菜单
Maquina autonoma de qualidade. Descobre app, testa 5 dimensoes (ALIVE/REAL/WORKS/LOOKS/FEELS), corrige, re-testa ate convergencia. Quality Certificate + Fix PR + baselines.
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。
基于 SOC 职业分类
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-meridian |
| description | Maquina autonoma de qualidade. Descobre app, testa 5 dimensoes (ALIVE/REAL/WORKS/LOOKS/FEELS), corrige, re-testa ate convergencia. Quality Certificate + Fix PR + baselines. |
| model | sonnet |
| context | fork |
| argument-hint | [URL ou path do projeto] [--threshold N] [--audit-only] [--resume] [--scope rotas] |
| allowed-tools | Read, Write, Edit, Glob, Grep, Bash, Agent, TaskCreate, TaskUpdate, TaskList, TeamCreate, TeamDelete, SendMessage |
| metadata | {"filePattern":"meridian-*.json,meridian-*.md","bashPattern":"meridian","priority":95} |
/meridian https://app.example.com # App deployada
/meridian ~/Claude/GitHub/raiz-platform # Projeto local
/meridian https://app.example.com --threshold 90 # Threshold customizado
/meridian https://app.example.com --audit-only # So diagnosticar
/meridian --resume # Retomar run interrompido
/meridian ~/Claude/GitHub/fgts-platform --scope "/dashboard,/consultas" # Rotas especificas
Executa QA completo AUTONOMO em 5 fases:
Spawnar o agent principal:
Agent({
subagent_type: "ag-meridian",
prompt: "{input do usuario}",
run_in_background: true,
mode: "auto"
})
| Artefato | Arquivo |
|---|---|
| Quality Certificate | docs/meridian-certificate-YYYY-MM-DD.md |
| Fix PR | GitHub PR com before/after |
| Baselines | meridian-baselines.json |
| State (recovery) | meridian-state.json |
| KB Update | ~/.claude/shared/meridian-kb/ |
| Issues backlog | GitHub Issues com label meridian-finding |
| Agent | Fase | Proposito |
|---|---|---|
| ag-explorar-codigo | SCOUT | Explorar codebase |
| ag-testar-ux-qualidade | SIEGE D4 | Visual quality testing |
| ag-corrigir-bugs | FORGE | Bugfix em sprints |
| ag-corrigir-tipos | FORGE | Fix TypeScript errors |
| ag-depurar-erro | FORGE | Debug de bugs obscuros |