yishan-io/yishan-mono/starting-task
- community
Checking that the SKILL.md file is still in the repository.
Use when outcome-oriented engineering work may need a tracked Local Task before planning or implementation.
Install
npx vskill@latest install yishan-io/yishan-mono/starting-taskSKILL.md
Starting Task
Use this skill to decide if work needs a Local Task.
The SQLite-backed Local Task daemon owns task metadata and status. Pi uses the managed daemon through its configured endpoint. The daemon generates UUIDs. Imported IDs stay opaque.
Decision Rule
Create Local Tasks for independently meaningful, outcome-oriented engineering work:
- Bug fixes
- Features
- Refactors
- Improvements to reliability, performance, UX, or maintainability
- Migrations, security work, or investigations with a durable conclusion
Before automatically creating a task, confirm all of these:
- The request is not merely an operational action or a question.
- It has a concrete, independently meaningful outcome that can be stated and verified.
- It needs durable tracking because it may span sessions or needs a plan, tests, decisions, or acceptance criteria.
Do not create a task merely because work has multiple steps or touches multiple files. Do not create one for operational requests, even when they have several steps: git rebase, commit, push, branch switching, checking status, running a known command, or a small one-time edit.
Routine initial reading or investigation is not a task by itself. If it reveals a distinct outcome-oriented work item, create a task then. When uncertain, do not create a task. Reuse a related active task instead of creating a duplicate.
Always create a task when the user explicitly asks to track the work.
Start a Task
Use task_start with an outcome-focused title. Prefer separate goal, context, and independently checkable acceptanceCriteria fields for multi-step work; use description only when a short free-form brief is clearer. Do not provide both forms.
Use this concise creation shape when the information is known; omit sections that would be speculation:
Before creating a task automatically (without explicit user direction), call task_template_read to read the Agent default section structure. Use those headings and their order when writing the task description. Fall back to the built-in structure (## Goal, ## Context, ## Acceptance Criteria, ## Notes) only when the tool is unavailable.
Include context/background when it is known and materially affects the work; omit it rather than inventing history. Keep detailed research logs and changing decisions out of the task description. Put them in notes after task creation. The user can update task metadata with task_update at any time.
You can set priority and tags at creation. New tasks start with new status. Do not provide an ID.
Use workspace_list before task_start to resolve the current workspace. Use YISHAN_WORKSPACE_ID when it is set. Otherwise, select the workspace whose localPath matches the current worktree path. When that workspace is non-primary, pass its ID as workspaceId. Do not associate a task started in the primary workspace.
For delegated work, associate the task with an existing target workspace. If no target workspace exists, use workspace_create first, then pass its returned workspace ID to task_start.
If YISHAN_PROJECT_ID is set, the task belongs to that project. Without it, the task is global.
Work After Start
A tracked Local Task may contain multiple independently deliverable plan subtasks in its plan.md. A plan subtask is an execution unit within the tracked Local Task; its independent deliverability does not by itself require a new Local Task.
Before execution, the controller must create and freeze a brief for each plan subtask. Each brief must define its scope, non-goals, acceptance criteria or check, and a finite execution budget. Do not begin a plan subtask without its brief. The controller is the sole authority to create or replace scope-changing plan content.
Executors may record only within-brief execution detail in plan.md; they must not replace scope-changing plan content. When a newly discovered issue is related to the active plan subtask, a non-scope-changing clarification may stay in scope. If the discovery changes its frozen scope, requires a new decision or dependency, or changes assumptions for a later plan subtask, record it in notes and stop the current attempt. The controller decides whether to continue and, if so, must replace the affected scope-changing plan content with a newly frozen brief before resuming. Ask the user when the change affects unapproved behavior, commitments, or tradeoffs. For an unrelated issue, ask the user whether to create a separate Local Task.
Use task_read for the synthetic, read-only brief. Use task_update to change metadata, new, progressing, or cancelled status, priority, or tags. It cannot set done. Use task_list or task_search to find tasks.
Use task_write for plan, notes, or outcome. Use task_append_note for task-specific discoveries.
Use task_finish only when the user explicitly asks to complete the task. It writes the outcome and marks the task done.
Boundaries
Do not create or parse task IDs. Do not bypass the daemon. Do not write the synthetic brief. Do not use task_update for completion.
Security
- Security scanPassed, 90 of 100Pattern checks on the skill's files.
- Model reviewPassed
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingpackages/pi-task/skills/starting-task/SKILL.md
Popularity
- 7-day trend score
- 0
Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/yishan-io/yishan-mono/starting-task)