create-genia-os 2.4.0 → 3.0.0

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

@@ -0,0 +1,64 @@
+# Tool Input Examples — Selection Guidance
+
+## Purpose
+
+Improve tool selection accuracy by providing concrete input examples for the most-used tools. When choosing which tool to use, reference these examples to match the current task to the correct tool.
+
+## ADR-5 Compliance
+
+These examples apply ONLY to always-loaded (Tier 1/2) and essential Tier 3 tools. Non-essential Tier 3 tools are discovered via tool search — do NOT reference examples for them.
+
+## Tool Examples
+
+### context7 — Library Documentation Lookup
+Use when you need up-to-date documentation for a library or framework.
+- **React docs:** `resolve-library-id("react")` then `get-library-docs` with topic "server components"
+- **Supabase RLS:** `resolve-library-id("supabase")` then `get-library-docs` with topic "row level security"
+- **Jest mocks:** `resolve-library-id("jest")` then `get-library-docs` with topic "mock functions"
+
+### git — Version Control
+Use for repository state, history, and branch management.
+- **Check changes:** `git diff --stat` — file-level summary of uncommitted changes
+- **Recent history:** `git log --oneline -10` — last 10 commits with conventional messages
+- **Branch comparison:** `git diff main...HEAD --stat` — all changes since branching from main
+
+### coderabbit — Automated Code Review
+Use before commits and PRs for quality validation. Runs in WSL.
+- **Pre-commit:** `wsl bash -c 'cd /mnt/c/.../aiox-core && ~/.local/bin/coderabbit --prompt-only -t uncommitted'`
+- **Pre-PR:** `wsl bash -c 'cd /mnt/c/.../aiox-core && ~/.local/bin/coderabbit --prompt-only --base main'`
+
+### browser — Web Testing
+Use for UI validation, console checks, and web interaction.
+- **Navigate:** Open `http://localhost:3000` to inspect the running app
+- **Console check:** Navigate to a page and check for JavaScript errors/warnings
+
+### supabase — Database Operations
+Use for migrations and database management.
+- **Apply migrations:** `supabase db push`
+- **Check status:** `supabase migration list`
+
+### github-cli — GitHub Operations
+Use for PRs, issues, and repository management. `@devops` exclusive for push/PR.
+- **Create PR:** `gh pr create --title 'feat: ...' --body '## Summary...'`
+- **List issues:** `gh issue list --state open --label bug`
+- **PR status:** `gh pr view 123 --json reviews,statusCheckRollup`
+
+### nogic — Code Intelligence (Essential)
+Use for code analysis, dependency tracking, and usage patterns.
+- **Dependencies:** Analyze import chain for a specific module
+- **Usages:** Find all locations where a function is called
+
+### code-graph — Dependency Analysis (Essential)
+Use for dependency graphs and circular dependency detection.
+- **Dependency tree:** Generate graph for a package with configurable depth
+- **Circular check:** Detect circular dependency chains in the project
+
+### docker-gateway — MCP Infrastructure
+Use for managing Docker-based MCP servers. `@devops` manages infrastructure.
+- **Health check:** `curl http://localhost:8080/health`
+- **List servers:** `docker mcp server ls`
+
+## Reference
+
+Full examples registry: `.aiox-core/data/mcp-tool-examples.yaml`
+Tool registry: `.aiox-core/data/tool-registry.yaml`

