sinapse-ai 9.3.0 → 10.0.0-rc.1

package/.claude/CLAUDE.md

@@ -1,384 +1,103 @@
 # CLAUDE.md - SINAPSE
 
-Este arquivo configura o comportamento do Claude Code ao trabalhar neste repositório.
-
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
 ## Constitution
 
-O SINAPSE possui uma **Constitution formal** com princípios inegociáveis e gates automáticos.
+Full doc: `.sinapse-ai/constitution.md`. Gates auto-block violations.
 
-**Documento completo:** `.sinapse-ai/constitution.md`
-
-**Princípios fundamentais:**
-
-| Artigo | Princípio | Severidade |
-|--------|-----------|------------|
+| Art. | Principle | Severity |
+|------|-----------|----------|
 | I | CLI First | NON-NEGOTIABLE |
 | II | Agent Authority | NON-NEGOTIABLE |
-| III | Documentation-First Development | NON-NEGOTIABLE |
+| III | Documentation-First | NON-NEGOTIABLE |
 | IV | No Invention | MUST |
 | V | Quality First | MUST |
 | VI | Absolute Imports | SHOULD |
-| VII | Ecosystem Metrics Accuracy | NON-NEGOTIABLE |
+| VII | Metrics Accuracy | NON-NEGOTIABLE |
 | VIII | Mandatory Delegation | NON-NEGOTIABLE |
 | IX | Safe Collaboration | NON-NEGOTIABLE |
 | X | Security & Data Protection | NON-NEGOTIABLE |
 
-**Gates automáticos bloqueiam violações.** Consulte a Constitution para detalhes completos.
-
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Language Configuration
-
-Language preference is handled by Claude Code's native `language` setting (v2.1.0+).
-Configure in `~/.claude/settings.json` (global) or `.claude/settings.json` (project):
-
-```json
-{ "language": "portuguese" }
-```
-
-The installer writes this automatically during `npx sinapse-ai install`. No language config in `core-config.yaml`.
-
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Premissa Arquitetural: CLI First
-
-O SINAPSE segue uma hierarquia clara de prioridades que deve guiar **TODAS** as decisões:
-
-```
-CLI First → Observability Second → UI Third
-```
-
-| Camada | Prioridade | Descrição |
-|--------|------------|-----------|
-| **CLI** | Máxima | Onde a inteligência vive. Toda execução, decisões e automação. |
-| **Observability** | Secundária | Observar e monitorar o que acontece no CLI em tempo real. |
-| **UI** | Terciária | Gestão pontual e visualizações quando necessário. |
-
-### Princípios Derivados
-
-1. **A CLI é a fonte da verdade** - Dashboards apenas observam, nunca controlam
-2. **Funcionalidades novas devem funcionar 100% via CLI** antes de ter qualquer UI
-3. **A UI nunca deve ser requisito** para operação do sistema
-4. **Observabilidade serve para entender** o que o CLI está fazendo, não para controlá-lo
-5. **Ao decidir onde implementar algo**, sempre prefira CLI > Observability > UI
-
-> **Referência formal:** Constitution Artigo I - CLI First (NON-NEGOTIABLE)
-
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Estrutura do Projeto
-
-```
-sinapse-ai/
-├── .sinapse-ai/              # Core do framework
-│   ├── core/                # Módulos principais (orchestration, memory, etc.)
-│   ├── data/                # Knowledge base, entity registry
-│   ├── development/         # Agents, tasks, templates, checklists, scripts
-│   └── infrastructure/      # CI/CD templates, scripts
-├── bin/                     # CLI executables (sinapse-init.js, sinapse.js)
-├── docs/                    # Documentação
-│   └── stories/             # Development stories (active/, completed/)
-├── packages/                # Shared packages
-├── pro/                     # Pro submodule (proprietary)
-├── squads/                  # Squad expansions
-└── tests/                   # Testes
-```
-
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Framework vs Project Boundary
-
-O SINAPSE usa um modelo de 4 camadas (L1-L4) para separar artefatos do framework e do projeto. Deny rules em `.claude/settings.json` reforçam isso deterministicamente.
-
-| Camada | Mutabilidade | Paths | Notas |
-|--------|-------------|-------|-------|
-| **L1** Framework Core | NEVER modify | `.sinapse-ai/core/`, `.sinapse-ai/constitution.md`, `bin/sinapse.js`, `bin/sinapse-init.js` | Protegido por deny rules |
-| **L2** Framework Templates | NEVER modify | `.sinapse-ai/development/tasks/`, `.sinapse-ai/development/templates/`, `.sinapse-ai/development/checklists/`, `.sinapse-ai/development/workflows/`, `.sinapse-ai/infrastructure/` | Extend-only |
-| **L3** Project Config | Mutable (exceptions) | `.sinapse-ai/data/`, `agents/*/MEMORY.md`, `core-config.yaml` | Allow rules permitem |
-| **L4** Project Runtime | ALWAYS modify | `docs/stories/`, `packages/`, `squads/`, `tests/` | Trabalho do projeto |
-
-**Toggle:** `core-config.yaml` → `boundary.frameworkProtection: true/false` controla se deny rules são ativas (default: true para projetos, false para contribuidores do framework).
-
-> **Referência formal:** `.claude/settings.json` (deny/allow rules), `.claude/rules/agent-authority.md`
+## CLI First
 
