bps-kit 1.3.1 → 1.4.0

package/.bps-kit.json

@@ -1,6 +1,6 @@
 {
   "mode": "normal",
   "vscode": false,
-  "version": "1.3.0",
-  "installedAt": "2026-03-11T00:38:59.884Z"
+  "version": "1.3.1",
+  "installedAt": "2026-03-11T00:51:23.317Z"
 }

package/CHANGELOG.md

@@ -5,6 +5,21 @@ Todas as mudanças notáveis ​​neste projeto serão documentadas neste arqui
 O formato é baseado no [Keep a Changelog](https://keepachangelog.com/pt-BR/1.0.0/), 
 e este projeto adere ao [Versionamento Semântico](https://semver.org/lang/pt-BR/).
 
+## [1.4.0] - 2026-03-11
+### Adicionado
+- **AGENTS.md**: Novo arquivo de regras dedicado ao routing de agentes, extraído do `GEMINI.md`. Contém a tabela Keyword→Agent completa com 22 agentes mapeados, regras de boundary enforcement e file type ownership. Distribuído automaticamente em `.agents/rules/AGENTS.md` (Antigravity) e `.github/AGENTS.md` (VS Code Copilot via `--vscode`).
+- **`convert_to_vscode.js`**: Suporte nativo ao `AGENTS.md` — o arquivo é convertido e copiado para `.github/AGENTS.md` com os mesmos path replacements do `GEMINI.md` (skills, vault, agents, scripts, frontmatter trigger).
+
+### Corrigido
+- **GEMINI.md**: 5 causas raiz que forçavam `Agent: orchestrator | Skill: none` como default foram eliminadas:
+  - `Auto-Selection Protocol` genérico substituído por referência à tabela concreta do `AGENTS.md`.
+  - Step 2 do checklist aponta explicitamente para `AGENTS.md` com instrução `NUNCA default orchestrator`.
+  - `PRE-FLIGHT OBRIGATÓRIO` reposicionado para pós-roteamento (passos 1-7) com revalidação anti-orchestrator.
+  - Step 5 agora carrega skills do frontmatter `skills:` do agente além do Intent Map.
+  - Gemini Mode `edit` não mais hardcoded para `orchestrator` — redireciona para Keyword→Agent table.
+  - Seção `Core Rule — Skills First` redundante removida (liberando espaço).
+- `GEMINI.md` reduzido de 279 → 182 linhas (−35%) sem perda de informação.
+
 ## [1.2.0] - 2026-03-09
 ### Adicionado
 - Novo template `ARCHITECTURE.md` distribuído junto ao kit, mapeando os 20 agentes com suas skills, skills organizadas por tier (basic/normal/extra), 12 workflows e scripts de validação disponíveis. Referenciado obrigatoriamente pelo `GEMINI.md` no startup da sessão.

package/README.md

@@ -43,7 +43,8 @@ npx bps-kit@latest --upgrade
 Se você adicionou a flag `--vscode`, a arquitetura compilada será organicamente moldada para as regras da nuvem do GitHub (`.github/`):
 - O manifesto base e as Workflows irão habitar o `.github/copilot-instructions.md`.
 - As Skills ativas receberão glob patterns específicos (`applyTo: "**/*"`) protegendo seu ciclo de vida.
-- As famosas 20 Personas se fundirão em um único registro local `.github/AGENTS.md`.
+- As 22 Personas são mapeadas individualmente em `.github/agents/*.agent.md` — com frontmatter Copilot nativo.
+- O manifesto de routing de agentes (Keyword→Agent) é exportado para `.github/AGENTS.md`.
 - A gigantesca Vault contendo +1000 skills inativas será armazenada inteligentemente em isolamento (`.copilot-vault/`), salvando permanentemente seu limite de Chat no VS Code de travar.
 
