菜单
This skill teaches agents how to collect and verify evidence before marking tasks complete. Inspired by production-grade development practices, it ensures all claims are backed by executable proof:...
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/ComeOnOliver/skillshub --skill evidence-verification命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
基于 SOC 职业分类
正在显示 SKILL.md
| name | evidence-verification |
| description | This skill teaches agents how to collect and verify evidence before marking tasks complete. Inspired by production-grade development practices, it ensures all claims are backed by executable proof:... |
Version: 1.0.0 Type: Quality Assurance Auto-activate: Code review, task completion, production deployment
This skill teaches agents how to collect and verify evidence before marking tasks complete. Inspired by production-grade development practices, it ensures all claims are backed by executable proof: test results, coverage metrics, build success, and deployment verification.
Key Principle: Show, don't tell. No task is complete without verifiable evidence.
Test Evidence
Build Evidence
Deployment Evidence
Code Quality Evidence
## Evidence Collection Steps
1. **Identify Verification Points**
- What needs to be proven?
- What could go wrong?
- What does "complete" mean?
2. **Execute Verification**
- Run tests
- Run build
- Run linters
- Check deployments
3. **Capture Results**
- Record exit codes
- Save output snippets
- Note timestamps
- Document environment
4. **Store Evidence**
- Add to shared context
- Reference in task completion
- Link to artifacts
Minimum Evidence Requirements:
Production-Grade Requirements:
Use this template when running tests:
## Test Evidence
**Command:** `npm test` (or equivalent)
**Exit Code:** 0 ✅ / non-zero ❌
**Duration:** X seconds
**Results:**
- Tests passed: X
- Tests failed: X
- Tests skipped: X
- Coverage: X%
**Output Snippet:**
[First 10 lines of test output]
**Timestamp:** YYYY-MM-DD HH:MM:SS
**Environment:** Node vX.X.X, OS, etc.
Use this template when building:
## Build Evidence
**Command:** `npm run build` (or equivalent)
**Exit Code:** 0 ✅ / non-zero ❌
**Duration:** X seconds
**Artifacts Created:**
- dist/bundle.js (XXX KB)
- dist/styles.css (XXX KB)
**Errors:** X
**Warnings:** X
**Output Snippet:**
[First 10 lines of build output]
**Timestamp:** YYYY-MM-DD HH:MM:SS
Use this template for linting and type checking:
## Code Quality Evidence
**Linter:** ESLint / Ruff / etc.
**Command:** `npm run lint`
**Exit Code:** 0 ✅ / non-zero ❌
**Errors:** X
**Warnings:** X
**Type Checker:** TypeScript / mypy / etc.
**Command:** `npm run typecheck`
**Exit Code:** 0 ✅ / non-zero ❌
**Type Errors:** X
**Timestamp:** YYYY-MM-DD HH:MM:SS
Use this comprehensive template for task completion:
## Task Completion Evidence
### Task: [Task description]
### Agent: [Agent name]
### Completed: YYYY-MM-DD HH:MM:SS
### Verification Results
| Check | Command | Exit Code | Result |
|-------|---------|-----------|--------|
| Tests | `npm test` | 0 | ✅ 45 passed, 0 failed |
| Build | `npm run build` | 0 | ✅ Bundle created (234 KB) |
| Linter | `npm run lint` | 0 | ✅ No errors, 2 warnings |
| Types | `npm run typecheck` | 0 | ✅ No type errors |
### Coverage
- Statements: 87%
- Branches: 82%
- Functions: 90%
- Lines: 86%
### Evidence Files
- Test output: `.claude/quality-gates/evidence/tests-2025-XX-XX.log`
- Build output: `.claude/quality-gates/evidence/build-2025-XX-XX.log`
### Conclusion
All verification checks passed. Task ready for review.
