verification-protocol
Verify AI Task Completion with Evidence
AI agents often claim success without proof. This protocol forces independent verification with concrete evidence before marking tasks complete. It eliminates false completions and ensures quality through objective validation.
Baixar o ZIP da skill
Upload no Claude
Vá em Configurações → Capacidades → Skills → Upload skill
Ative e comece a usar
Testar
A utilizar "verification-protocol". Verify that the new user authentication feature is complete
Resultado esperado:
- File exists: /src/lib/auth.ts (2,847 bytes, modified 2025-01-10 14:30:00)
- No placeholders found: 0 TODO/TBD/FIXME markers
- TypeScript compilation: 0 errors
- ESLint check: 0 errors, 1 warning
- Test results: 18 passed, 0 failed
- API endpoint test: POST /api/login returns 200 with JWT token
- Verification result: PASSED - All criteria met with evidence
Auditoria de Segurança
SeguroThis is a documentation-only skill defining verification protocols. No executable code exists. All 73 static findings are false positives: scanner misinterpreted documentation examples and metadata as security threats. Contains only Markdown documentation describing independent verification methods for AI task completion.
Fatores de risco
🌐 Acesso à rede (1)
📁 Acesso ao sistema de arquivos (6)
⚙️ Comandos externos (52)
Pontuação de qualidade
O Que Você Pode Construir
Prevent False Task Completions
Stop agents from marking tasks complete without proof. Independent verification catches incomplete work before it reaches production.
Standardize AI Verification
Consistent verification criteria across all AI agents. Every completion requires measurable evidence, not subjective claims.
Audit AI Work Quality
Track actual completion rates with evidence trails. Identify which agents need improvement through verification failure patterns.
Tente Estes Prompts
Use verification protocol to check if /src/lib/feature.ts exists, has no placeholders, compiles without errors, passes linting, and has working tests. Return evidence for each criterion.
Verify feature implementation across multiple files: check all claimed outputs exist, contain no TODOs, compile successfully, and pass all tests. Collect evidence for each file.
Check if API endpoints respond correctly: GET /api/users returns 200, POST /api/users creates records, error handling works. Provide status codes and response samples as evidence.
Verify complete system: database migrations applied, API routes accessible, frontend builds without errors, all tests pass, monitoring endpoints respond. Provide comprehensive evidence trail.
Melhores Práticas
- Always use independent verifier agent, never verify your own work
- Collect specific evidence for each verification criterion
- Block task completion when any verification fails
Evitar
- Self-attestation: Agent claiming its own work is complete
- Assumed success: Marking complete without checking evidence
- Loose verification: Accepting partial or unclear results