 ### 🪄 Autocalibragem de Base de Código (Workflow Analyzer)

package/bin/convert_to_vscode.js

@@ -26,6 +26,7 @@ async function convertToVsCode(destAgents, destBase) {
         content = content.replace(/\.?\/?\.agents\/skills\//g, '.copilot-skills/');
         content = content.replace(/\.?\/?\.agents\/vault\//g, '.copilot-vault/');
         content = content.replace(/\.?\/?\.agents\/rules\/GEMINI\.md/g, '.github/copilot-instructions.md');
+        content = content.replace(/\.?\/?\.agents\/rules\/AGENTS\.md/g, '.github/AGENTS.md');
         content = content.replace(/\.?\/?\.agents\/VAULT_INDEX\.md/g, '.github/VAULT_INDEX.md');
         content = content.replace(/\.?\/?\.agents\/ARCHITECTURE\.md/g, '.github/ARCHITECTURE.md');
         content = content.replace(/\.?\/?\.agents\/agents\//g, '.github/agents/');
@@ -40,6 +41,22 @@ async function convertToVsCode(destAgents, destBase) {
         await fs.writeFile(path.join(gitHubDir, 'copilot-instructions.md'), content);
     }
 
+    // 1.1 Converter AGENTS.md (routing rules) com os mesmos path replacements
+    const agentsMdPath = path.join(destAgents, 'rules', 'AGENTS.md');
+    if (await fs.pathExists(agentsMdPath)) {
+        let content = await fs.readFile(agentsMdPath, 'utf8');
+        content = content.replace(/\.?\/?\.agents\/skills\//g, '.copilot-skills/');
+        content = content.replace(/\.?\/?\.agents\/vault\//g, '.copilot-vault/');
+        content = content.replace(/\.?\/?\.agents\/rules\/GEMINI\.md/g, '.github/copilot-instructions.md');
+        content = content.replace(/\.?\/?\.agents\/rules\/AGENTS\.md/g, '.github/AGENTS.md');
+        content = content.replace(/\.?\/?\.agents\/VAULT_INDEX\.md/g, '.github/VAULT_INDEX.md');
+        content = content.replace(/\.?\/?\.agents\/ARCHITECTURE\.md/g, '.github/ARCHITECTURE.md');
+        content = content.replace(/\.?\/?\.agents\/agents\//g, '.github/agents/');
+        content = content.replace(/\.?\/?\.agents\/scripts\//g, '.github/scripts/');
+        content = content.replace(/trigger:\s*always_on/g, 'applyTo: "**/*"');
+        await fs.writeFile(path.join(gitHubDir, 'AGENTS.md'), content);
+    }
+
     // 2. Mover as skills ativas inteiras (em vez de achatar) para preservar scripts em python embutidos e sub documentações!
     // Importante: NÃO alocamos em .github/skills pois o Copilot engole todos os Markdowns de lá e causa sobrecarga de +60 referências!
     const skillsDest = path.join(destAgents, 'skills');

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bps-kit",
-  "version": "1.3.1",
+  "version": "1.4.0",
   "description": "BPS Kit - The Ultimate Antigravity Brain",
   "bin": {
     "bps-kit": "./bin/cli.js"