When: After writing code for a feature or bug fix
Steps:
Save all files - Ensure changes are written
Run tests
npm test
# or: pytest, cargo test, go test, etc.
Run build (if applicable)
npm run build
# or: cargo build, go build, etc.
Run linter
npm run lint
# or: ruff check, cargo clippy, golangci-lint run
Run type checker (if applicable)
npm run typecheck
# or: mypy, tsc --noEmit
Document evidence
quality_evidenceMark task complete (only if all evidence passes)
When: Reviewing another agent's code or user's PR
Steps:
Read the code changes
Verify tests exist
Run tests
Check build
Verify code quality
Document review evidence
Approve or request changes
When: Deploying to production or staging
Steps:
Pre-deployment checks
Execute deployment
Post-deployment checks
Document deployment evidence
## Deployment Evidence
**Environment:** production
**Timestamp:** YYYY-MM-DD HH:MM:SS
**Version:** vX.X.X
**Pre-Deployment:**
- Tests: ✅ Exit 0
- Build: ✅ Exit 0
- Security: ✅ No critical issues
**Deployment:**
- Command: `kubectl apply -f deployment.yaml`
- Exit Code: 0 ✅
**Post-Deployment:**
- Health Check: ✅ 200 OK
- Smoke Tests: ✅ All passed
- Error Rate: <0.1%
Verify rollback capability
Shared Context (Primary)
{
"quality_evidence": {
"tests_run": true,
"test_exit_code": 0,
"coverage_percent": 87,
"build_success": true,
"build_exit_code": 0,
"linter_errors": 0,
"linter_warnings": 2,
"timestamp": "2025-11-02T10:30:00Z"
}
}
Evidence Files (Secondary)
.claude/quality-gates/evidence/ directory{type}-{timestamp}.logtests-2025-11-02-103000.logTask Completion Messages
✅ Tests executed with captured exit code ✅ Timestamp recorded ✅ Evidence stored in context
✅ Tests pass (exit code 0) ✅ Coverage ≥70% (or project standard) ✅ Build succeeds (exit code 0) ✅ No critical linter errors ✅ Type checker passes ✅ Security scan shows no critical issues
✅ All production-grade requirements ✅ Coverage ≥80% ✅ No linter warnings ✅ Performance benchmarks within thresholds ✅ Accessibility audit passes (WCAG 2.1 AA) ✅ Integration tests pass ✅ Deployment verification complete
Bad:
"I've implemented the login feature. It should work correctly."
Good:
"I've implemented the login feature. Evidence:
- Tests: Exit code 0, 12 tests passed, 0 failed
- Build: Exit code 0, no errors
- Coverage: 89%
Task complete with verification."
Bad:
"Tests passed" (without actually running them)
Good:
"Tests passed. Exit code: 0
Command: npm test
Output: Test Suites: 3 passed, 3 total
Timestamp: 2025-11-02 10:30:15"
Bad:
"Build failed with exit code 1, but the code looks correct so marking complete."
Good:
"Build failed with exit code 1. Errors:
- TypeError: Cannot read property 'id' of undefined (line 42)
Fixing the error now before marking complete."
Bad:
"Tests passed yesterday, so the code is still good."
Good:
"Re-running tests after today's changes.
New evidence: Exit code 0, 45 tests passed, coverage 87%"
Evidence is automatically tracked in shared context:
// Context structure includes:
{
quality_evidence?: {
tests_run: boolean;
test_exit_code?: number;
coverage_percent?: number;
build_success?: boolean;
linter_errors?: number;
timestamp: string;
}
}
Evidence collection feeds into quality gates:
In parallel execution:
Before marking task complete:
- [ ] Tests executed
- [ ] Test exit code captured (0 = pass)
- [ ] Build executed (if applicable)
- [ ] Build exit code captured (0 = pass)
- [ ] Code quality checks run (linter, types)
- [ ] Evidence documented with timestamp
- [ ] Evidence added to shared context
- [ ] Evidence summary in completion message
JavaScript/TypeScript:
npm test # Run tests
npm run build # Build project
npm run lint # Run ESLint
npm run typecheck # Run TypeScript compiler
Python:
pytest # Run tests
pytest --cov # Run tests with coverage
ruff check . # Run linter
mypy . # Run type checker
Rust:
cargo test # Run tests
cargo build # Build project
cargo clippy # Run linter
Go:
go test ./... # Run tests
go build # Build project
golangci-lint run # Run linter
See /skills/evidence-verification/examples/ for:
v1.0.0 - Initial release
Remember: Evidence-first development prevents hallucinations, ensures production quality, and builds confidence. When in doubt, collect more evidence, not less.