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Sistema de Agentes
-
-### Ativação de Agentes
-Use `@agent-name` ou `/SINAPSE:agents:agent-name`:
-
-| Agente | Persona | Escopo Principal |
-|--------|---------|------------------|
-| `@developer` | Pixel | Implementacao de codigo |
-| `@quality-gate` | Litmus | Testes e qualidade |
-| `@architect` | Stratum | Arquitetura e design tecnico |
-| `@project-lead` | Beacon | Product Management |
-| `@product-lead` | Axis | Product Owner, stories/epics |
-| `@sprint-lead` | Sync | Scrum Master |
-| `@analyst` | Scope | Pesquisa e analise |
-| `@data-engineer` | Tensor | Database design |
-| `@ux-design-expert` | Mosaic | UX/UI design |
-| `@devops` | Pipeline | CI/CD, git push (EXCLUSIVO) |
-
-### Comandos de Agentes
-Use prefixo `*` para comandos:
-- `*help` - Mostrar comandos disponíveis
-- `*create-story` - Criar story de desenvolvimento
-- `*task {name}` - Executar task específica
-- `*exit` - Sair do modo agente
-
-### Mapeamento Agente → Codebase
+`CLI First → Observability Second → UI Third`. New features MUST work 100% via CLI before any UI.
 
-| Agente | Diretórios Principais |
-|--------|----------------------|
-| `@developer` | `packages/`, `.sinapse-ai/core/`, `bin/` |
-| `@architect` | `docs/architecture/`, system design |
-| `@data-engineer` | `packages/db/`, migrations, schema |
-| `@quality-gate` | `tests/`, `*.test.js`, quality gates |
-| `@product-lead` | Stories, epics, requirements |
-| `@devops` | `.github/`, CI/CD, git operations |
+## Project Structure
 
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Documentation-First Development (NON-NEGOTIABLE)
-
-1. **Pipeline automatico** - Todo briefing segue: Epic → Story → Validacao → Implementacao
-2. **Story validada obrigatoria** - Nenhum codigo sem story com status >= Ready
-3. **Trabalhe a partir de stories** - Todo desenvolvimento começa com uma story em `docs/stories/`
-4. **Atualize progresso** - Marque checkboxes conforme completa: `[ ]` → `[x]`
-5. **Rastreie mudanças** - Mantenha a seção File List na story
-6. **Siga critérios** - Implemente exatamente o que os acceptance criteria especificam
-
-### Workflow de Story
 ```
-@product-lead *create-story → @developer implementa → @quality-gate testa → @devops push
+.sinapse-ai/           # Framework core, data, agents, tasks, templates
+bin/                   # CLI executables
+docs/stories/          # Development stories
+packages/              # Shared packages
+squads/                # Squad expansions
+tests/                 # Tests
 ```
 
