schedule
Manages recurring and one-off scheduled tasks. Session-scoped scheduling via CronCreate/CronDelete/CronList. Documents the cloud path for tasks that need to survive machine sleep or network drops.
- License 许可
- MIT
- 最近更新
- 2026-11-13
- Stars
- ★ 931
- Forks
- 84
01 前言 Front Matter
- name
- schedule
- license
- MIT
- description
- Manages recurring and one-off scheduled tasks. Session-scoped scheduling via CronCreate/CronDelete/CronList. Documents the cloud path for tasks that need to survive machine sleep or network drops.
- auto-trigger
- false
- last-updated
- 1774483200000
- user-invocable
- true
- trigger_keywords
- [ "schedule", "recurring", "every N minutes", "cron", "set a reminder", "run periodically" ]
02 资源文件 2
- 参考文档__benchmarks__/list-empty.md716 B
- 参考文档__benchmarks__/natural-language-interval.md1.1 KB
03 SKILL.md 正文
/schedule — Task Scheduling
Orientation
Use when: scheduling a recurring or one-off remote agent run (cron-style, outside the current session). Don't use when: repeating work within the current session (use /loop); continuous unattended campaigns (use /daemon).
Default execution path (READ FIRST)
/schedule add does NOT call CronCreate by default. It shells out to
node scripts/local-schedule.js which installs a native OS entry (Windows
Task Scheduler or Unix cron). Only pass --remote to use Anthropic's routine
system, and only after explicit user confirmation.
Why: CronCreate counts against the account-wide 15 routine runs / 24h
cap; every fire of the scheduled task counts. See
docs/ROUTINE-QUOTA.md.
Default flow — /schedule add "<expr>" "<command>" (no --remote)
Run:
node scripts/local-schedule.js add "<expr>" "<command>" --confirm
Before running, show the cadence, project, prompt and persistence and obtain approval to create the OS task. Normal Claude permissions remain enabled. Do not add a bypass flag to make unattended execution succeed.
Then report the returned ID and the removal command. This survives session
end, machine reboot, and consumes zero routine quota. Use
/schedule list and /schedule remove {id} (which also shell out to
local-schedule.js) by default. Remove tasks before uninstalling. New jobs require
their .citadel/schedules/<id>.json record; deleting it disables execution but leaves
an inert OS entry. Existing jobs retain their old bypass flags until explicitly
removed and recreated. See runner safety and migration.
Codex automation lane
When running in Codex and the user wants the schedule to survive the current terminal session, create a Codex Automation plan instead of a local OS entry:
node scripts/codex-automation.js plan --type schedule --cadence "<expr>" --command "<command>" --write
Give the returned prompt to the Codex app automation surface and keep the generated .planning/codex-automations/{id}.json as the durable Citadel record.
If .planning/ does not exist, create it before writing automation state.
Opt-in routine flow — /schedule add --remote ...
Only when --remote is explicitly passed:
- Confirm: "This will use
CronCreate, which counts against your 15 routine runs / 24h quota and is cleared at session end. Continue? (y/N)" - On confirmation, run the
CronCreate-based flow documented below.
The rest of the protocol documents the full CronCreate flow for reference
and for --remote invocations.
When to Route Here
- "run pr-watch every hour"
- "check my PRs automatically"
- "schedule a thing"
- "remind me to run tests every 30 minutes"
- "set up a recurring task"
- "list my scheduled tasks"
- "cancel the PR check"
- Any mention of "schedule", "recurring", "every N minutes/hours", "cron"
Protocol
/schedule list
List all currently scheduled tasks using CronList.
Output format:
Active schedules (N):
[id] {description} — {cron expression} — next run: {time}
No schedules active.
If CronList is not available: output a helpful error (see Fringe Cases).
/schedule add "{description}" {/skill-or-command}
Create a recurring task.
Steps:
- Parse the user's description to extract:
- Natural language interval: "every 30 minutes", "hourly", "every day at 9am"
- The skill or command to run:
/pr-watch,/do status, etc.
- Convert natural language to a cron expression (see Conversion Table below)
- Confirm with user: "I'll run
{command}{natural-language-interval} (cron:{expression}). OK?" - If confirmed: call CronCreate with the expression and command
- Output: "Scheduled. ID: {id}. Use
/schedule remove {id}to cancel."
Cron Expression Conversion Table:
| Natural Language | Cron Expression |
|---|---|
| every minute | * * * * * |
| every 5 minutes | */5 * * * * |
| every 15 minutes | */15 * * * * |
| every 30 minutes | */30 * * * * |
| every hour / hourly | 0 * * * * |
| every 2 hours | 0 */2 * * * |
| every 6 hours | 0 */6 * * * |
| every day / daily | 0 9 * * * (default 9am) |
| every day at {H}am/pm | 0 {H} * * * |
| every weekday | 0 9 * * 1-5 |
| every Monday | 0 9 * * 1 |
If the user provides a raw cron expression directly, use it as-is without converting. Validate it has 5 fields before accepting.
