execute

📁 majiayu000/claude-skill-registry 📅 Feb 4, 2026
3
总安装量
3
周安装量
#58837
全站排名
安装命令
npx skills add https://github.com/majiayu000/claude-skill-registry --skill execute

Agent 安装分布

opencode 3
codex 3
amp 2
github-copilot 2
claude-code 2
replit 1

Skill 文档

Execution System

Purpose: Do the work at high quality with velocity. When to trigger: After clarity is established and priorities are set.

Pre-Flight Checklist

Before writing any code, verify:

  • Clarity exists – active-context.md is current with clear success criteria
  • Task is prioritized – This is the right thing to work on now
  • Success criteria defined – I know exactly what “done” looks like
  • Boundaries set – I know what NOT to do

If any are missing → Return to appropriate system.

Execution Protocol

1. Track Progress Visibly

Use TodoWrite at the start:

1. [ ] Step one of implementation
2. [ ] Step two of implementation
3. [ ] Validation step

Update status as you go. The user should always know where you are.

2. Make the Smallest Viable Change

  • Don’t over-engineer
  • Don’t add features that weren’t requested
  • Don’t refactor unrelated code
  • Don’t add “nice to haves”
  • Solve the exact problem, nothing more

3. Validate Continuously

After every significant change:

npx tsc --noEmit  # Catch type errors immediately

Don’t let errors accumulate. Fix as you go.

4. One Thing at a Time

  • Complete one task before starting another
  • Don’t context-switch mid-implementation
  • If blocked, acknowledge it rather than switching
  • Mark complete immediately when done

5. Parallel When Independent

Use subagents (Task tool) for truly independent work:

  • Research while implementing
  • Multiple unrelated fixes
  • Exploration while writing

But NOT for dependent steps – those must be sequential.

Quality Gates

Before marking any task complete:

Code Quality

  • Types pass (npx tsc --noEmit)
  • No lint errors
  • Follows existing patterns in codebase
  • No unintended side effects

Functionality

  • Works for the happy path
  • Edge cases considered
  • Error states handled
  • Loading states present (if UI)

Integration

  • Doesn’t break existing functionality
  • Works with existing data
  • API contracts maintained

Documentation (if applicable)

  • Complex logic commented
  • API changes noted in SOURCE_OF_TRUTH.md
  • Breaking changes flagged

Velocity Principles

Do:

  • Edit existing files over creating new ones
  • Use existing patterns and utilities
  • Ask early if blocked (don’t spin)
  • Ship incrementally
  • Commit working states frequently

Don’t:

  • Invent new patterns when existing ones work
  • Gold-plate solutions
  • Wait until everything is perfect
  • Make changes outside the task scope

Output Tracking

During execution, maintain:

  1. Todos – Current task breakdown with status
  2. Active context – Updated if scope changes
  3. Issues – Any new problems discovered → Identity System

Completion Criteria

A task is ONLY complete when:

  1. All acceptance criteria from clarity are met
  2. All quality gates pass
  3. Type check passes
  4. Functionality verified
  5. No regressions introduced

Handling Blocks

If you hit a wall during execution:

  1. STOP – Don’t keep trying the same failing approach
  2. DOCUMENT – What was attempted, what failed
  3. TRANSITION – Go to Reset System if blocked, or ask user

Never mark something complete that isn’t actually done.

Transition

After execution:

  • Success → Mark complete, update SOURCE_OF_TRUTH.md if significant
  • Partial success → Document what remains, continue or handoff
  • Failure → Proceed to Reset System
  • New issues discovered → Log to Identity System

Capture Learning: If something important was learned (new pattern, gotcha, or insight), add one line to learnings.md. Keep it brief – this is memory, not documentation.


This is System 4 of 5: Clarity → Identity → Priority → Execution → Reset