----
-
-<!-- PROJECT-CUSTOMIZED: Safe to modify for your project -->
-## Padrões de Código
+## Framework Boundary (L1-L4)
 
-### Convenções de Nomenclatura
+| Layer | Mutability | Key Paths |
+|-------|-----------|-----------|
+| L1 Core | NEVER | `.sinapse-ai/core/`, `bin/sinapse*.js` |
+| L2 Templates | NEVER | `.sinapse-ai/development/{tasks,templates,checklists,workflows}/` |
+| L3 Config | Mutable | `.sinapse-ai/data/`, `core-config.yaml` |
+| L4 Runtime | ALWAYS | `docs/stories/`, `packages/`, `squads/`, `tests/` |
 
-| Tipo | Convenção | Exemplo |
-|------|-----------|---------|
-| Componentes | PascalCase | `WorkflowList` |
-| Hooks | prefixo `use` | `useWorkflowOperations` |
-| Arquivos | kebab-case | `workflow-list.tsx` |
-| Constantes | SCREAMING_SNAKE_CASE | `MAX_RETRIES` |
-| Interfaces | PascalCase + sufixo | `WorkflowListProps` |
+## Agents
 
-### Imports
-**Sempre use imports absolutos.** Nunca use imports relativos.
-```typescript
-// ✓ Correto
-import { useStore } from '@/stores/feature/store'
+Use `@agent-name` or `/SINAPSE:agents:agent-name`. Commands: `*help`, `*create-story`, `*task {name}`, `*exit`.
 
-// ✗ Errado
-import { useStore } from '../../../stores/feature/store'
-```
-
-**Ordem de imports:**
-1. React/core libraries
-2. External libraries
-3. UI components
-4. Utilities
-5. Stores
-6. Feature imports
-7. CSS imports
+| Agent | Persona | Scope |
+|-------|---------|-------|
+| @developer | Pixel | Code implementation |
+| @quality-gate | Litmus | Testing & quality |
+| @architect | Stratum | Architecture |
+| @project-lead | Beacon | Product Management |
+| @product-lead | Axis | Product Owner |
+| @sprint-lead | Sync | Scrum Master |
+| @analyst | Scope | Research |
+| @data-engineer | Tensor | Database |
+| @ux-design-expert | Mosaic | UX/UI |
+| @devops | Pipeline | CI/CD, push (EXCLUSIVE) |
 
-### TypeScript
-- Sem `any` - Use tipos apropriados ou `unknown` com type guards
-- Sempre defina interface de props para componentes
-- Use `as const` para objetos/arrays constantes
-- Tipos de ref explícitos: `useRef<HTMLDivElement>(null)`
+## Documentation-First (NON-NEGOTIABLE)
 
-### Error Handling
-```typescript
-try {
-  // Operation
-} catch (error) {
-  logger.error(`Failed to ${operation}`, { error })
-  throw new Error(`Failed to ${operation}: ${error instanceof Error ? error.message : 'Unknown'}`)
-}
-```
+Pipeline: Epic → Story → Validation → Implementation. No code without a story at status >= Ready.
 
----
+## Code Patterns
 
-<!-- PROJECT-CUSTOMIZED: Safe to modify for your project -->
-## Testes & Quality Gates
+- **Naming:** PascalCase components, `use` prefix hooks, kebab-case files, SCREAMING_SNAKE constants
+- **Imports:** Always absolute (`@/...`), never relative. Order: React → external → UI → utils → stores → features → CSS
+- **TypeScript:** No `any`, always define props interfaces, `as const` for constants
+- **Commits:** Conventional Commits + story ref: `feat: feature [Story 2.1]`
+- **Push:** Only `@devops` can push to remote
 
-### Comandos de Teste
-```bash
-npm test                    # Rodar testes
-npm run test:coverage       # Testes com cobertura
-npm run lint                # ESLint
-npm run typecheck           # TypeScript
-```
+## Tests
 