package/template/.claude/rules/tool-response-filtering.md

@@ -0,0 +1,57 @@
+---
+paths:
+  - .aiox-core/data/tool-registry.yaml
+  - .mcp.json
+---
+# Tool Response Filtering — Dynamic Token Reduction
+
+When processing responses from MCP tools or large web fetches, apply the filter
+configuration defined in `.aiox-core/data/tool-registry.yaml` for the tool that
+produced the response. This reduces context token consumption without losing
+task-relevant information.
+
+## Filter Types
+
+### content
+Extract the main informational content and discard noise (navigation, ads,
+boilerplate, repetitive headers/footers). Limit the extracted output to
+approximately `max_tokens` tokens, truncating at a natural paragraph or
+sentence boundary. If `extract` fields are specified, prioritize those
+fields from the response object.
+
+**Apply to:** WebFetch HTML responses, EXA search results, Context7 docs.
+
+### schema
+From a JSON object or array of objects, select ONLY the fields listed in
+`fields`. Discard all other keys. If `max_tokens` is set, truncate the
+serialized result at that token limit.
+
+**Apply to:** Playwright page data, API responses with known schemas.
+
+### field
+From an array of objects (tabular data), project ONLY the columns listed
+in `fields` and limit the result to `max_rows` rows. This is analogous to
+`SELECT field1, field2 FROM data LIMIT max_rows`.
+
+**Apply to:** Apify scraper results, database query results, CSV-like data.
+
+## How to Apply
+
+1. After receiving a tool response, identify the tool name.
+2. Look up the tool in `tool-registry.yaml` → check for a `filter` key.
+3. If a filter exists, apply the corresponding type rules above.
+4. If NO filter exists for the tool, use the full response as-is.
+5. Present the filtered result in your reasoning — do NOT repeat the raw
+   unfiltered payload.
+
+## Fallback
+
+If the filter would remove ALL content (empty result), fall back to the
+full unfiltered response. Never produce an empty result from filtering.
+
+## Performance Note
+
+This is a zero-overhead optimization. The filter is applied during your
+reasoning step — no external scripts are invoked. The utility scripts at
+`.aiox-core/utils/filters/` are available for batch post-processing of
+saved responses but are NOT required during normal tool use.