/schedule remove {id}
Remove a scheduled task by ID using CronDelete.
If the user doesn't know the ID: run /schedule list first, show the list,
and ask which one to remove.
Output: "Removed schedule {id} ({description})."
/schedule status
Show all active schedules and their next run times. Equivalent to /schedule list
with additional context about what each task does and when it last ran (if available).
Session-Scoped vs. Cloud-Persistent Scheduling
Session-Scoped (CronCreate)
CronCreate schedules tasks that run during the current Claude Code session only. When the session ends (Claude Code closes or the conversation is reset), all session-scoped schedules are cleared.
Use session-scoped when:
- Running checks during an active work session ("remind me every 30min to commit")
- Polling for PR feedback while you're at the computer
- Triggering skill runs during a long coding session
Cloud-Persistent (RemoteTrigger)
For tasks that need to survive machine sleep, network drops, or session restarts, use RemoteTrigger — a one-off cloud trigger that fires from Anthropic's infrastructure rather than your local session.
Use cloud-persistent when:
- The task needs to run overnight or while you're away
- You want notifications when you return to your machine
- The interval spans multiple days or calendar dates
How to set up a one-off cloud trigger:
- Call RemoteTrigger with the desired delay and the command to run
- Claude Code registers the trigger in Anthropic's cloud scheduler
- When the trigger fires, it wakes a new Claude Code session and runs the command
- Results are delivered as a notification
Note: RemoteTrigger requires Claude Code with cloud features enabled (Pro or Team plan). CronCreate works on all plans but is session-scoped only.
Fringe Cases
CronCreate not available: Output error; suggest OS cron/Task Scheduler. Never fail silently.
Ambiguous interval: Ask for clarification before proceeding.
Raw cron expression: Accept without conversion; validate 5 fields.
Every-minute schedule: Warn about 60 fires/hour; suggest 5m or 15m instead.
No schedules when listing: "No active schedules. Use /schedule add to create one."
Pause requested: Explain pause isn't supported; remove and recreate instead.
Contextual Gates
Disclosure: "Creating schedule: {command} at {cron expression}. OK?"
Reversibility: amber — creates OS cron entries or CronCreate sessions (side effects outside the repo); undo with /schedule remove {id}
Trust gates:
- Any: confirms before creating; shows cron expression before accepting
Quality Gates
- Always confirm before creating (CronCreate is a side effect)
- Always show the cron expression alongside the natural-language description
- Always provide the ID after creation so the user can remove it
- Never leave a user unable to remove a schedule they created
Exit Protocol
/schedule does not produce a HANDOFF block. After each action, output a concise confirmation or list and wait for the next command.
- After
add: "Scheduled. ID: {id}. Use/schedule remove {id}to cancel." - After
remove: "Removed schedule {id}." - After
listorstatus: the active schedule list (or "No active schedules.") - After any error: a clear message and actionable suggestion.
04 同仓库更多 Skills
workspace
Multi-repo campaign coordinator. Same lifecycle as fleet -- scope claims, discovery relay, wave-based execution -- but the unit of work is a repo, not a file. Coordinates campaigns across repositories with shared context.
wiki
Markdown-first knowledge base where the LLM acts as librarian. Ingests raw sources, compiles and interlinks topic files, self-maintains an index. No vector DB or embeddings required -- uses LLM-native navigation over structured markdown up to ~400K words.
watch
File sentinel that monitors the working directory for changes and marker comments, then auto-triggers appropriate skills. Poll-based via git diff against the last scan commit. Writes intake items for batch processing and routes marker actions through /do. Use for automatic reactions to file changes; do NOT use for one-off inspection or tasks needing human judgment per file.
verify
Self-test the Citadel hook pipeline from within a live session. Exercises real tool calls (Write, Edit, Bash, Read) and checks that hooks fired, telemetry accumulated, and no errors occurred. Reports HOOK HEALTH: PASS or HOOK HEALTH: FAIL with a per-hook breakdown.
unharness
Safely leave Citadel using the active adoption receipt. Produces a no-write, reviewable plan, preserves a portable archive, removes only exact owned material, and reports modified or externally registered surfaces as retained or unknown. Legacy installs must be imported before exact leave is claimed.
triage
GitHub issue and PR investigator. Pulls open issues/PRs, classifies them, searches the codebase for root cause or reviews contributed code, proposes fixes with file:line references, and optionally implements fixes. Use for investigating GitHub issues and reviewing PRs; do NOT use for general code review unrelated to GitHub issues.