cron-scheduler

# Cron Scheduler > **Convention:** See `skills/conventions/test-before-bulk.md` — test every cron job on 3-5 items first. ## Contract This skill guarantees: - Schedule staggering: max 1 job per 5-minute slot, no collisions - Quiet hours gating: timezone-aware, with user-awake override - Thin job prompts: jobs say "Read skills/X/SKILL.md and run it" (no inline 3000-word prompts) - Idempotency: jobs can run twice without duplicate side effects - Results saved as reports: `reports/{job-name}/{YYYY-MM-DD-HHMM}.md` - One entry point: every job, workspace script and hook calls gbrain through one wrapper that sets `GBRAIN_HOME` and keeps stderr and the exit code ([the wrapper](https://github.com/garrytan/gbrain/blob/master/docs/guides/cron-schedule.md#calling-gbrain-unattended)) ## Phases 1. **Define job.** Name, schedule (cron expression), skill to run, timeout. 2. **Validate schedule.** Check no collision with existing jobs (5-minute offset rule). - Slots: :05, :10, :15, :20, :25, :30, :35, :40, :45, :50 - If collision detected, suggest the next available slot 3. **Check quiet hours.** Default: 11 PM - 8 AM local time. - Override: user-awake flag (if user is active, quiet hours suspended) - During quiet hours: save output to held queue - Morning contact releases the backlog 4. **Register with host scheduler.** OpenClaw cron, Railway cron, crontab, or process manager. **Each registered entry should execute via Minions, not `agentTurn`.** See `skills/conventions/cron-via-minions.md` for the rewrite pattern (PGLite uses `--follow`, Postgres uses fire-and-forget + `--idempotency-key` on the cycle slot). GBrain's v0.11.0 migration auto-rewrites entries for built-in handlers; host-specific handlers need a code-level registration per `docs/guides/plugin-handlers.md`. 5. **Write thin prompt.** Job prompt is one line: "Read skills/{name}/SKILL.md and run it." 6. **Call gbrain through the wrapper.** Cron runs with another `HOME` and environment than your shell, so a bare `gbrain` (or `bun run src/cli.ts`) can find no database (every lookup silently empty) or act as a second host (`refresh_not_owner`). Register `gbrain-run <args>`, the wrapper in [cron-schedule.md](https://github.com/garrytan/gbrain/blob/master/docs/guides/cron-schedule.md#calling-gbrain-unattended): it sets `GBRAIN_HOME` (never `HOME`), loads the brain's env file and runs the installed binary. Log stderr to a file, check the exit code, keep hook output off stdout, and give each call a timeout of at least 10 s. The lines below write `gbrain-run`; doctor prints the bare command. `gbrain doctor --only cron_gbrain_home` warns about any crontab line or script that runs `src/cli.ts` directly or calls gbrain without the brain's `GBRAIN_HOME`. ## Idempotency Requirement Every cron job MUST be idempotent: - Running the same job twice produces the same result (no duplicate pages, no duplicate timeline entries) - Use checkpoint state files to track progress and resume interrupted runs - Check for existing output before creating new output ## Output Format Job configuration saved. Report: "Job '{name}' scheduled at {cron expression}. Next run: {time}." ## Multi-source brains: use `sync --all`, not per-source entries When the brain has 2+ active sources (anything `gbrain sources list` shows with a non-null `local_path` that isn't archived), use one consolidated cron line instead of N per-source entries. **Preferred (multi-source)**: ```cron */5 * * * * gbrain-run sync --all --parallel 4 --workers 4 --skip-failed >> /tmp/gbrain-cron.log 2>&1 ``` This replaces N per-source lines AND auto-picks-up future sources without a crontab edit. Concurrency budget: `parallel × workers × 2 ≈ 32` connections during the wave (each per-file worker opens its own 2-connection pool). Stay under your Postgres `max_connections` setting. **Avoid (legacy)**: separate `gbrain sync --source default` and `gbrain sync --source zion-brain` entries staggered by 5 minutes. They require manual deconfliction every time a new source is added, and a slow source can race a fast source on the legacy global `gbrain-sync` lock (v0.40.3.0+ uses per-source `gbrain-sync:<sourceId>` locks but the per-source cron pattern doesn't benefit from the parallelism that `--all --parallel` actually delivers). `gbrain doctor` surfaces the recommended line as a `sync_consolidation` check whenever it detects 2+ active sources. Paste-ready from there. ### Managed brain A managed brain refuses the line above: managed sync needs `--no-pull` and rejects `--skip-failed` (`writer_coordinator_required`). Check the mode first: `gbrain sources writer status --json` reports `"mode": "managed"` or `"mode": "classic"`, and its `bindings` name each source's owner host. On a managed brain, install this line instead, on the host that owns the sources: ```cron */15 * * * * gbrain-run sources refresh <id> >> /tmp/gbrain-cron.log 2>&1; gbrain-run sync --all --no-pull --hard-deadline 13m >> /tmp/gbrain-cron.log 2>&1 ``` - `gbrain sources refresh <id>` is the only way a managed checkout takes upstream commits (a fetch and `git merge --ff-only`, then