AgentSkillsCN

pm-do

执行一项任务——如有需要则获取资源，加载上下文，实施并完成

SKILL.md
--- frontmatter
name: pm-do
description: Execute a task — grab if needed, load context, implement, and complete
user_invocable: true
disable-model-invocation: true
args: "<task-id>"

/pm-do — Execute Task

Phase 1: Claim & Context

  1. •Call pm_get(task_id) to read the task
  2. •If the task is todo with no assignee:
    • •Call pm_grab(task_id) to claim it (validates readiness)
    • •If grab fails, stop and show the blockers
  3. •If the task is in-progress with a different assignee:
    • •Warn: "This task is assigned to {assignee}. Proceed anyway?"
    • •Only continue if explicitly confirmed
  4. •Read the parent story for broader context:
    • •Call pm_get(story_id) to understand acceptance criteria
  5. •Review the task's Implementation section and DoD

Phase 2: Execute

  1. •Read project documentation if touching new areas:
    • •Call pm_docs("project") for architecture context
  2. •Implement the work described in the task:
    • •Follow the implementation instructions
    • •Write/modify the specified files
    • •Run tests as described
  3. •Verify ALL DoD criteria are met — check off each item

Phase 3: Complete

  1. •Call pm_update(task_id, status="review") if the task needs human review, OR call pm_update(task_id, status="done") if all DoD criteria are met
  2. •Check if all sibling tasks in the story are done:
    • •Call pm_active or list tasks for the story
    • •If all done, suggest: "All tasks for {story_id} are complete — update story to done?"
  3. •Summarize what was done, files changed, tests run
  4. •Suggest next action: "Check the board for more work: /pm board"