AgentSkillsCN

openspec-apply

当实施经批准的OpenSpec更改时，请使用此技能。阅读tasks.md并完全按照执行。不要偏离、添加功能或跳过任务。与每个代码任务一起使用test-tdd。

SKILL.md
--- frontmatter
name: openspec-apply
description: >-
  Use when implementing an APPROVED OpenSpec change. Read tasks.md and follow
  EXACTLY. Do not deviate, add features, or skip tasks. Use with test-tdd for
  each code task.

OpenSpec Apply

Implement approved changes by following the task list exactly.

Announce: "I'm using openspec-apply to implement change [id] following the approved tasks."

Iron Law

code
FOLLOW THE TASK LIST EXACTLY - NO ADDITIONS, NO SKIPS

The tasks were approved. Your job is to execute them, not improve them.

Process

Phase 1: Load Context

  1. •

    Read the proposal:

    bash
    openspec show [change-id]
  2. •

    Read tasks.md completely before starting

  3. •

    Identify current progress:

    • •Which tasks are [x] complete?
    • •Which is the next [ ] pending task?

Phase 2: Execute Tasks

For EACH task:

Step 1: Announce

code
Starting Task N.M: [task description]

Step 2: Load TDD skill → REQUIRED SUB-SKILL: Load test-tdd for any code task

Step 3: Follow task steps exactly

  • •Create/modify the exact files listed
  • •Write the exact code specified
  • •Run the exact verification commands

Step 4: Verify

  • •Test passes?
  • •No other tests broken?
  • •Types check?

Step 5: Mark complete Update tasks.md: - [ ] → - [x]

Step 6: Commit

bash
git add [files]
git commit -m "[type](scope): [description]"

Phase 3: Report Progress

After each task:

code
Completed Task N.M: [description]
- Files changed: [list]
- Tests: PASS
- Next: Task N.M+1

Phase 4: Completion

When all tasks are [x]:

  1. •

    Run full verification:

    bash
    bun run type-check
bun run lint
bun test
supabase test db
  2. •

    Report:

    code
    All tasks complete for change [id].
- [X] tasks completed
- All tests passing
- Ready for code review
  3. •

    Load code-review skill for final review

Working with Subagents

If dispatching domain experts:

  1. •Load subagent-workflow skill
  2. •Give them the specific task from tasks.md
  3. •Track their session_id
  4. •Verify their output
  5. •If issues found, recall their session for fixes

REQUIRED SUB-SKILL

  • •For any code task → Load test-tdd
  • •For subagent orchestration → Load subagent-workflow
  • •After completion → Load code-review

Red Flags - STOP

If you catch yourself:

  • •Adding features not in tasks
  • •Skipping tasks "to save time"
  • •Implementing differently than specified
  • •Not running verification steps
  • •Marking tasks complete without verification

STOP. Follow the task list.

Common Rationalizations

ExcuseReality
"I see a better way"The plan was approved. Follow it.
"This task is unnecessary"The plan was approved. Do it anyway.
"I'll combine these tasks"Tasks are sized deliberately. Do them separately.
"Tests are passing, skip verification"Run ALL verification. Every time.

What To Do When Stuck

If a task is unclear or blocked:

  1. •STOP - do not guess
  2. •State what's unclear
  3. •Ask for clarification
  4. •Wait for response

Do NOT proceed with assumptions.