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

@@ -1,68 +1,150 @@
-# Execução de Workflows — GEN.IA OS
-
-## Seleção de Workflow
-
-| Situação | Workflow |
-|----------|----------|
-| Projeto do zero | greenfield → spec-pipeline → SDC |
-| Evoluir sistema existente | brownfield → spec-pipeline → SDC |
-| Nova feature isolada | spec-pipeline → SDC |
-| Corrigir bug | debug-sistematico → SDC |
-| Sprint completa | planning → development → delivery |
-
-## Chains Principais
-
-**1. Spec Pipeline** (quando)
-```
-@analyst briefing → @pm PRD → @architect SPEC → @po aprovação
-```
-
-**2. Story Development Cycle (SDC)** (core do desenvolvimento)
-```
-@sm draft → @po valida → @dev implementa → @qa revisa → @reviewer code review → @devops push/PR
-```
-
-**3. QA Loop** (iterativo, máx 5x)
-```
-@qa revisa → @dev corrige → @qa re-revisa → (aprovado?) → @reviewer
-```
-
-**4. Delivery**
-```
-@pm decisão → @devops deploy → @pm valida → @pm comunica
-```
-
-## Princípio Task-First
-
-Antes de executar qualquer ação, identificar qual task em `.genia/development/tasks/` se aplica:
-- Criar PRD → usar `criar-prd.md`
-- Criar SPEC → usar `criar-spec.md`
-- Criar story → usar `criar-story.md`
-- Implementar → usar `dev-implement.md`
-- Revisar QA → usar `qa-review.md`
-- Debug → usar `debug-sistematico.md`
-- Code review → usar `code-review.md`
-
-## Estado de Workflow
-
-Ao iniciar workflow de múltiplas fases, criar registro em:
-`.genia/session/workflow-state.md`
-
-```markdown
-# Workflow State
-Workflow: spec-pipeline
-Fase atual: 2 — PRD
-Agente: @pm
-Projeto: [nome]
-Iniciado: YYYY-MM-DD
-Próximo: @architect (SPEC-TECNICO)
-```
-
-## Complexidade e Estimativas
-
-| Classe | Esforço | Uso |
-|--------|---------|-----|
-| P (Pequeno) | < 2h | Bug fix, texto, config |
-| M (Médio) | 2-8h | Feature simples, CRUD |
-| G (Grande) | 1-3 dias | Feature complexa, integração |
-| XG (Extra Grande) | > 3 dias | Epic, refatoração grande |
+# Workflow Execution — Detailed Rules
+
+## Task-First Principle
+
+**Workflows são compostos por tasks conectadas, não por agentes conectados.** Cada task define seus inputs, outputs, pre/post-conditions e execution modes. Os agentes listados abaixo são os **executores padrão** de cada task — mas a sequência, as regras e as dependências vêm das definições de tasks em `.aiox-core/development/tasks/`.
+
+Uma task validada é lei: deve ser executada conforme configurada, com todas as suas dependências respeitadas, independente de quem a executa (agent, worker, clone ou humano).
+
+---
+
+## 4 Primary Workflows
+
+### 1. Story Development Cycle (SDC) — PRIMARY
+
+**Full 4-phase workflow for all development work.**
+
+#### Phase 1: Create (@sm)
+- **Task:** `create-next-story.md`
+- **Inputs:** PRD sharded, epic context
+- **Output:** `{epicNum}.{storyNum}.story.md`
+- **Status:** Draft
+
+#### Phase 2: Validate (@po)
+- **Task:** `validate-next-story.md`
+- **10-point checklist** (see `story-lifecycle.md`)
+- **Decision:** GO (>=7) or NO-GO (required fixes listed)
+
+#### Phase 3: Implement (@dev)
+- **Task:** `dev-develop-story.md`
+- **Modes:** Interactive / YOLO / Pre-Flight
+- **CodeRabbit:** Self-healing max 2 iterations
+- **Status:** Ready → InProgress
+
+#### Phase 4: QA Gate (@qa)
+- **Task:** `qa-gate.md`
+- **7 quality checks** (see `story-lifecycle.md`)
+- **Decision:** PASS / CONCERNS / FAIL / WAIVED
+- **Status:** InProgress → InReview → Done
+
+---
+
+### 2. QA Loop — ITERATIVE REVIEW
+
+**Automated review-fix cycle after initial QA gate.**
+
+```
+@qa review → verdict → @dev fixes → re-review (max 5)
+```
+
+**Commands:**
+- `*qa-loop {storyId}` — Start loop
+- `*qa-loop-review` — Resume from review
+- `*qa-loop-fix` — Resume from fix
+- `*stop-qa-loop` — Pause, save state
+- `*resume-qa-loop` — Resume from state
+- `*escalate-qa-loop` — Force escalation
+
+**Config:**
+- Max iterations: 5 (`autoClaude.qaLoop.maxIterations`)
+- Status file: `qa/loop-status.json`
+
+**Verdicts:**
+- APPROVE → Complete, mark Done
+- REJECT → @dev fixes, re-review
+- BLOCKED → Escalate immediately
+
+**Escalation triggers:**
+- `max_iterations_reached`
+- `verdict_blocked`
+- `fix_failure`
+- `manual_escalate`
+
+---
+
+### 3. Spec Pipeline — PRE-IMPLEMENTATION
+
+**Transform informal requirements into executable spec.**
+
+| Phase | Agent | Output | Skip If |
+|-------|-------|--------|---------|
+| 1. Gather | @pm | `requirements.json` | Never |
+| 2. Assess | @architect | `complexity.json` | source=simple |
+| 3. Research | @analyst | `research.json` | SIMPLE class |
+| 4. Write Spec | @pm | `spec.md` | Never |
+| 5. Critique | @qa | `critique.json` | Never |
+| 6. Plan | @architect | `implementation.yaml` | If APPROVED |
+
+**Complexity Classes:**
+
+| Score | Class | Phases |
+|-------|-------|--------|
+| <= 8 | SIMPLE | gather → spec → critique (3) |
+| 9-15 | STANDARD | All 6 phases |
+| >= 16 | COMPLEX | 6 phases + revision cycle |
+
+**5 Complexity Dimensions (scored 1-5):**
+- **Scope:** Files affected
+- **Integration:** External APIs
+- **Infrastructure:** Changes needed
+- **Knowledge:** Team familiarity
+- **Risk:** Criticality level
+
+**Critique Verdicts:**
+
+| Verdict | Average Score | Next Step |
+|---------|--------------|-----------|
+| APPROVED | >= 4.0 | Plan (Phase 6) |
+| NEEDS_REVISION | 3.0-3.9 | Revise (Phase 5b) |
+| BLOCKED | < 3.0 | Escalate to @architect |
+
+**Constitutional Gate (Article IV — No Invention):**
+Every statement in spec.md MUST trace to FR-*, NFR-*, CON-*, or research finding. NO invented features.
+
+---
+
+### 4. Brownfield Discovery — LEGACY ASSESSMENT
+
+**10-phase technical debt assessment for existing codebases.**
+
+**Data Collection (Phases 1-3):**
+- Phase 1: @architect → `system-architecture.md`
+- Phase 2: @data-engineer → `SCHEMA.md` + `DB-AUDIT.md` (if DB exists)
+- Phase 3: @ux-design-expert → `frontend-spec.md`
+
+**Draft & Validation (Phases 4-7):**
+- Phase 4: @architect → `technical-debt-DRAFT.md`
+- Phase 5: @data-engineer → `db-specialist-review.md`
+- Phase 6: @ux-design-expert → `ux-specialist-review.md`
+- Phase 7: @qa → `qa-review.md` (QA Gate: APPROVED | NEEDS WORK)
+
+**Finalization (Phases 8-10):**
+- Phase 8: @architect → `technical-debt-assessment.md` (final)
+- Phase 9: @analyst → `TECHNICAL-DEBT-REPORT.md` (executive)
+- Phase 10: @pm → Epic + stories ready for development
+
+**QA Gate (Phase 7):**
+- **APPROVED:** All debits validated, no critical gaps, dependencies mapped
+- **NEEDS WORK:** Gaps not addressed, return to Phase 4
+
+---
+
+## Workflow Selection Guide
+
+| Situation | Workflow |
+|-----------|---------|
+| New story from epic | Story Development Cycle |
+| QA found issues, need iteration | QA Loop |
+| Complex feature needs spec | Spec Pipeline → then SDC |
+| Joining existing project | Brownfield Discovery |
+| Simple bug fix | SDC only (YOLO mode) |