-### Quality Gates (Pre-Push)
-Antes de push, todos os checks devem passar:
 ```bash
-npm run lint        # ESLint
-npm run typecheck   # TypeScript
-npm test            # Jest
+npm test && npm run lint && npm run typecheck
 ```
 
----
-
-<!-- PROJECT-CUSTOMIZED: Safe to modify for your project -->
-## Convenções Git
-
-### Commits
-Seguir Conventional Commits:
-- `feat:` - Nova funcionalidade
-- `fix:` - Correção de bug
-- `docs:` - Documentação
-- `test:` - Testes
-- `chore:` - Manutenção
-- `refactor:` - Refatoração
-
-**Referencie story ID:** `feat: implement feature [Story 2.1]`
-
-### Branches
-- `main` - Branch principal
-- `feat/*` - Features
-- `fix/*` - Correções
-- `docs/*` - Documentação
-
-### Push Authority
-**Apenas `@devops` pode fazer push para remote.**
+## Tool Usage
 
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## Otimização Claude Code
-
-### Uso de Ferramentas
-| Tarefa | Use | Não Use |
-|--------|-----|---------|
-| Buscar conteúdo | `Grep` tool | `grep`/`rg` no bash |
-| Ler arquivos | `Read` tool | `cat`/`head`/`tail` |
-| Editar arquivos | `Edit` tool | `sed`/`awk` |
-| Buscar arquivos | `Glob` tool | `find` |
-| Operações complexas | `Task` tool | Múltiplos comandos manuais |
+Use Grep (not grep), Read (not cat), Edit (not sed), Glob (not find). Prefer native tools over MCP. Tool registry: `.sinapse-ai/data/tool-registry.yaml`.
 
-### Performance
-- Prefira chamadas de ferramentas em batch
-- Use execução paralela para operações independentes
-- Cache dados frequentemente acessados durante a sessão
+## Context Management
 
-### Context Management (NOG-18 + TOK-4A)
-- Use `/compact` when context feels heavy or approaching limits
-- Context brackets (SYNAPSE) replaced by native compaction
-- Agent memory persists in `.sinapse-ai/development/agents/{id}/MEMORY.md`
 - Rules with `paths:` frontmatter only load when working on matching files
-- **Agent Handoff:** On agent switch (`@agent`), previous agent is compacted to ~379-token handoff artifact. See `.claude/rules/agent-handoff.md` for full protocol. Max 3 retained summaries, oldest discarded.
-
-### Gerenciamento de Sessão
-- Rastreie progresso da story durante a sessão
-- Atualize checkboxes imediatamente após completar tasks
-- Mantenha contexto da story atual sendo trabalhada
-- Salve estado importante antes de operações longas
-
-### Recuperação de Erros
-- Sempre forneça sugestões de recuperação para falhas
-- Inclua contexto do erro em mensagens ao usuário
-- Sugira procedimentos de rollback quando apropriado
-- Documente quaisquer correções manuais necessárias
-
----
-
-<!-- PROJECT-CUSTOMIZED: Safe to modify for your project -->
-## Comandos Frequentes
-
-### Desenvolvimento
-```bash
-npm run dev                 # Iniciar desenvolvimento
-npm test                    # Rodar testes
-npm run lint                # Verificar estilo
-npm run typecheck           # Verificar tipos
-npm run build               # Build produção
-```
-
-### SINAPSE
-```bash
-npx sinapse-ai install       # Instalar SINAPSE
-npx sinapse-ai doctor        # Diagnóstico do sistema
-npx sinapse-ai info          # Informações do sistema
-```
-
-### Dashboard (apps/dashboard/)
-```bash
-cd apps/dashboard
-npm install
-npm run dev                 # Desenvolvimento
-npm run build               # Build produção
-```
-
----
-
-<!-- FRAMEWORK-OWNED: Generated by SINAPSE installer, do not customize -->
-## MCP Usage
-
-Ver `.claude/rules/mcp-usage.md` para regras detalhadas.
-
-**Resumo:**
-- Preferir ferramentas nativas do Claude Code sobre MCP
-- MCP Docker Gateway apenas quando explicitamente necessário
-- `@devops` gerencia toda infraestrutura MCP
-
----
+- Agent handoff compacts to ~379 tokens on switch
+- Agent memory in `.sinapse-ai/development/agents/{id}/MEMORY.md`
+- **Memory as hints:** Memory entries are hints, NOT ground truth. Always verify against actual codebase before acting on remembered facts.
 
