AgentSkillsCN

copilot-taskmaster

防止智能体在多步骤任务中过早结束。当此功能启用时，智能体将在每次回复结束前，运行一份包含5个要点的完成检查清单——确保用户的所有请求、计划步骤以及各项任务都已100%完成。当用户提出复杂且多步骤的工作需求时，若希望确保万无一失，可使用此功能。

SKILL.md
--- frontmatter
name: copilot-taskmaster
description: |
  Prevents the agent from finishing prematurely on multi-step tasks. When
  activated, the agent runs a 5-point completion checklist before every
  response end — verifying that all user requests, plan steps, and tasks
  are 100% complete. Use when the user asks for complex, multi-step work
  and you want to ensure nothing is missed.
license: MIT

Copilot Taskmaster

A behavioral skill that prevents the agent from stopping prematurely. Before finishing any response, run the completion checklist below to verify all work is truly done.

Rules

  1. •

    Always maintain a task list. Use the update_todo tool to track every discrete task. Update it as you start and complete work.

  2. •

    Before finishing ANY response, run the completion checklist (see below).

  3. •

    If anything remains, continue working immediately. Do not just describe what is left — ACTUALLY DO IT.

  4. •

    Respect user intent. If the user explicitly changed their mind, withdrew a request, said to stop, or told you to skip something, treat that item as resolved. Do NOT force completion of work the user no longer wants.

  5. •

    Loop protection. Track how many times you have run the checklist consecutively with no new issues found. After 3 consecutive clean checks, trust the result and finish. Do not loop forever.

Completion Checklist

Run these checks before every response end:

1. RE-READ THE ORIGINAL USER MESSAGE(S)

List every discrete request or acceptance criterion. For each one, confirm it is fully addressed — not just started, FULLY done.

2. CHECK THE TASK LIST

Review every task in your update_todo list. Any task not marked completed? Do it now — unless the user indicated it is no longer wanted.

3. CHECK THE PLAN

Walk through each step of the plan (if one exists). Any step skipped or partially done? Finish it — unless the user redirected or deprioritized it.

4. CHECK FOR ERRORS

Did any tool call, build, test, or lint fail? If so, fix it before finishing.

5. CHECK FOR LOOSE ENDS

Any TODO comments, placeholder code, missing tests, or follow-ups noted but not acted on? Address them.

After the Checklist

  • •If everything is genuinely 100% done (or explicitly deprioritized by the user), briefly confirm completion for each user request, then finish.
  • •If anything remains, immediately continue working on whatever is left.
  • •Increment your internal "clean check" counter each time the checklist passes with zero issues. Reset the counter whenever you find and fix something.

Example Flow

code
User: "Add input validation to the form, write tests, and update the README."

Agent thinks:
  ✓ Checklist pass 1 — validation done, tests written, README not updated → continue
  ✓ Checklist pass 2 — README updated, all tests pass → clean check #1
  ✓ Checklist pass 3 — re-verified, everything complete → clean check #2
  ✓ Checklist pass 4 — still all good → clean check #3 → safe to finish