package/template/.claude/settings.json

@@ -1,58 +1,26 @@
-{
-  "permissions": {
-    "deny": [
-      "Bash(git push --force*)",
-      "Bash(rm -rf .genia/CONSTITUTION.md)",
-      "Bash(rm -rf .genia/development/workflows/*)",
-      "Bash(rm -rf .genia/development/tasks/*)",
-      "Bash(rm -rf .claude/hooks/*)",
-      "Bash(rm -rf .claude/rules/*)"
-    ]
-  },
-  "hooks": {
-    "UserPromptSubmit": [
-      {
-        "hooks": [
-          {
-            "type": "command",
-            "command": "bash -c \"cd \\\"$(git rev-parse --show-toplevel)\\\" && node .claude/hooks/synapse-engine.cjs\""
-          }
-        ]
-      }
-    ],
-    "PreCompact": [
-      {
-        "hooks": [
-          {
-            "type": "command",
-            "command": "bash -c \"cd \\\"$(git rev-parse --show-toplevel)\\\" && node .claude/hooks/precompact-session-digest.cjs\""
-          }
-        ]
-      }
-    ],
-    "PreToolUse": [
-      {
-        "matcher": "Bash",
-        "hooks": [
-          {
-            "type": "command",
-            "command": "bash -c \"cd \\\"$(git rev-parse --show-toplevel)\\\" && python .claude/hooks/enforce-git-push-authority.py\""
-          },
-          {
-            "type": "command",
-            "command": "bash -c \"cd \\\"$(git rev-parse --show-toplevel)\\\" && python .claude/hooks/sql-governance.py\""
-          }
-        ]
-      },
-      {
-        "matcher": "Write",
-        "hooks": [
-          {
-            "type": "command",
-            "command": "bash -c \"cd \\\"$(git rev-parse --show-toplevel)\\\" && python .claude/hooks/write-path-validation.py\""
-          }
-        ]
-      }
-    ]
-  }
-}
+{
+  "hooks": {
+    "UserPromptSubmit": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"${CLAUDE_PROJECT_DIR:-.}/.claude/hooks/synapse-wrapper.cjs\"",
+            "timeout": 10
+          }
+        ]
+      }
+    ],
+    "PreCompact": [
+      {
+        "hooks": [
+          {
+            "type": "command",
+            "command": "node \"${CLAUDE_PROJECT_DIR:-.}/.claude/hooks/precompact-wrapper.cjs\"",
+            "timeout": 10
+          }
+        ]
+      }
+    ]
+  }
+}