-<!-- PROJECT-CUSTOMIZED: Token Optimization TOK-2 -->
-## Tool Selection Guidance
+## Token Economy & Response Format (NON-NEGOTIABLE)
 
-Prefer native Claude Code tools over MCP for common operations. The tool-registry at `.sinapse-ai/data/tool-registry.yaml` defines the 3-Tier Tool Mesh:
+Auto-applied to all agents: `~/.claude/rules/token-economy.md` + `~/.claude/rules/response-format.md`. Compact at 60%, model route haiku/sonnet/opus, no preamble, no trailing summary.
 
-| Tier | When Loaded | Examples |
-|------|-------------|---------|
-| **1** (Always) | Session start | Read, Write, Edit, Bash, Grep, Glob, Task |
-| **2** (Deferred) | Agent activation | git, coderabbit, context7, supabase |
-| **3** (Deferred) | Via tool search | EXA, Playwright, Apify, Nogic, Code-Graph |
+## Delegation & Anti-Hallucination
 
-**Guidelines:**
-- Use Tier 1 native tools for file ops, search, and commands — avoid MCP equivalents
-- Tier 3 MCP tools are deferred via Tool Search — only loaded when explicitly needed
-- Limit tool search to maximum 2 searches per turn to avoid overhead
-- Essential MCP servers (nogic, code-graph) are never disabled
-- Non-essential servers (EXA, Apify, Playwright) load on-demand via tool search
-
-**Tool Examples (TOK-4B):** Concrete input examples improve tool selection accuracy by +18pp. See `.claude/rules/tool-examples.md` for examples of top-10 tools. Full registry: `.sinapse-ai/data/mcp-tool-examples.yaml`. ADR-5: examples for always-loaded tools only, search for deferred tools.
-
-**Runtime capabilities:** `.sinapse/runtime-capabilities.json` (generated by capability-detection.js)
-
----
-
-<!-- PROJECT-CUSTOMIZED: Safe to modify for your project -->
-## Debug
-
-### Habilitar Debug
-```bash
-export SINAPSE_DEBUG=true
-```
-
-### Logs
-```bash
-tail -f .sinapse/logs/agent.log
-```
+- Persona switch for sequential work, sub-agent only for parallel (20K+ tokens each)
+- Model routing: `haiku` routine, `sonnet` standard, `opus` complex
+- Sub-agents announce their model for visual verification via statusline
+- `npm view {pkg}` before adding deps. Cite file:line for claims.
+- Mark uncertain claims with [NEEDS VERIFICATION]. Compact at 60%.
 
 ---
-
-*SINAPSE Claude Code Configuration v4.0*
-*CLI First | Observability Second | UI Third*
+*SINAPSE v6.0 — CLI First | Observability Second | UI Third*

package/.claude/hooks/enforce-git-push-authority.sh

@@ -1,13 +1,34 @@
 #!/bin/bash
 # enforce-git-push-authority.sh
-# PreToolUse hook: blocks "git push" commands in Bash tool
-# Only meant to run when agent is NOT @devops
+# PreToolUse hook: blocks "git push" commands in Bash tool unless active agent is @devops
+# Detects active agent via .sinapse/session-state.json (same pattern as enforce-delegation.cjs)
 # Uses node (not jq) for JSON parsing — works on Windows/Git Bash
-# FAIL-CLOSED: if parsing fails, blocks the command (exit 2)
-# Hardened v2: also detects indirect execution via script files and pipes
+# FAIL-CLOSED: if parsing fails OR session-state is missing/unreadable, blocks the command
+# Hardened v3: adds real agent detection (Story 10.15), retains indirect-execution checks
 
 INPUT=$(cat)
 
