crons

name	crons
description	Runs cron-program lifecycle workflows for wall-clock scheduling, recurring automation, and scheduler diagnostics. Use when creating, tuning, or repairing time-based automation.

crons

Use this skill when the user asks to schedule, inspect, tune, or debug mu cron automation.

Contents

• Core contract
• Preflight checks
• Schedule kinds
• Cron lifecycle workflow
• Prompt design for scheduled runs
• Diagnostics and recovery
• Evaluation scenarios

Core contract

1. Use explicit schedule semantics

   • Choose the right schedule kind (at, every, or cron) instead of overloading one pattern.
   • Keep schedule intent readable in title and reason fields.

2. CLI-first lifecycle control

   • Create/update/trigger/enable/disable/delete via mu cron ... commands.
   • Do not hand-edit cron.jsonl.

3. Bounded execution prompts

   • Scheduled runs should execute one bounded control-loop pass and exit.
   • Avoid prompts that imply unbounded autonomous execution.

4. Read -> mutate -> verify

   • Inspect scheduler/program state first.
   • After changes, re-read and trigger a smoke run.

Preflight checks

mu status --pretty
mu control status --pretty
mu control identities --all --pretty
mu cron stats --pretty
mu cron list --limit 20 --pretty

If user expects channel delivery, verify linked operator identity for that channel before diagnosing cron execution as broken.

Schedule kinds

1. One-shot (at)

   • Run once at an absolute time.
   • Flags: --schedule-kind at --at <iso8601> (or --at-ms <epoch-ms>)

2. Fixed interval (every)

   • Run repeatedly on fixed millisecond cadence.
   • Flags: --schedule-kind every --every-ms <ms>
   • Optional alignment: --anchor-ms <epoch-ms>

3. Cron expression (cron)

   • Run by cron expression.
   • Flags: --schedule-kind cron --expr "<cron-expr>" --tz <timezone>

Cron lifecycle workflow

1) Inspect scheduler and programs

mu cron stats --pretty
mu cron list --limit 20 --pretty
mu cron get <program-id> --pretty

Use --json --pretty for full records.

2) Create a cron program

One-shot example:

mu cron create \
  --title "One-shot audit" \
  --schedule-kind at \
  --at 2026-02-22T02:00:00Z \
  --reason oneshot_audit

Fixed interval example:

mu cron create \
  --title "Every 10m check" \
  --schedule-kind every \
  --every-ms 600000 \
  --reason periodic_check

Cron-expression example:

mu cron create \
  --title "Nightly maintenance" \
  --schedule-kind cron \
  --expr "0 2 * * *" \
  --tz UTC \
  --reason nightly_maintenance

3) Update schedule or enablement

mu cron update <program-id> --enabled false
mu cron update <program-id> --schedule-kind every --every-ms 300000
mu cron update <program-id> --schedule-kind cron --expr "0 3 * * *" --tz UTC
mu cron enable <program-id>
mu cron disable <program-id>

4) Trigger smoke run

mu cron trigger <program-id> --reason smoke_test

Then verify:

mu cron get <program-id> --pretty
mu store tail events --limit 40 --pretty
mu store tail cp_operator_turns --limit 40 --pretty

5) Delete obsolete programs

mu cron delete <program-id>

Prompt design for scheduled runs

Use concise prompts with explicit bounded-pass behavior.

Example:

Review open issues for root <root-id>. Perform exactly one bounded step:
select one ready task (or report blocked), apply one action, verify state,
post concise summary, then exit.

For DAG execution workloads, combine with:

• planning
• protocol
• control-flow (when explicit loop/termination policy is required)
• model-routing (when per-issue model/provider/thinking policy is required)
• execution
• heartbeats (for short-cadence wake loops)

Diagnostics and recovery

When cron automation appears stalled or misfiring:

1. Confirm scheduler + program state:

mu cron stats --pretty
mu cron list --enabled true --limit 50 --pretty
mu cron get <program-id> --pretty

2. Trigger manually to separate scheduler timing issues from execution issues:

mu cron trigger <program-id> --reason manual_recovery_test

3. Inspect runtime evidence:

mu store tail events --limit 60 --pretty
mu store tail cp_operator_turns --limit 60 --pretty
mu store tail cp_outbox --limit 40 --pretty

4. Apply smallest recovery step:

• fix schedule-kind/flags mismatch
• correct timezone (--tz)
• simplify prompt scope
• toggle enablement (disable -> enable)
• relink channel identity if delivery is absent

Evaluation scenarios

1. One-shot schedule execution (at)

   • Setup: mu cron create --schedule-kind at --at <future-iso>.
   • Expected: program executes once, records deterministic status, then does not re-fire.

2. Interval schedule retune (every)

   • Setup: active interval schedule at 10m.
   • Expected: mu cron update ... --every-ms 300000 changes cadence to 5m without recreating program ID.

3. Cron expression + timezone correctness

   • Setup: cron expression schedule with explicit timezone.
   • Expected: next-run alignment matches timezone expectations; manual trigger succeeds and logs are auditable.