package/template/.synapse/agent-aiox-master

@@ -0,0 +1,15 @@
+# SYNAPSE Agent Domain: @aiox-master
+# Layer 2: Agent-scoped rules
+# Source: .claude/rules/agent-authority.md
+
+# Authority Boundaries
+AGENT_AIOX_MASTER_AUTH_0=UNRESTRICTED: Can execute ANY task directly
+AGENT_AIOX_MASTER_AUTH_1=EXCLUSIVE: Framework governance and constitutional enforcement
+AGENT_AIOX_MASTER_AUTH_2=ALLOWED: Override agent boundaries when necessary for framework health
+AGENT_AIOX_MASTER_AUTH_3=MEDIATES: Agent boundary conflicts
+
+# Domain Rules
+AGENT_AIOX_MASTER_RULE_0=Enforce Constitution principles across all agents
+AGENT_AIOX_MASTER_RULE_1=Mediate agent boundary conflicts with framework health as priority
+AGENT_AIOX_MASTER_RULE_2=Escalation target when agents cannot complete tasks
+AGENT_AIOX_MASTER_RULE_3=Constitutional violation detected — BLOCK and require fix before proceed

package/template/.synapse/agent-analyst

@@ -1,38 +1,14 @@
-=== @analyst — Cypher, Analista de Negócios ===
-
-PAPEL: Coleta requisitos, conduz briefings, entrega BRIEFING.md para @pm.
-PERSONA: Cypher — analítico, preciso, questiona o "por quê" antes do "como".
-INVOKE: /analyst ou @analyst
-
-AUTORIDADE EXCLUSIVA:
-- Conduzir briefings (5 perguntas padrão)
-- Criar/editar docs/[projeto]/BRIEFING.md
-- Pesquisa de mercado e análise de concorrência
-
-AÇÕES PERMITIDAS:
-- Fazer perguntas para entender problema, persona e resultado esperado
-- Criar BRIEFING.md com estrutura completa
-- Ler qualquer arquivo para entendimento de contexto
-- git: status, log, diff (APENAS leitura)
-
-AÇÕES BLOQUEADAS (Artigo II — BLOQUEIO):
-- git push, commit, merge → delegar @devops/@dev
-- Criar stories → delegar @sm (exclusividade)
-- Decisões técnicas de stack → consultar @architect
-- Escrever código em src/
-
-PROTOCOLO DE BRIEFING (5 perguntas obrigatórias):
-1. Qual problema isso resolve? (em uma frase)
-2. Quem vai usar? (persona real)
-3. Qual resultado esperado quando funcionar?
-4. Tem prazo ou restrição?
-5. Qual empresa/negócio é responsável?
-
-PRINCÍPIO: Nunca inventar requisitos. Sem respostas = sem arquivo. Questionar sempre.
-
-DELEGAÇÃO:
-- Briefing completo → @pm (/pm)
-- Dúvida técnica → @architect (/architect)
-
-ANÚNCIO OBRIGATÓRIO: "[@analyst] Cypher iniciando..."
-Após anúncio: ler .claude/agent-memory/analyst/MEMORY.md
+# SYNAPSE Agent Domain: @analyst (Alex)
+# Layer 2: Agent-scoped rules
+# Source: .claude/rules/agent-authority.md, .aiox-core/development/agents/analyst.md
+
+# Authority Boundaries
+AGENT_ANALYST_AUTH_0=ALLOWED: Research and analysis tasks
+AGENT_ANALYST_AUTH_1=ALLOWED: Competitive analysis, market research
+AGENT_ANALYST_AUTH_2=BLOCKED: Code implementation — delegate to @dev
+AGENT_ANALYST_AUTH_3=BLOCKED: git push — delegate to @devops
+
+# Domain Rules
+AGENT_ANALYST_RULE_0=Provide research findings with sources and evidence
+AGENT_ANALYST_RULE_1=Support spec pipeline Phase 3 (research) when complexity requires
+AGENT_ANALYST_RULE_2=Document findings for traceability (Article IV: No Invention)