package/templates/agents-template/rules/AGENTS.md

@@ -0,0 +1,81 @@
+---
+trigger: always_on
+---
+
+# AGENTS.md — Agent Routing & Boundary Rules
+
+> Este arquivo define COMO rotear requests para o agente correto e suas regras de fronteira.
+
+---
+
+## 🔴 REGRA #1: orchestrator NÃO é default
+
+> **orchestrator = SOMENTE quando 2+ domínios distintos precisam trabalhar juntos.**
+> Se o request cabe em 1 agente → use esse agente diretamente.
+
+---
+
+## 🗺️ Keyword → Agent (primeira correspondência)
+
+| Keywords no request | Agent | Skills (frontmatter) |
+|---|---|---|
+| component, react, ui, ux, css, tailwind, frontend, hook, state | `frontend-specialist` | clean-code, frontend-design, react-patterns, tailwind-patterns |
+| landing page, site, website, hero, parallax, one-pager | `site-builder` | scroll-experience, enhance-prompt, frontend-design, design-md |
+| backend, server, api, endpoint, auth, express, fastapi, hono | `backend-specialist` | api-patterns, database-design, backend-dev-guidelines |
+| schema, SQL, migration, prisma, drizzle, database design | `database-architect` | database-design, prisma-expert |
+| bug, error, crash, not working, broken, fix, investigate | `debugger` | systematic-debugging |
+| test, testing, coverage, TDD, E2E, jest, vitest | `test-engineer` | testing-patterns, test-driven-development |
+| mobile, iOS, Android, React Native, Flutter, Expo | `mobile-developer` | mobile-design |
+| security, vulnerability, audit, OWASP | `security-auditor` | vulnerability-scanner |
+| pentest, red team, exploit | `penetration-tester` | vulnerability-scanner |
+| deploy, CI/CD, Docker, infrastructure, PM2 | `devops-engineer` | docker-expert |
+| performance, speed, Web Vitals, optimize, profiling | `performance-optimizer` | performance-profiling |
+| SEO, meta tags, ranking, sitemap | `seo-specialist` | seo-fundamentals |
+| plan, roadmap, task breakdown, milestones | `project-planner` | brainstorming, plan-writing |
+| requirements, user stories, backlog | `product-manager` | plan-writing, brainstorming |
+| MVP, strategy, product vision | `product-owner` | plan-writing, brainstorming |
+| game, unity, godot, phaser, multiplayer | `game-developer` | — |
+| n8n, webhook, automation, workflow automation | `automation-specialist` | n8n-mcp-tools-expert, n8n-workflow-patterns |
+| docs, README, documentation, manual | `documentation-writer` | — |
+| legacy code, refactor, tech debt | `code-archaeologist` | clean-code |
+| codebase analysis, map, discovery, explore | `explorer-agent` | — |
+| E2E automation, CI pipeline, QA pipeline | `qa-automation-engineer` | testing-patterns |
+| coordinate, orchestrate, 2+ domains simultâneos | `orchestrator` | plan-writing, behavioral-modes |
+
+### Protocolo de Uso
+
+1. **Match keywords** do request na tabela acima (primeira correspondência)
+2. **Ler** `.agents/agents/{agent}.md` → carregar skills do frontmatter `skills:`
+3. **Anunciar**: `🤖 **Applying knowledge of @[agent-name]...**`
+4. **Carregar skills**: Ler cada SKILL.md listada no frontmatter
+5. **Anunciar**: `📖 Using skill: [nome]` — para CADA skill carregada
+6. **Se Agent=orchestrator** → revalidar: o request REALMENTE precisa de 2+ agentes?
+
+---
+
+## 🔴 Agent Boundary Enforcement
+
+**Cada agente DEVE ficar no seu domínio. Cross-domain = VIOLAÇÃO.**
+
+| Agent | ✅ CAN Do | ❌ CANNOT Do |
+|-------|----------|-------------|
+| `frontend-specialist` | Components, UI, styles, hooks | Test files, API routes, DB |
+| `backend-specialist` | API, server logic, DB queries | UI components, styles |
+| `test-engineer` | Test files, mocks, coverage | Production code |
+| `mobile-developer` | RN/Flutter, mobile UX | Web components |
+| `database-architect` | Schema, migrations, queries | UI, API logic |
+| `security-auditor` | Audit, vulnerabilities | Feature code, UI |
+| `devops-engineer` | CI/CD, deployment, infra | Application code |
+| `debugger` | Bug fixes, root cause | New features |
+| `explorer-agent` | Codebase discovery | Write operations |
+
+### File Type Ownership
+
+| File Pattern | Owner Agent |
+|---|---|
+| `**/*.test.{ts,tsx,js}`, `**/__tests__/**` | `test-engineer` |
+| `**/components/**` | `frontend-specialist` |
+| `**/api/**`, `**/server/**` | `backend-specialist` |
+| `**/prisma/**`, `**/drizzle/**` | `database-architect` |
+
+> Se um agente precisa escrever fora do seu domínio → INVOCAR o agente correto para aquele arquivo.

package/templates/agents-template/rules/GEMINI.md

@@ -18,10 +18,10 @@ trigger: always_on
 ANTES de escrever QUALQUER código ou resposta:
 
