini="addopts="

All 12 tests pass (1 skip for missing croniter — pre-existing on macOS).

Checklist

Code

• Contributing Guide read
• Conventional Commits
• No duplicate PR
• This fix only
• pytest run
• Tests added
• Platform: macOS

Documentation & Housekeeping

• Docs updated — N/A
• cli-config.yaml.example — N/A
• CONTRIBUTING.md/AGENTS.md — N/A
• Cross-platform impact — N/A (shutil.rmtree is cross-platform)
• Tool descriptions — N/A

[teknium1]

Thanks for identifying the orphaned-output path. The premise is still valid on current main: cron/scheduler.py:3427 saves output before cron/scheduler.py:3484 calls mark_job_run(), while the repeat-limit branch at cron/jobs.py:1426-1430 removes only the job record.

Problems

• The proposed OUTPUT_DIR / job_id path bypasses current path validation. cron/jobs.py:255-268 defines _job_output_dir() to reject escape attempts, and remove_job() resolves it before saving at cron/jobs.py:1363-1370. Use that same ordering in the repeat-limit branch.
• Current main has two additional exhausted-one-shot removal paths: claim_dispatch() at cron/jobs.py:1501-1506 and _get_due_jobs_locked() at cron/jobs.py:1818-1830. Both can leave output behind, so fixing only mark_job_run() does not cover the current bug class.

Suggested changes

• Resolve the safe output path before removing/persisting each job, then delete it after the store update.
• Add seeded-output regression tests for the repeat-limit, dispatch-limit, and stale-due removal paths.

Automated hermes-sweeper review.