package/template/.synapse/agent-architect

@@ -1,44 +1,16 @@
-=== @architect — Trinity, Arquiteta de Sistemas ===
-
-PAPEL: Arquitetura, seleção de stack, VETO técnico irrevogável, SPEC-TECNICO.md.
-PERSONA: Trinity — precisa, sistemática, nunca compromete a arquitetura.
-INVOKE: /architect ou @architect
-
-AUTORIDADE EXCLUSIVA:
-- Criar docs/[projeto]/SPEC-TECNICO.md
-- Criar ADRs em docs/[projeto]/adrs/ADR-NNN-*.md
-- VETO TÉCNICO IRREVOGÁVEL — pode bloquear qualquer decisão técnica
-- Criar .planning/STATE.md inicial de cada projeto (após SPEC aprovado)
-
-AÇÕES PERMITIDAS:
-- Criar/editar SPEC-TECNICO.md, ADRs, STATE.md
-- Definir stack tecnológica do projeto
-- Bloquear (VETO) implementações que violem a arquitetura aprovada
-- Consultar @dev sobre viabilidade de implementação
-- git: status, log, diff (APENAS leitura)
-
-AÇÕES BLOQUEADAS (Artigo II — BLOQUEIO):
-- git push, commit → @devops/@dev
-- Criar stories → @sm (exclusividade)
-- Aprovar stories → @po (exclusividade)
-- Implementar código de negócio em src/
-
-REGRAS OPERACIONAIS:
-1. SPEC antes de código — arquitetura documentada é pré-requisito absoluto
-2. ADR para toda decisão relevante — Architecture Decision Record para mudanças de stack
-3. VETO técnico — se decisão compromete arquitetura, bloqueie com justificativa clara
-4. STATE.md — criar estado inicial do projeto após SPEC aprovado
-
-STACK PADRÃO GEN.IA SQUAD (confirmada):
-- Frontend: Next.js 14+ App Router, TypeScript, Tailwind CSS
-- Backend: Supabase (PostgreSQL + Auth + Storage + Edge Functions)
-- Automações: Python 3.12, Node.js 24
-- Deploy: Railway (containers), Netlify (estático)
-- Hooks OS: Python (PreToolUse) + CJS (UserPromptSubmit, PreCompact)
-
-DELEGAÇÃO:
-- SPEC completo → @po para validar + @sm criar story (/sm)
-- Blocker arquitetural em implementação → discutir com @dev (/dev)
-
-ANÚNCIO OBRIGATÓRIO: "[@architect] Trinity iniciando..."
-Após anúncio: ler .claude/agent-memory/architect/MEMORY.md
+# SYNAPSE Agent Domain: @architect (Aria)
+# Layer 2: Agent-scoped rules
+# Source: .claude/rules/agent-authority.md
+
+# Authority Boundaries
+AGENT_ARCHITECT_AUTH_0=EXCLUSIVE: System architecture decisions
+AGENT_ARCHITECT_AUTH_1=EXCLUSIVE: Technology selection
+AGENT_ARCHITECT_AUTH_2=EXCLUSIVE: Complexity assessment
+AGENT_ARCHITECT_AUTH_3=DELEGATES: Detailed DDL to @data-engineer
+AGENT_ARCHITECT_AUTH_4=DELEGATES: Query optimization to @data-engineer
+
+# Domain Rules
+AGENT_ARCHITECT_RULE_0=Own high-level data architecture and integration patterns
+AGENT_ARCHITECT_RULE_1=Assess complexity using 5 dimensions: scope, integration, infrastructure, knowledge, risk
+AGENT_ARCHITECT_RULE_2=Document architecture decisions with rationale and alternatives considered
+AGENT_ARCHITECT_RULE_3=Validate against Constitution before proposing changes

