القائمة
Maquina autonoma de teste final. QAT textual, UX-QAT visual, benchmark comparativo, ciclo test-fix-retest, E2E batch — PDCA convergente com scores.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
Use quando Codex, Claude Code, outro agente, plataforma ou automacao precisar consultar o Data Engine Knowledge Gateway e pedir acesso governado a dados/APIs via Access Broker CLI. Cobre descoberta de recursos, manifesto, validacao, preview, request, status, drift e descriptor, sem aprovar ou revelar segredos.
Entry point do sistema. Recebe qualquer pedido, classifica, roteia para a melhor combinação de skills/agents/plugins, e monitora. Vai além do óbvio — sugere combos compostos, ativa auxiliares proativos, e delega a plugins canonicals (ADR-0001) quando apropriado.
Maquina especialista em dados Raiz: data-engine API canonical (/v1/kpis/*, /reports/<panel>/data, Envelope v2, KPI Registry Ouro M01..A*) + SQL TOTVS RM + Neon PostgreSQL + Zeev BPM + PBI_RAIZ. Use para: descobrir KPI Ouro, consumir painel canonical, escrever query nova (cascade painel>KPI>dinamica>raw), otimizar SQL lenta, anti-patterns, relatorios, processos Zeev. Tier 0: data-engine. Aplica ADR-041 (TOTVS via Neon Mirror enforced).
Maquina de design UI/UX. 67 estilos, 96 paletas, 57 font pairings, 13 stacks. Design system, componentes, landing pages, dashboards. Integra com shadcn, Figma, v0.
Maquina autonoma de limpeza de dead code (Knip + AST + bundle, confidence tiers, PRs atomicos). Use para tech debt, hand-off, reducao de bundle, cleanup pos-spike, auditoria de saude. Modo --target=harness audita dead skills/hooks/MCPs do proprio Claude Code.
Auditoria completa de software (wrapper FORTRESS). Roda MERIDIAN + SENTINEL + ARCHITECT + CONDUCTOR + LIGHTHOUSE + HARNESS (opcional) em sequencia. Fortress Score = laudo completo.
| name | ag-4-teste-final |
| description | Maquina autonoma de teste final. QAT textual, UX-QAT visual, benchmark comparativo, ciclo test-fix-retest, E2E batch — PDCA convergente com scores. |
| model | sonnet |
| context | fork |
| argument-hint | [qat|ux-qat|benchmark|ciclo|e2e] [URL ou path] [--threshold N] [--resume] |
| allowed-tools | Read, Write, Edit, Glob, Grep, Bash, Agent, TaskCreate, TaskUpdate, TaskList, TeamCreate, TeamDelete, SendMessage |
| metadata | {"filePattern":"teste-final-state.json,tests/qat/**,tests/ux-qat/**,tests/qat-benchmark/**","bashPattern":"teste-final|qat|ux-qat|benchmark","priority":95} |
/teste-final qat ~/Claude/GitHub/raiz-platform
/teste-final ux-qat https://app.example.com
/teste-final benchmark https://app.example.com
/teste-final ciclo ~/Claude/GitHub/raiz-platform
/teste-final e2e ~/Claude/GitHub/raiz-platform
/teste-final --resume
| Modo | Agent interno | O que faz |
|---|---|---|
| qat | ag-testar-qualidade-qat + ag-criar-cenario-qat | QAT textual PDCA (L1-L4, dual-score) |
| ux-qat | ag-testar-ux-qualidade + ag-criar-cenario-ux-qat + ag-capturar-tela (apps nativos) | UX-QAT visual PDCA (screenshots, AI judge) |
| benchmark | ag-benchmark-qualidade + ag-criar-cenario-benchmark | Benchmark comparativo (dual-run, parity index) |
| ciclo | ag-ciclo-testes | Test-fix-retest completo (max 3 ciclos) |
| e2e | ag-testar-e2e-batch | E2E em batches com auto-fix |