+# Resolve project root consistently with other hooks
+PROJECT_ROOT="${CLAUDE_PROJECT_DIR:-$(pwd)}"
+
+# Detect active agent from session state (fail-closed: empty string on any error)
+AGENT=$(node -e "
+  const fs = require('fs');
+  const path = require('path');
+  try {
+    const p = path.join(process.argv[1], '.sinapse', 'session-state.json');
+    const s = JSON.parse(fs.readFileSync(p, 'utf8'));
+    console.log(s.lastAgent || '');
+  } catch (e) {
+    console.log('');
+  }
+" "$PROJECT_ROOT" 2>/dev/null)
+
+# If active agent is @devops, allow the command immediately (AC 1)
+if [ "$AGENT" = "devops" ]; then
+  exit 0
+fi
+
 # Extract command from JSON using node (available on all SINAPSE systems)
 COMMAND=$(echo "$INPUT" | node -e "
   let d='';

package/.claude/hooks/verify-packages.cjs

@@ -0,0 +1,83 @@
+#!/usr/bin/env node
+/**
+ * verify-packages.cjs — Slopsquatting Prevention Hook
+ *
+ * Blocks `npm install`/`npm add` commands that reference packages
+ * not found on the npm registry. Prevents installation of hallucinated
+ * (fabricated) packages that attackers may register with malicious code.
+ *
+ * Research: 19.7% of packages recommended by LLMs are fabricated.
+ * Source: arXiv study, 576K samples across 16 models.
+ *
+ * Hook type: PreToolUse (Bash)
+ * Exit 0 = allow, Exit 2 = block
+ */
+'use strict';
+
+const { execSync } = require('child_process');
+
+let input = '';
+process.stdin.setEncoding('utf8');
+process.stdin.on('data', (d) => { input += d; });
+process.stdin.on('end', () => {
+  try {
+    const data = JSON.parse(input);
+    const command = (data.tool_input && data.tool_input.command) || '';
+
+    // Only check direct npm install/add commands (not text in PR bodies, etc.)
+    // Skip if command is gh, curl, echo, or other non-npm commands
+    const trimmed = command.trim();
+    if (trimmed.startsWith('gh ') || trimmed.startsWith('curl ') || trimmed.startsWith('echo ')) process.exit(0);
+    const installMatch = trimmed.match(/^npm\s+(install|add|i)\s+(.+)/m);
+    if (!installMatch) process.exit(0);
+
+    const argsStr = installMatch[2];
+
+    // Extract package names, skipping flags (--save-dev, -D, etc.)
+    const tokens = argsStr.split(/\s+/).filter(t => !t.startsWith('-'));
+    if (tokens.length === 0) process.exit(0);
+
+    const failed = [];
+    for (const token of tokens) {
+      // Strip version specifier: pkg@1.0.0 -> pkg, @org/pkg@^2 -> @org/pkg
+      let pkgName;
+      if (token.startsWith('@')) {
+        // Scoped package: @org/pkg@version
+        const slashIdx = token.indexOf('/');
+        if (slashIdx === -1) continue; // malformed, skip
+        const afterSlash = token.substring(slashIdx + 1);
+        const atIdx = afterSlash.indexOf('@');
+        pkgName = atIdx > 0
+          ? token.substring(0, slashIdx + 1 + atIdx)
+          : token;
+      } else {
+        const atIdx = token.indexOf('@');
+        pkgName = atIdx > 0 ? token.substring(0, atIdx) : token;
+      }
+
+      // Skip if it looks like a local path or URL
+      if (pkgName.startsWith('.') || pkgName.startsWith('/') || pkgName.includes('://')) continue;
+      if (pkgName.endsWith('.tgz') || pkgName.endsWith('.tar.gz')) continue;
+
+      try {
+        execSync(`npm view "${pkgName}" name`, { timeout: 8000, stdio: 'pipe' });
+      } catch {
+        failed.push(pkgName);
+      }
+    }
+
+    if (failed.length > 0) {
+      const names = failed.map(n => `'${n}'`).join(', ');
+      const msg = failed.length === 1
+        ? `BLOCKED: Package ${names} not found on npm. This may be a hallucinated package (slopsquatting).`
+        : `BLOCKED: Packages ${names} not found on npm. These may be hallucinated packages (slopsquatting).`;
+      process.stderr.write(msg + '\n');
+      process.exit(2);
+    }
+
+    process.exit(0);
+  } catch {
+    // Fail-open: if hook crashes, allow the operation
+    process.exit(0);
+  }
+});