a managed sync of the sources on that checkout). Add one refresh per checkout that tracks a remote, with its real source id; drop it for checkouts with no remote. The `;` is deliberate: the sync still runs when a refresh refuses. - `refresh_diverged`: the checkout has commits its upstream lacks, so something commits outside gbrain (usually a script that writes files and runs `git commit` in the checkout). Fix that writer so it commits with `gbrain git commit-push -- <paths>`, then reconcile as the refusal says (`gbrain git commit-push --push-only --keep-checkout`, after the user agrees); never force the checkout ([scripts and the managed checkout](https://github.com/garrytan/gbrain/blob/master/docs/guides/shared-checkout-git.md#scripts-and-the-managed-checkout)). - `--hard-deadline 13m` stays under the 15-minute interval. Managed sync takes no per-source lock, so a run that outlives the interval would overlap the next tick. A run stopped at its deadline ends `resumable` and the next tick continues it. Sources drain one at a time, so `--parallel` does not apply. - `owner_unavailable` for a source: read its `detail`. `unbound_source` on a classic Postgres brain is normal and does not stop classic sync; keep the sync line. Any other detail means another host owns the source: install a per-source line on that host (`gbrain sync --source <id> --no-pull --hard-deadline 13m`, plus its refresh) instead of claiming it here. - Failed files: after fixing the files and committing them, rerun the plain `gbrain sync --source <id> --no-pull …` with the failed run's options; the next sync retries what the last one recorded. Never add `--skip-failed` (`--retry-failed` is a deprecated alias of the plain sync on a managed brain). `gbrain doctor` prints this managed line in `sync_consolidation` on a managed brain with 2+ active sources. ## When it fails Follow the [agent operator protocol](../conventions/agent-operator-protocol.md) for any gbrain error `code`, exit code, `[AGENT]` block or notice block. Specific to this skill: - A scheduled `gbrain sync` hits `sync_in_progress` / `lock_busy`: an earlier tick still runs. Widen the interval or stagger the job; never add a second overlapping schedule. - Doctor reports a stale source after the cron change: check `gbrain sources status <id>` for held items or errors before changing the schedule again. - `checkpoint_validation_timeout` in a sync log: run the retry command the error prints; do not cancel the request. - `writer_coordinator_required` from a scheduled sync: the brain is managed. Replace the line with the [managed brain](#managed-brain) recipe; never retry with `--skip-failed` or without `--no-pull`. - `sync_in_progress` from `gbrain sources refresh`: a managed sync cursor is unfinished, and the refresh never waits on it. Run the `gbrain sync --source <id> --no-pull …` command the refusal prints; the next tick's refresh then runs. - `worktree_refreshing` from a sync: a refresh is draining or merging that checkout. Leave it; the next tick syncs it. - A job's lookups all come back empty, or its log shows `no_brain` or `GBRAIN_DB_ACCESS no_url`: the job runs without the brain's config. Call gbrain through the wrapper; an empty result from a job whose stderr went to `/dev/null` is not evidence of "no notes". - `refresh_not_owner` from cron while the same refresh works by hand: the job resolves another gbrain home, so it is another host. The refusal names the identity file, `HOME` and `GBRAIN_HOME` it read; set the owner's `GBRAIN_HOME` in the wrapper. Never copy or delete a `host.json`. - A `--json` caller fails to parse gbrain's output: something the wrapper runs prints to stdout. Send that hook's output to a log file. - Calls time out in the job but finish by hand: a cold start takes seconds. Raise the timeout to at least 10 s in both the shell and the runtime that spawns it, and use the [warm CLI socket](https://github.com/garrytan/gbrain/blob/master/docs/guides/cli-socket.md). - Exit 10 from a scheduled write: it was accepted and is still pending, not lost or "not found". Poll `gbrain write-request <request_id>` before retrying or reporting it. ## Anti-Patterns - Scheduling jobs at the same minute (:00 for everything) - Inline 3000-word prompts in cron jobs (use skill file references) - Running cron jobs without testing on 3-5 items first - Jobs that produce different output on re-run (not idempotent) - Sending notifications during quiet hours (save to held queue instead) - A bare `gbrain` or `bun run src/cli.ts` in a cron line, script or hook instead of the wrapper; a wrapper that sets `HOME` instead of `GBRAIN_HOME` - `2>/dev/null` on a gbrain call, or ignoring its exit code - A script that writes files and runs plain `git commit` in a managed checkout instead of `gbrain git commit-push -- <paths>` - Printing `env` or a process environment into a log or transcript while debugging (it holds the database URL and keys) - Separate per-source `gbrain sync --source <id>` cron entries when `gbrain sync --all --parallel N --workers N` would replace them with one line that auto-picks-up future sources.