| name | everything-claude-code |
| description | Bundle de ~60 agents + ~221 skills cobrindo reviewers, build-resolvers, TDD, E2E, security, autonomous loops. Padrão do projeto — usar o MÁXIMO de agents possível em paralelo a cada alteração. |
| status | always-on |
| source | https://github.com/affaan-m/everything-claude-code |
Skill: everything-claude-code
Catálogo + protocolo de uso da suite ECC. Padrão do projeto: toda alteração dispara o máximo de agents ECC aplicáveis em paralelo. Single message, múltiplas tool calls.
Sempre ativo. Não opcional. Ignorar = task incompleta.
Trigger
- Qualquer edição de código → review agent da stack + security-reviewer (se toca auth/input/segredo).
- Build falhou →
*-build-resolver da stack.
- Bug →
engineering:debug skill + reviewer agent.
- Feature nova / refactor amplo →
planner + architect paralelo antes de codar.
- E2E →
e2e-runner.
- Docs →
doc-updater + docs-lookup (Context7).
- Performance →
performance-optimizer.
- Dead code →
refactor-cleaner.
Protocolo de invocação (HARD RULE)
Paralelo agressivo é o padrão. Single message com 3-5+ agents simultâneos sempre que trabalho independente.
Fluxo por fase
| Fase | Agents (paralelo) |
|---|
| Plan (feature/refactor) | Plan + everything-claude-code:planner + everything-claude-code:architect |
| Search/Explore | Explore (quick/medium/thorough) + general-purpose |
| Stack review (após edit) | everything-claude-code:<lang>-reviewer + everything-claude-code:code-reviewer + everything-claude-code:security-reviewer |
| Build error | everything-claude-code:<lang>-build-resolver |
| Testes | everything-claude-code:tdd-guide + test-runner |
| E2E | everything-claude-code:e2e-runner |
| Performance | everything-claude-code:performance-optimizer |
| Cleanup | everything-claude-code:refactor-cleaner |
| Docs | everything-claude-code:doc-updater + docs-lookup |
Reviewers por stack
| Stack | Reviewer |
|---|
| TS/JS | everything-claude-code:typescript-reviewer |
| Python | everything-claude-code:python-reviewer |
| Go | everything-claude-code:go-reviewer |
| Rust | everything-claude-code:rust-reviewer |
| Java/Spring | everything-claude-code:java-reviewer |
| Kotlin/Android | everything-claude-code:kotlin-reviewer |
| C# / .NET | everything-claude-code:csharp-reviewer |
| C++ | everything-claude-code:cpp-reviewer |
| Flutter/Dart | everything-claude-code:flutter-reviewer |
| SQL/migration | everything-claude-code:database-reviewer |
| Healthcare/PHI | everything-claude-code:healthcare-reviewer |
| Genérico | everything-claude-code:code-reviewer, code-architect |
Build resolvers por stack
| Stack | Resolver |
|---|
| TS/genérico | everything-claude-code:build-error-resolver |
| Go | everything-claude-code:go-build-resolver |
| Rust | everything-claude-code:rust-build-resolver |
| Java/Maven/Gradle | everything-claude-code:java-build-resolver |
| Kotlin/Gradle | everything-claude-code:kotlin-build-resolver |
| C++/CMake | everything-claude-code:cpp-build-resolver |
| Dart/Flutter | everything-claude-code:dart-build-resolver |
| PyTorch | everything-claude-code:pytorch-build-resolver |
Skills stack-specific (carregar via Skill tool ANTES de codar)
| Stack | Skills |
|---|
| Python | python-patterns + python-testing |
| TS/Node | frontend-patterns + nextjs-turbopack + bun-runtime |
| Go | golang-patterns + golang-testing |
| Rust | rust-patterns + rust-testing |
| Java/Spring | springboot-patterns + springboot-tdd + springboot-security + jpa-patterns |
| Kotlin | kotlin-patterns + kotlin-testing + kotlin-coroutines-flows |
| C# / .NET | dotnet-patterns + csharp-testing |
| C++ | cpp-coding-standards + cpp-testing |
| Flutter | dart-flutter-patterns |
| Laravel | laravel-patterns + laravel-tdd + laravel-security |
| Django | django-patterns + django-tdd + django-security |
| Postgres | postgres-patterns + database-migrations |
Regra de invocação (passo a passo)
- Identifica stack do arquivo tocado.
- Carrega
*-patterns + *-security + *-testing da stack (Skill tool, paralelo).
- Para feature/refactor amplo: dispara
planner + architect paralelo.
- Edita código.
- Dispara reviewer(s) da stack +
security-reviewer (se aplicável) paralelo num único message.
- Build falhou →
*-build-resolver.
- Testes via
tdd-guide + test-runner. E2E via e2e-runner.
- Docs via
doc-updater + Context7 (docs-lookup).
- Endereça feedback de TODOS reviewers ou justifica trade-off.
Padrões
- Paralelo > sequencial sempre que independente. Single message com múltiplas Agent calls.
- Trust mas verify — agent summary é intenção, não fato. Conferir diff real.
- Subagent prompt self-contained — paths, line numbers, contexto. Não "based on findings".
- Foreground vs background — foreground quando precisa resultado; background quando paralelo livre.
- Hard limit: nunca delegar entendimento. "Implementa baseado nas findings" = anti-pattern.
Slash commands úteis (ECC)
/ecc:plan, /multi-plan, /multi-execute, /loop-start, /loop-status, /code-review, /security-scan, /quality-gate, /test-coverage, /refactor-clean, /update-docs, /update-codemaps, /skill-create, /skill-stocktake.
Definition of Done
Notas
- Origem: https://github.com/affaan-m/everything-claude-code (v2.0.0-rc.1).
- Requer Claude Code CLI ≥ 2.1.0.
- Plugin global já registrado — agents/skills disponíveis sob namespace
everything-claude-code:*.
- Runtime knobs:
ECC_HOOK_PROFILE=standard|strict|minimal, ECC_DISABLED_HOOKS, ECC_SESSION_START_MAX_CHARS.
