Closes #5439

Summary

Adds native cron job output chaining via context_from field. When set, the most recent output of the referenced job(s) is automatically injected into the prompt before each run, enabling job A to collect data and job B to process it without manual file wiring.

Files changed

• cron/jobs.py — added context_from parameter to create_job(), with normalization to a list
• cron/scheduler.py — output injection in _build_job_prompt()
• tools/cronjob_tools.py — schema, validation, handler wiring
• tests/cron/test_cron_context_from.py — 14 new tests

Implementation

Issue #5439 Option C proposed a new ~/.hermes/cron/shared/ directory with named files. Instead, the implementation reads from the existing ~/.hermes/cron/output/{job_id}/ store that save_job_output() already writes atomically.

This avoids introducing a new filesystem surface while delivering stronger guarantees:

• Per-job isolation, no naming collisions between jobs
• Atomic writes already in place (temp → fsync → os.replace())
• Automatic latest semantics via mtime sorting
• No new cleanup logic needed

Safety:

• Path traversal guard: job IDs validated as 12-char hex strings
• Truncation at 8K chars to prevent prompt bloat
• Graceful degradation on OSError (covers PermissionError and FileNotFoundError)

Usage:

# Job A collects data
cronjob(action='create', prompt='Find top news', schedule='every 1h', name='news_finder')
# Returns job_id: abc123def456

# Job B processes it
cronjob(action='create', prompt='Summarize the news', schedule='every 2h',
        context_from=['abc123def456'])

Tests

Tested on

macOS, Linux, Python 3.11, Hermes v0.11.0