sinapse-ai 9.3.0 → 9.5.0

package/.claude/rules/squad-awareness.md

@@ -1,71 +1,99 @@
 ---
-paths: **/*
+paths:
+  - "squads/**"
+  - ".sinapse-ai/development/agents/**"
 ---
 
-# Sinapse — Orchestration Rules
-
-> **CRITICAL:** This project has specialized AI agent squads installed. When a user request falls within a domain covered by a squad, you MUST delegate to the appropriate specialist agent instead of handling it yourself.
-
-## Delegation Rule
-
-When a user request matches a squad domain (see table below):
-1. **Acknowledge** the domain is covered by a specialized squad
-2. **Recommend** activating the squad's orchestrator or specialist agent
-3. **Provide** the invocation command (e.g., `/ca:agents:ca-orchestrator`)
-4. **Do NOT** handle the request yourself if a dedicated agent exists
-
-**Exception:** If the user explicitly asks you to handle it anyway, proceed — but note the specialized squad exists.
-
-## Squads Instaladas
-
-| Squad | Capacidade |
-|-------|-----------|
-
-
-## Mapa de Delegacao por Dominio
-
-| Dominio | Squad | Agente Lead | Invocacao |
-|---------|-------|-------------|-----------|
-| Branding e identidade visual | squad-brand | brand-orchestrator (Meridian) | `/brand:agents:brand-orchestrator` |
-| Vendas e estrategia comercial | squad-commercial | cs-orchestrator (Pipeline) | `/commercial:agents:cs-orchestrator` |
-| Conteudo e editorial | squad-content | content-orchestrator | `/content:agents:content-orchestrator` |
-| Copywriting e persuasao | squad-copy | copy-strategist (Quill) | `/copywriting:agents:copy-strategist` |
-| Animacoes web, Three.js, shaders, motion | squad-animations | ca-orchestrator (Kinetic) | `/ca:agents:ca-orchestrator` |
-| UX/UI e experiencia digital | squad-design | dx-orchestrator (Nexus) | `/digital-experience:agents:dx-orchestrator` |
-| Inteligencia financeira e pricing | squad-finance | fi-orchestrator (Ledger) | `/finance:agents:fi-orchestrator` |
-| Growth organico, SEO e analytics | squad-growth | ga-orchestrator (Catalyst) | `/growth:agents:ga-orchestrator` |
-| Midia paga (Meta Ads, Google Ads, CRO) | squad-paidmedia | pm-orchestrator (Apex) | `/pm:agents:pm-orchestrator` |
-| Produto e discovery | squad-product | ps-orchestrator (Vector) | `/product:agents:ps-orchestrator` |
-| Pesquisa e inteligencia competitiva | squad-research | research-orchestrator (Prism) | `/research:agents:research-orchestrator` |
-| Claude Code mastery e automacao | squad-claude | cm-orchestrator (Orion) | `/claude:agents:cm-orchestrator` |
-| Conselho estrategico e modelos mentais | squad-council | council-orchestrator (Zenith) | `/council:agents:council-orchestrator` |
-| Narrativa, storytelling e pitch | squad-storytelling | narrative-orchestrator (Arc) | `/narrative:agents:narrative-orchestrator` |
-| Seguranca cibernetica e compliance | squad-cybersecurity | cyber-orchestrator (Fortress) | `/cyber:agents:cyber-orchestrator` |
-
-## Quando Delegar
-
-| Situacao | Squad |
-|----------|-------|
-| Animacao/motion/Three.js/shader | squad-animations |
-| Copy/headline/persuasao | squad-copy |
-| Branding/identidade/logo | squad-brand |
-| Pesquisa/benchmark/analise competitiva | squad-research |
-| Conteudo/editorial/blog/social media | squad-content |
-| UX/UI/design system/wireframe | squad-design |
-| Growth/SEO/analytics/metricas organicas | squad-growth |
-| Vendas/proposta/pitch/CRM | squad-commercial |
-| Financeiro/pricing/P&L/budget | squad-finance |
-| Midia paga/Meta Ads/Google Ads/CRO | squad-paidmedia |
-| Produto/discovery/roadmap | squad-product |
-| Claude Code/prompt engineering/MCP | squad-claude |
-| Conselho estrategico/modelos mentais/decisao | squad-council |
-| Storytelling/narrativa/pitch/apresentacao | squad-storytelling |
-| Seguranca/pentest/compliance/incident response | squad-cybersecurity |
-
-## Handoff Protocol
-
-1. **Identificar** o dominio do pedido
-2. **Informar** qual squad cobre e como invocar: `/{prefix}:agents:{agent-id}`
-3. **Fornecer contexto** do handoff se necessario
-4. Squads sao **autonomas** — o orchestrator coordena internamente
-5. Squads possuem **knowledge bases**, **tasks** e **workflows** proprios em `./squads/{squad-name}/`
+# Sinapse — Intelligent Auto-Routing (NON-NEGOTIABLE)
+
+> Users are NOT AI experts. They should NEVER need to know agent names, commands, or squad structures.