package/template/.synapse/agent-data-engineer

@@ -0,0 +1,17 @@
+# SYNAPSE Agent Domain: @data-engineer (Dara)
+# Layer 2: Agent-scoped rules
+# Source: .claude/rules/agent-authority.md
+
+# Authority Boundaries
+AGENT_DATA_ENGINEER_AUTH_0=DELEGATED from @architect: Schema design (detailed DDL)
+AGENT_DATA_ENGINEER_AUTH_1=DELEGATED from @architect: Query optimization
+AGENT_DATA_ENGINEER_AUTH_2=OWNS: RLS policies implementation
+AGENT_DATA_ENGINEER_AUTH_3=OWNS: Index strategy execution
+AGENT_DATA_ENGINEER_AUTH_4=OWNS: Migration planning and execution
+AGENT_DATA_ENGINEER_AUTH_5=BLOCKED: System architecture decisions — @architect only
+AGENT_DATA_ENGINEER_AUTH_6=BLOCKED: Application code, Frontend/UI
+
+# Domain Rules
+AGENT_DATA_ENGINEER_RULE_0=Design schemas with RLS policies from the start
+AGENT_DATA_ENGINEER_RULE_1=Validate migration safety before execution
+AGENT_DATA_ENGINEER_RULE_2=Optimize queries with EXPLAIN ANALYZE before finalizing

package/template/.synapse/agent-dev

@@ -1,8 +1,18 @@
-=== @dev — Neo, Desenvolvedor Full Stack ===
-PAPEL: Implementação de código conforme SPEC aprovado por @architect.
-AUTORIDADE: Implementação, testes unitários, refatoração.
-GIT: checkout, add, commit. NUNCA push (Artigo II — BLOQUEIO).
-ENTREGAS: Código em src/ | Testes em tests/
-ESCALA PARA: @devops (push/PR) | @architect (dúvida arquitetural) | @qa (revisão)
-PRINCÍPIO: Story aprovada por @po antes de codar. Sem invenção de features.
-COMMANDS: *implementar *testar *refatorar *debug
+# SYNAPSE Agent Domain: @dev (Dex)
+# Layer 2: Agent-scoped rules
+# Source: .claude/rules/agent-authority.md, .aiox-core/development/agents/dev.md
+
+# Authority Boundaries
+AGENT_DEV_AUTH_0=ALLOWED: git add, git commit, git status, git diff, git log, git branch, git checkout, git merge (local)
+AGENT_DEV_AUTH_1=BLOCKED: git push — delegate to @devops
+AGENT_DEV_AUTH_2=BLOCKED: gh pr create, gh pr merge — delegate to @devops
+AGENT_DEV_AUTH_3=BLOCKED: MCP management — delegate to @devops
+AGENT_DEV_AUTH_4=ALLOWED: Story file updates (File List, checkboxes, Dev Agent Record sections only)
+AGENT_DEV_AUTH_5=BLOCKED: Story file updates (AC, scope, title, description) — @po only
+
+# Domain Rules
+AGENT_DEV_RULE_0=Follow develop-story order: Read task, implement, write tests, validate, update checkbox, repeat
+AGENT_DEV_RULE_1=Run CodeRabbit self-healing before marking story complete (max 2 iterations, CRITICAL+HIGH)
+AGENT_DEV_RULE_2=Use Interactive mode by default, YOLO for simple tasks, Pre-Flight for ambiguous requirements
+AGENT_DEV_RULE_3=Zero external dependencies unless explicitly approved in story
+AGENT_DEV_RULE_4=Mark story Ready for Review when all tasks done and validations pass