AgentSkillsCN

verification-loop

在构建或审查前端组件、落实无障碍设计，或与前端架构师合作时，确保UI组件符合WCAG 2.1 AA标准。

SKILL.md
--- frontmatter
name: verification-loop
description: Ensure no completion claims without fresh evidence by verifying changes actually work before marking tasks complete

Verification Loop

Purpose

Ensure no completion claims without fresh evidence. Verify changes actually work before marking tasks complete.

Core Principle

Never claim completion without verification in THIS session.

Previous runs, cached results, or assumptions are not valid evidence.

Verification Checklist

Before Claiming "Done"

  • • Build passes (run build command, show output)
  • • Tests pass (run test command, show output)
  • • Linter passes (run lint command, show output)
  • • No regressions (existing functionality still works)

Evidence Requirements

Each verification requires:

  1. •Command executed in current session
  2. •Output shown (not summarized from memory)
  3. •Timestamp implicit in tool call

Framework Detection

JavaScript/TypeScript Projects

bash
npm run build && npm test && npm run lint

Python Projects

bash
# Activate virtual environment first if present
source venv/bin/activate 2>/dev/null || source .venv/bin/activate 2>/dev/null || true
python -m pytest && python -m flake8

Go Projects

bash
go build ./... && go test ./... && go vet ./...

Failure Handling

On Build Failure

  1. •Parse error message
  2. •Identify root cause
  3. •Apply fix
  4. •Re-run verification
  5. •Do NOT claim done until build passes

On Test Failure

  1. •Identify failing test(s)
  2. •Determine if test or code is wrong
  3. •Fix appropriately
  4. •Re-run ALL tests
  5. •Ensure no regressions

On Lint Failure

  1. •Address each linting issue
  2. •Prefer auto-fix where safe
  3. •Re-run linter
  4. •Verify all issues resolved

Never Skip

  • •Don't skip verification "because it worked last time"
  • •Don't assume changes are safe without testing
  • •Don't rely on IDE indicators without command verification
  • •Don't claim partial completion as full completion

Output Format

When verification passes:

markdown
## Verification Complete

### Build
✓ `npm run build` - Success (0 errors)

### Tests
✓ `npm test` - 42 tests passed, 0 failed, 85% coverage

### Lint
✓ `npm run lint` - No issues

**Status: Ready to commit**

When verification fails:

markdown
## Verification Failed

### Build
✗ `npm run build` - 2 errors

### Analysis
[Error details and fix plan]

**Status: Fixing issues before completion**