+> The system MUST understand natural language and route automatically.
+
+## Golden Rule
+
+**User types anything → SINAPSE understands → Routes to the right specialist → Delivers result.**
+
+The user NEVER needs to:
+- Know which agent handles what
+- Type `@agent-name` or `/SINAPSE:agents:...`
+- Understand greenfield vs brownfield
+- Know about squads, orchestrators, or handoffs
+
+## Auto-Detection at Session Start
+
+On EVERY session start, automatically detect:
+1. **Project State:** Greenfield (empty/new) | Brownfield (existing code) | SINAPSE-managed
+2. **Tech Stack:** Framework, language, database, testing, CI/CD
+3. **Maturity:** Score 0-10 based on tests, docs, CI presence
+
+Use this context to adjust behavior:
+- **Greenfield:** Start with scaffolding, architecture decisions, project setup
+- **Brownfield:** Start with discovery, tech debt assessment, understanding existing code
+- **SINAPSE-managed:** Continue normal SDC workflow
+
+## Automatic Request Routing
+
+When ANY user message arrives, classify and route:
+
+| User Says (examples) | Route To | How |
+|----------------------|----------|-----|
+| "cria um site", "novo projeto", "scaffold" | @architect → @developer | Greenfield workflow |
+| "arruma esse bug", "fix isso" | @developer | Fast-track if trivial |
+| "cria uma marca", "logo", "identidade" | squad-brand | Auto-delegate |
+| "escreve um copy", "headline", "landing page" | squad-copy | Auto-delegate |
+| "pesquisa sobre X", "analise competitiva" | squad-research | Auto-delegate |
+| "faz deploy", "publica", "push" | @devops | Exclusive authority |
+| "testa isso", "quality check" | @quality-gate | QA gate |
+| "cria uma story", "nova feature" | @sprint-lead → @product-lead | SDC workflow |
+| "animacao", "three.js", "shader" | squad-animations | Auto-delegate |
+| "SEO", "growth", "analytics" | squad-growth | Auto-delegate |
+| "ads", "campanha", "meta ads" | squad-paidmedia | Auto-delegate |
+| "financeiro", "pricing", "P&L" | squad-finance | Auto-delegate |
+| "seguranca", "pentest", "LGPD" | squad-cybersecurity | Auto-delegate |
+| "conteudo", "blog", "editorial" | squad-content | Auto-delegate |
+| "design system", "UI", "wireframe" | squad-design | Auto-delegate |
+| "estrategia", "conselho", "decisao" | squad-council | Auto-delegate |
+| "storytelling", "pitch", "narrativa" | squad-storytelling | Auto-delegate |
+| "vende isso", "proposta", "CRM" | squad-commercial | Auto-delegate |
+| "produto", "roadmap", "discovery" | squad-product | Auto-delegate |
+
+## Handoff Protocol (Automatic)
+
+When routing between agents/squads:
+1. **Never ask permission** — just route
+2. **Provide context** — pass the user's original request + project state
+3. **Confirm briefly** — "Delegando para @specialist que e o expert nisso."
+4. **Return result** — bring the deliverable back to the user
+
+## Brownfield Auto-Behavior
+
+When project is detected as BROWNFIELD and user hasn't run discovery:
+1. **First interaction:** "Detectei um projeto existente. Vou analisar a estrutura antes de comecar."
+2. **Auto-run:** Quick tech stack scan (< 30 seconds, not full brownfield discovery)
+3. **Inform:** "Projeto {framework} com {database}. Pronto para trabalhar."
+4. **Then proceed** with the user's original request
+
+## Greenfield Auto-Behavior
+
+When project is detected as GREENFIELD:
+1. **First interaction:** "Projeto novo detectado. Vou configurar a estrutura ideal."
+2. **Ask minimal:** "Que tipo de projeto? (web app, API, landing page)"
+3. **Auto-scaffold:** Apply templates, CI/CD, .env.example
+4. **Then proceed** with implementation
+
+## Tool & Command Handoff
+
+When ANY agent encounters a task outside its domain:
+- **Git push needed** → Auto-delegate to @devops (NEVER ask user)
+- **Database work** → Auto-delegate to @data-engineer
+- **Test needed** → Auto-delegate to @quality-gate
+- **Architecture decision** → Auto-delegate to @architect
+- **Story needed** → Auto-create via fast-track (trivial) or @sprint-lead (complex)
+
+## Anti-Patterns (FORBIDDEN)
+
+- Asking user "which agent should handle this?"
+- Showing agent invocation commands to the user
+- Requiring user to type `@agent-name` for routing
+- Leaving a task unfinished because "that's another agent's job" without auto-delegating
+- Asking user if project is greenfield or brownfield (auto-detect it)

