Cron

cron — scheduled jobs

Jobs fire a prompt into the live session while pi is running. Past-due jobs catch up once on the next tick (~30s). Storage: <agentDir>/cron/jobs.json.

Creating jobs

1. Convert the user's natural language to a 5-field cron expression.
2. Verify it with cron action:"test" schedule:"..." — check the next 5 fire times.
3. Add it: cron action:"add" name:"<short-name>" schedule:"..." prompt:"...".

Cheatsheet: minute hour day-of-month month day-of-week

• 0 9 * * mon — Mondays 09:00
• */30 * * * * — every 30 minutes
• 0 9 1 * * — first of the month, 09:00
• 30 8 * * 1-5 — weekdays 08:30

Names: letters/digits/_/./-, max 64 chars. Cap: 20 jobs.

Job prompts

The prompt becomes a standalone turn — no conversation context. Write it self-contained and short: what to do, where, and what to report. Bad: "check it again"; good: "Run npm test in this repo and summarize failures".

Loop guard: a fired turn cannot add, remove, or run jobs. Never write job prompts that schedule further jobs.

Managing

• cron action:"list" — name, schedule, next/last fire, enabled.
• cron action:"run" name:"..." — manual fire (result arrives as a follow-up turn).
• cron action:"disable" name:"..." — pause a job (schedule kept; skipped by ticks and export).
• cron action:"enable" name:"..." — resume; the next fire is recomputed from now (no surprise catch-up fire).
• cron action:"add" ... enabled:false — create the job already disabled.
• cron action:"remove" name:"...".
• cron action:"export" — crontab lines for 24/7 coverage while pi is closed; hand the block to the user to install (crontab -l | cat - cron.txt | crontab -). Exported lines run headless pi -p --no-session and log under <agentDir>/cron/logs/.

Config

settings.json → "cron": { "enabled": true, "tickMs": 30000 } (global kill switch + tick interval). timeoutMs caps headless (pinned) child runs (SIGTERM at the cap, SIGKILL after a 5s grace; clamped 1min–24h, default 10min) — raise it for long pinned jobs.