scheduled-tasks
Schedule agent work for later, create or change recurring tasks, pause or resume future scheduling, inspect task execution history, or explicitly report an outcome from a task-dispatched conversation in PI WEB.
- License 许可
- MIT
- 最近更新
- 2026-11-15
- Stars
- ★ 942
- Forks
- 189
01 前言 Front Matter
- name
- scheduled-tasks
- description
- Schedule agent work for later, create or change recurring tasks, pause or resume future scheduling, inspect task execution history, or explicitly report an outcome from a task-dispatched conversation in PI WEB.
03 SKILL.md 正文
Scheduled agent tasks
Discover the deferred scheduled_tasks tools with tool_search or codemode searchTools. Use describeNamespace("scheduled_tasks") for guidance and inspect each tool's schema before calling it. These tools require this package's enabled PI WEB backend on the session's machine; they do not use the browser's selected machine.
- Start with
scheduled_tasks_overviewfor tasks and the configured default timezone. Usescheduled_tasks_targetsto list registered projects, then supply a project ID to list its workspaces. Do not substitute filesystem paths for IDs. - Translate the user's request into an initial prompt and either a one-off offset-bearing ISO timestamp or a five-field cron schedule. Store an explicit IANA timezone. If the user omitted it, use the configured default; if there is no default, ask rather than guessing. Resolve material schedule ambiguity before creating a task.
- Preserve intent; do not plan execution. Task creation is transcription plus scheduling, not task execution or investigation. Keep the prompt close to the user's request. Do not add goals, deliverables, obligations, workflows, assessment criteria, limits, tools, files, documentation references, or other implementation choices unless the user explicitly requested them. Leave investigation and execution decisions to the dispatched agent. Ask before making assumptions that materially change the work.
- Set optional
model(provider/model ID) andthinkingLevel(off,minimal,low,medium,high,xhigh,max) only when requested. Omit either independently to use the session default. If the model is unavailable at dispatch, the scheduler falls back to the default model and records an execution notice. Unavailable thinking levels similarly fall back to the session default, with a notice. Inspect execution notices and effective selections in history or execution details. Do not remove or rewrite the requested task settings because one execution fell back. - Populate optional
contextPathsonly when the user explicitly asks to include files. Paths are workspace-relative; the dispatched agent reads their current contents, not snapshots. - Before creating a task or updating its definition, show the proposed name, exact prompt, schedule, timezone, target project/workspace, any context paths, and any model/thinking overrides; for updates, highlight what changes. Use
ask_userto request explicit confirmation and wait for the answer before callingscheduled_tasks_createorscheduled_tasks_update. The initial request to schedule or change work is not confirmation of the proposed definition. Request confirmation again if substantive changes follow confirmation; harmless formatting or normalization that preserves meaning does not require renewed confirmation. - Inspect before edits.
scheduled_tasks_updatereplaces the full definition and requires its current revision; preserve unchanged fields. Pause/resume, archive, and delete also require the current revision. A stale revision means inspect again, not overwrite blindly. - Pausing affects future scheduled dispatches only. Archiving also prevents manual dispatch. Neither affects existing conversations. A recurring task dispatches each occurrence independently, even when earlier executions have no outcome report.
scheduled_tasks_deletepermanently removes an archived task, all definition revisions, execution records, and outcome reports. Reject deletion of a task that is not archived; archive it first and inspect its new revision before deleting. Existing conversations and transcripts are neither deleted nor stopped. Deletion rejects while a dispatch is still being admitted; wait and inspect again before retrying. Deleted executions no longer accept outcome reports.- Manual dispatch starts a normal independent conversation without consuming the next scheduled occurrence. Inspect history after a failed or uncertain launch; never blindly retry an ambiguous dispatch. There are no automatic retries after an uncertain launch; model/thinking fallback happens only after a typed rejection confirming that no initial prompt was submitted.
- Use
scheduled_tasks_historyfor paged records andscheduled_tasks_executionfor the exact dispatched task revision, prompt, context references, conversation ID, requested and effective model/thinking selections, dispatch notices and all outcome corrections.
When dispatched from a task
The initial message identifies your task and execution IDs. Retrieve your context with scheduled_tasks_execution, read any shared files, then do the requested work. You have the same task-management authority as any ordinary session; no extra plugin-specific approval applies.
Explicitly call scheduled_tasks_report with your execution ID, completed or errored, and a useful summary when you have an outcome. You may correct an earlier report; all reports remain in history. If you are still working or waiting for an answer, leave the execution dispatched. Session idle/error/close signals do not determine its outcome. Reporting does not stop your conversation or disable future occurrences.
Availability
The scheduler runs while its owning daemon is running, without a browser. It cannot wake the machine. After downtime it skips missed recurring occurrences and dispatches an overdue one-off once. Disabling the plugin stops future dispatches and preserves its SQLite tasks/history and existing conversations.
04 同仓库更多 Skills
relay
Carry work across a chain of fresh agent contexts using a shared goal, compact baton, and durable record. Use when a user invokes Relay, discusses the Relay method, prepares a Relay, or continues an active Relay. Do not load for generic multi-step work or ordinary delegation.
relay-runner
Prepare and run Relay legs in Pi: keep the goal, baton, and log durable, then hand off to one fresh session or stop visibly. Use with relay for /relay, /relay-worktree, or a handoff naming relay-runner. Supplies continuity and a bounded fresh-context final review without prescribing the development route.
relay
暂无简介
relay-runner
暂无简介
ux-guide
Use whenever designing, implementing, changing, or reviewing PI WEB UI interactions, controls, layouts, styling, or accessibility, including plugin surfaces. Guides reuse and extension of shared UI primitives and browser verification rather than prescribing per-screen styling. For interface copy, also use ux-writing.
npm-release-via-github-actions
Use this skill whenever the user asks for a new npm version, npm release, package release, new release, version bump, publishing to npm, cutting a GitHub release, tagging a release, or anything similar. It publishes through GitHub Actions and GitHub Releases, not from the local machine, and uses Changesets to generate CHANGELOG.md/release notes. Trigger even for casual phrasing like "ship a release", "bump npm", "publish the package", or "make a new version".