package/.claude/rules/token-economy.md

@@ -0,0 +1,148 @@
+# Token Economy & Response Format — NON-NEGOTIABLE
+
+> Consolidado: economia de contexto + formato de resposta em uma única regra.
+> Princípio: desperdício de contexto degrada qualidade E custa caro. Mesmo inimigo.
+> Detalhes: `.sinapse-ai/development/knowledge-base/token-economy-guide.md`.
+
+---
+
+## 1. Compactação
+
+| Trigger | Threshold |
+|---|---|
+| Auto-compact | **60%** do contexto (não 83%) |
+| Manual `/compact` | Troca de agente; pós-leitura de arquivo grande |
+| Pre-agent-switch | Handoff artifact (~379 tok) via `agent-handoff.md` |
+
+60%: acima disso perde coerência sobre instruções iniciais ("context amnesia").
+
+---
+
+## 2. Model Routing (MUST)
+
+**Regra zero:** Execute direto sempre que der. Sub-agente custa ~20K mínimo.
+
+**Opus 4.7:** Effort default `xhigh`. `thinking_budget` fixo NÃO suportado — adaptive.
+
+| Task | Modelo | Effort |
+|---|---|---|
+| Arquitetura cross-system, debug complexo, refactor multi-file | **opus** | `xhigh` |
+| Spec Pipeline COMPLEX (score >= 16) | **opus** | `max` |
+| Feature do spec, code review, bug fix, testes, stories | **sonnet** | `high` |
+| Análise single-file, pergunta factual | **sonnet** | `medium` |
+| Lint, rename, YAML, lookup, bulk | **haiku** | `low` |
+
+- Em dúvida: menor tier. Escala se falhar.
+- Nunca opus pra task que haiku resolve. Nunca `max` fora de COMPLEX.
+- Sub-agente anuncia modelo ao spawnar.
+
+---
+
+## 3. Subagent Threshold (Opus 4.7)
+
+Spawn APENAS se: `>= 8 tool calls` previstos OU fan-out paralelo real. Abaixo → inline.
+
+---
+
+## 4. Anti-Patterns (FORBIDDEN)
+
+| Anti-Pattern | Fix |
+|---|---|
+| Ler mesmo arquivo 2x | Uma vez, guarda line numbers |
+| Re-ler após Edit/Write | Edit confirma sucesso, não releia |
+| Persona completo em agent switch | Handoff protocol |
+| Grep/Glob sem `head_limit` | Sempre setar |
+| Sub-agente pra task <8 tool calls | Inline |
+| Sequential reads independentes | Paralelo (uma mensagem, N calls) |
+| Cole payload bruto no raciocínio | Extrai só relevante |
+| `thinking_budget` fixo em 4.7 | Adaptive — não suportado |
+| Preamble ("Claro!", "Vou te ajudar...") | Ação direta |
+| Trailing summary ("Em resumo...") | Só próximo passo, 1 linha |
+| Narração de plano | Executa em paralelo |
+| Restating da pergunta | Responde direto |
+
+---
+
+## 5. Hierarquia de Tool (MANDATORY)
+
+```
