Summary

Adds a lightweight trigger system for cron jobs — a $0 pre-run gate that evaluates a condition before spawning an AIAgent. If the trigger returns false, the job is skipped and rescheduled. No LLM call, no tokens burned.

Problem: Cron jobs run on fixed schedules regardless of whether there's new data to process. A job scheduled every 30m fires 48 times/day even if only 2 of those runs have meaningful work. The other 46 runs waste inference costs.

Solution: An optional trigger field on cron jobs that supports three evaluation types:

Examples

# Only summarize when there are new conversations
cronjob(action="create",
  prompt="Summarize new conversations",
  schedule="every 2h",
  trigger={"type": "sql",
           "query": "SELECT COUNT(*) FROM messages WHERE timestamp > strftime('%s','now','-2 hours')",
           "threshold": 1})

# Only process when feed file has been updated
cronjob(action="create",
  prompt="Process updated data feed",
  schedule="every 30m",
  trigger={"type": "file_changed", "path": "~/data/feed.json"})

# Only run when an external flag is set
cronjob(action="create",
  prompt="Run nightly analysis",
  schedule="0 9 * * *",
  trigger={"type": "command", "command": "test -f /tmp/new-data-ready"})

Design decisions

• Fail-closed: Trigger errors (SQL exceptions, missing files, timeouts) skip the job rather than running it. You never accidentally burn tokens on a broken trigger.
• Read-only SQL: Mutation queries (INSERT/UPDATE/DELETE/DROP/ALTER/CREATE) are blocked.
• 10s timeout: Command triggers have a hard 10-second timeout to prevent blocking the scheduler.
• Backward compatible: Jobs without a trigger field run exactly as before. Existing jobs, configs, and the [SILENT] mechanism are untouched.
• Minimal surface: 3 files changed + 1 test file. The trigger check is entirely contained in the cron subsystem — AIAgent, model_tools, gateway, and config are untouched.

Files changed

• cron/scheduler.py — check_trigger() function + 3 type-specific evaluators, wired into tick() before run_job()
• cron/jobs.py — trigger param on create_job(), persisted in job dict
• tools/cronjob_tools.py — trigger param on cronjob() tool + schema update
• tests/cron/test_triggers.py — 18 tests covering all trigger types, edge cases, and failure modes

Test plan

• 18 unit tests passing (pytest tests/cron/test_triggers.py)
• Manual test: create a cron job with SQL trigger, verify it skips when query returns 0
• Manual test: create a cron job with file_changed trigger, verify it skips when file is stale
• Manual test: verify existing cron jobs without triggers still run normally