Usage

Model-facing tools, registered globally in every agent:

cron_add — a prompt plus exactly one selector: cron (five fields, optional time_zone) or at (one-shot RFC 3339 with offset). Returns the job with its next three fire times; an identical active job is reused, not duplicated.
cron_list — every job with schedule, state, next fire time, and last run outcome.
cron_update — pause or resume.
cron_remove — remove by id.

The same store from the human side:

/cron list
/cron add 0 9 * * 1-5 Summarize overnight CI results
/cron add tz=Asia/Shanghai 0 9 * * 1-5 Prepare the morning standup
/cron add-at 2026-08-20T09:00:00+08:00 Prepare the release checklist
/cron pause cron-3
/cron resume cron-3
/cron remove cron-3

In the web profile, a ⏰ Cron action in the sidebar footer opens a panel with every job, its last run, and Run now / Pause / Delete actions, backed by a loopback /cron RPC channel. Other plugins can drive the store through the provided cron service.

Configuration

Key	Default	Meaning
`dataDir`	Harness-home `cron` directory	Directory holding `jobs.json` (atomic writes; a corrupt file is quarantined aside)
`defaultTimeZone`	host local zone	IANA zone for schedules that omit one
`maxJobs`	`64`	Maximum number of active jobs
`minIntervalMinutes`	`1`	Minimum gap between two occurrences of one recurring job
`coldWake`	`false`	Resume a due job's cold creating session so the task fires with no live session
`busyDelivery`	`followup`	Busy-target delivery: `followup` queues the task as the next turn; `inject` rides the running turn as context

dsh-cron

AI 分析

インストール

ドキュメント

Usage

Configuration