| name | cron-task |
| description | Create workspace cron tasks that run the agent on a schedule. Use when the user wants a repeating background task, periodic inbox triage, scheduled reports, or any recurring agent workflow. |
Cron Task Skill
Use the create_cron tool to create scheduled agent tasks under the workspace crons/ directory.
When To Use
Use this skill when the user wants the agent to:
- check something on a schedule
- run a repeated workflow every N minutes/hours/days
- build a recurring inbox triage or report task
- save a prompt as a workspace cron job instead of running it once
Do not use this skill for one-shot work. For one-time execution, run the task directly.
Scene Detection — Roleplay Cron
Before creating the task, check whether the cron involves the agent acting under a defined identity or persona. This includes ANY of the following:
- Acting as a specific character (fictional or otherwise)
- Operating as a professional role: programmer, manager, analyst, reviewer, support agent, moderator, etc.
- Maintaining a consistent personality, tone, or speaking style across runs
- Replying to messages or conversations in character
- Any task where "who the agent is" matters for behavior
If the task matches any of the above, call get_skill("roleplay-cron") and follow that skill's instructions for creating the cron. The roleplay-cron skill defines the correct file architecture (SOUL.md, SKILL.md, task.md separation) and anti-patterns to avoid. Do not proceed with the generic task skeleton below — use roleplay-cron instead.
If the task is purely mechanical (data fetching, report generation, cleanup, monitoring) with no identity requirements, continue with the generic flow below.
Required Output Shape
When creating a cron task, you must produce a complete task definition, not a vague reminder.
Every cron task should explicitly include:
- Which skill to read first, if the task depends on a workspace skill.
- The main execution flow in order.
- The decision rules for when to act, skip, or stop.
- Any required environment details that must be used inside the task.
- The expected reply or artifact behavior.
Cron Expression Rules
- Use standard 5-field cron expressions:
minute hour day-of-month month day-of-week
- Example every 5 minutes:
*/5 * * * *
- Example every day at 09:30:
30 9 * * *
Task Writing Rules
Write the task as an executable instruction block for a future agent run.
The task should:
- be self-contained
- state the target skill by name
- require
get_skill before using that skill's workflow
- tell the agent how to decide which items need action
- state what to do when there is nothing worth doing
- end with a clear stop condition
If the task involves replying to messages, require:
- reading context first
- an internal structured summary before replying
- a short reply plan
- then the actual send step
Recommended Task Skeleton
First call get_skill for <skill-name> and follow that skill strictly.
Execution flow:
1. ...
2. ...
3. ...
Decision rules:
1. reply when ...
2. skip when ...
3. stop when ...
If no action is needed, explicitly say so and stop.
QQ Inbox Example
For a recurring QQ inbox task, write the task so it says all of the following:
- First call
get_skill for qqinbox.
- Use the QQ inbox workflow from that skill.
- Read recent conversations before replying.
- Internally summarize topic, intent, and whether a reply is needed.
- State a short reply plan.
- Send replies only when justified by the conversation.
- If there is nothing worth replying to, explicitly skip and stop.
Tool Usage
Use create_cron with:
cron_id: stable workspace cron identifier
cron_expression: standard 5-field cron expression
task: the full task definition text
enabled: whether the cron should start enabled
Important Constraint
Do not create under-specified cron tasks like:
- "check qq"
- "run every 5 minutes"
- "look and reply if needed"
Those are too vague for later scheduled execution.