AgentSkillsCN

verify

确保工作成果完整——以结果为导向,鼓励超越最低要求。

SKILL.md
--- frontmatter
name: verify
description: Verify work is complete - outcome-focused, encourages going beyond minimum
allowed-tools: Bash, Read, Grep, Glob, TaskUpdate, TaskList
model: sonnet
user-invocable: false

Verify

Confirm the work is done well, not just done.

Quick Verification

bash
npm run typecheck && npm run build

If these fail, fix them first. No exceptions.

What "Complete" Means

A task is complete when:

  1. It works. Build passes, types check, feature functions.

  2. It solves the actual problem. Not just the literal requirements, but the underlying need.

  3. It's production-ready. Handles errors, edge cases, and real-world conditions.

Beyond Acceptance Criteria

Acceptance criteria are the minimum. If you see opportunities to:

  • Make the UX better
  • Improve performance
  • Add helpful error messages
  • Fix related issues you discover

Do it. A capable developer doesn't stop at "meets requirements."

For UI Tasks

Verify visually:

  • Does it look right?
  • Do all states work? (loading, error, empty, content)
  • Is it responsive?

Use agent-browser for verification when helpful:

bash
agent-browser open http://localhost:3000/path
agent-browser snapshot -i

Marking Complete

PASS:

code
TaskUpdate({
  taskId: "[id]",
  status: "completed",
  metadata: { passes: true, verified: "build" }
})

FAIL:

  • Report what's wrong
  • Keep task in_progress
  • Fix and re-verify

Report Format

code
Verify: [task-id] - [title]
═══════════════════════════
Build: ✓
Types: ✓
Works: ✓

Result: PASS

Extras: [Any improvements made beyond requirements]
Next: [next-task-id] - [title]

The Standard

Would a senior developer approve this PR?

If yes, it's done. If no, improve it.

Don't just tick boxes. Ensure the solution is genuinely good.