package/.claude/rules/agent-authority.md

@@ -1,3 +1,9 @@
+---
+paths:
+  - ".sinapse-ai/development/agents/**"
+  - "docs/stories/**"
+---
+
 # Agent Authority — Detailed Rules
 
 ## Delegation Matrix

package/.claude/rules/agent-handoff.md

@@ -1,3 +1,8 @@
+---
+paths:
+  - ".sinapse-ai/development/agents/**"
+---
+
 # Agent Handoff Protocol — Context Compaction
 
 ## Purpose

package/.claude/rules/cross-squad-routing.md

@@ -1,3 +1,8 @@
+---
+paths:
+  - "squads/**"
+---
+
 # Cross-Squad Routing Rules
 
 > Applies to Imperator (sinapse-orqx) and ALL squad orchestrators (*-orqx).

package/.claude/rules/hook-governance.md

@@ -1,3 +1,9 @@
+---
+paths:
+  - ".claude/hooks/**"
+  - ".claude/settings.json"
+---
+
 # Hook Governance Rules
 
 > Applies to ALL agents. Hooks are the enforcement layer of the Constitution.
@@ -8,6 +14,7 @@
 | Hook | Purpose | Behavior |
 |------|---------|----------|
 | `enforce-git-push-authority.sh` | Art. II — Only @devops can push | BLOCK (deny) |
+| `verify-packages.cjs` | Security — Block hallucinated npm packages (slopsquatting) | BLOCK (exit 2) |
 | `sql-governance.py` | Security — Block dangerous SQL | BLOCK (exit 2) |
 | `enforce-delegation.cjs` | Art. VIII — Orchestrators can't execute | BLOCK (exit 2) |
 

package/.claude/rules/mandatory-delegation.md

@@ -91,6 +91,30 @@ Everything OUTSIDE their orchestration domain MUST be delegated.
 - Saying "vou fazer isso eu mesmo" instead of delegating
 - Absorbing a request and executing it instead of routing
 
+## Universal Auto-Routing (users should NEVER call agents manually)
+
+Users are NOT AI experts. The system MUST understand natural language and route automatically.
+
+**On EVERY user message (not just orchestrator):**
+1. Detect the domain of the request
+2. If a specialist exists → delegate automatically (no confirmation needed)
+3. Brief acknowledgment: "Delegando para @specialist..."
+4. Return the result to the user
+
+**Auto-detect project state on first interaction:**
+- Check for `.sinapse-ai/` → SINAPSE-managed (continue SDC)
+- Check for `package.json` or `.git` → Brownfield (run quick tech scan first)
+- Empty directory → Greenfield (ask project type, scaffold)
+
+**Cross-agent handoff (automatic, never ask user):**
+- Agent needs git push → auto-delegate to @devops
+- Agent needs tests → auto-delegate to @quality-gate
+- Agent needs schema → auto-delegate to @data-engineer
+- Agent needs story → auto-create via fast-track or @sprint-lead
+- Agent needs architecture decision → auto-delegate to @architect
+
 ## Enforcement
 
 Any response from an orchestrator that contains direct domain work (code, schema, copy, etc.) without having first delegated to the appropriate specialist is a **constitutional violation** and must be corrected immediately.
+
+Any response that asks the user to manually invoke an agent (showing `@agent-name` or `/SINAPSE:agents:...` commands) instead of auto-routing is a **UX violation** — the system should just do it.

package/.claude/rules/mcp-usage.md

@@ -1,5 +1,7 @@
 ---
-paths: **/*
+paths:
+  - ".mcp.json"
+  - ".sinapse-ai/data/tool-registry.yaml"
 ---
 
 # MCP Server Usage Rules - SINAPSE Architecture