+Caminho conhecido → Read   (não Bash cat)
+Padrão conhecido  → Grep   (não Bash grep)
+Lista de arquivos → Glob   (não Bash find)
+Edição pontual    → Edit   (não Write)
+```
+
+Bash só pra operações que nenhuma tool dedicada cobre. Leitura cirúrgica: `offset`+`limit`.
+
+---
+
+## 6. Response Format
+
+### Comprimento por complexidade
+
+| Task | Máximo |
+|---|---|
+| Confirmação / Yes-No | 1 linha |
+| Pergunta factual | 1-3 linhas |
+| Task simples concluída | 5-10 linhas |
+| Multi-step | Só o que mudou |
+| Análise profunda | Estruturado, sem padding |
+
+### Padrão requerido
+
+```
+[ação ou resposta direta]
+[detalhe crítico se não-óbvio]
+[próximo passo se aplicável, 1 linha]
+```
+
+### Formato por situação
+
+| Situação | Use |
+|---|---|
+| Comparação, métrica | Tabela |
+| Sequência de passos | Lista numerada |
+| Itens sem ordem | Bullets |
+| Código | Code block |
+| Narrativa | Parágrafo curto (máx 3 frases) |
+
+Default: tabela ou bullet.
+
+---
+
+## 7. Linguagem
+
+- **Português** pra Caio e Matheus
+- Sem jargão ("salvei" não "commitei no HEAD")
+- Sem nomes de agentes em conversa (usuário vê "implementei", não "@developer")
+- Sem comandos pro usuário rodar — agentes fazem tudo (Safe Collaboration)
+
+**Exceção:** usuário pede explicitamente ("explica em detalhes", "modo educativo"). Mesmo assim, sem preamble nem trailing summary.
+
+---
+
+## 8. Budget de Contexto (200K)
+
+| Alocação | Target | % |
+|---|---:|---:|
+| System + CLAUDE.md + Rules | ~10K | 5% |
+| Agente + tools | ~5K | 3% |
+| Working memory | **≤80K** | 40% |
+| Histórico | ~58K | 29% |
+| Safety buffer (compacta 60%) | ~40K | 20% |
+
+Working memory >80K = sinal pra compactar.
+
+---
+
+## 9. Memory Anti-Patterns
+
+- Não salvar estado efêmero (PR atual, task de hoje)
+- Não salvar o que já tá em código/git
+- Memory = fatos que surpreenderiam uma sessão futura
+
+---
+
+## Enforcement
+
+NON-NEGOTIABLE. Violação = falha de qualidade (Constitution Art. V).

package/.claude/rules/tool-examples.md

@@ -1,3 +1,9 @@
+---
+paths:
+  - ".sinapse-ai/data/tool-registry.yaml"
+  - ".mcp.json"
+---
+
 # Tool Input Examples — Selection Guidance
 
 ## Purpose

package/.claude/rules/workflow-execution.md

@@ -1,3 +1,10 @@
+---
+paths:
+  - "docs/stories/**"
+  - ".sinapse-ai/development/tasks/**"
+  - ".sinapse-ai/development/workflows/**"
+---
+
 # Workflow Execution — Detailed Rules
 
 ## Task-First Principle