 □ 1. CLASSIFICAR o request (QUESTION / SIMPLE / COMPLEX / DESIGN)
-□ 2. ROTEAR o agente correto (frontend-specialist / backend-specialist / etc)
+□ 2. ROTEAR via AGENTS.md — Keyword→Agent table (NUNCA default orchestrator)
 □ 3. ANUNCIAR: 🤖 **Applying knowledge of `@[agent-name]`...**
 □ 4. LER o arquivo .md do agente (ex: .agents/agents/frontend-specialist.md)
-□ 5. IDENTIFICAR skills relevantes no Intent → Skill Routing Map
+□ 5. CARREGAR skills do frontmatter `skills:` do agente (+ Intent Map p/ extras)
 □ 6. LER cada SKILL.md relevante
 □ 7. ANUNCIAR: 📖 Using skill: [nome] — para CADA skill usada
 □ 8. SOCRATIC GATE: Se build/feature → PERGUNTAR mínimo 3 questões estratégicas
@@ -35,7 +35,7 @@ ANTES de escrever QUALQUER código ou resposta:
 - **📖 SKILLS FIRST**: Se não anunciou `📖 Using skill:` → VOLTE e releia este bloco.
 - **⚠️ SEM SKILL**: Se nenhuma skill for relevante, diga: `⚠️ No skill used — responding from base knowledge.`
 - **🛑 NUNCA** pule o Socratic Gate em requests de build/feature/create.
-- **📣 PRE-FLIGHT OBRIGATÓRIO**: Antes de CADA resposta que não seja QUESTION, produza na primeira linha: `⚙️ Agent: [nome] | Skill: [nome ou none] | 🇧🇷 PT-BR` — isso é seu re-check forçado a cada turno.
+- **📣 PRE-FLIGHT OBRIGATÓRIO**: Após rotear agente+skill (passos 1-7), confirme: `⚙️ Agent: [nome] | Skill: [nome ou none] | 🇧🇷 PT-BR` — se Agent=orchestrator, revalide: é MESMO multi-domínio?
 - **🆘 /recall TRIGGER**: Se o usuário digitar `/recall` → PARE tudo, releia este arquivo imediatamente, e responda: `✅ Re-ancorado. Agent=[X] | Skill=[X] | PT-BR=SIM` antes de continuar.
 
 ---
@@ -69,9 +69,6 @@ Agent activated → Check frontmatter "skills:" → Read SKILL.md (INDEX) → Re
 - **Active skills**: in `.agents/skills/` — see `ARCHITECTURE.md` for full list
 - **Vault skills** (~1200+): in `.agents/vault/` — discoverable via index
 
-### Core Rule — Skills First
-After invoking any skill, explicitly say: '📖 Using skill: [skill-name]' before proceeding. If no skill was used, say: '⚠️ No skill used — responding from base knowledge.'
-
 ### Routing Flow
 1. Check if an **active skill** covers the request → use it directly
 2. If not → open `.agents/VAULT_INDEX.md` to find a vault skill
@@ -121,26 +118,10 @@ After invoking any skill, explicitly say: '📖 Using skill: [skill-name]' befor
 
 ## 🤖 INTELLIGENT AGENT ROUTING (STEP 2 - AUTO)
 
-### Auto-Selection Protocol
-1. **Analyze (Silent)**: Detect domains from user request.
-2. **Select Agent(s)**: Choose the most appropriate specialist(s).
-3. **Inform User**: State which expertise is being applied.
-4. **Apply**: Generate response using the selected agent's persona and rules.
-
-### Response Format (MANDATORY)
-```markdown
-🤖 **Applying knowledge of `@[agent-name]`...**
-[Continue with specialized response]
-```
-
-### Agent Routing Checklist (Before ANY code/design)
+> 🔴 **Leia `.agents/rules/AGENTS.md` para a tabela Keyword→Agent completa.**
+> Regra absoluta: orchestrator = SOMENTE multi-domínio (2+ agentes). NUNCA como default.
 
-| Step | Check | If Unchecked |
-|------|-------|--------------|
-| 1 | Identified correct agent? | → STOP. Analyze domain first. |
-| 2 | Read agent's `.md` file? | → STOP. Open `.agents/agents/{agent}.md` |
-| 3 | Announced agent? | → STOP. Add announcement. |
-| 4 | Loaded required skills? | → STOP. Check `skills:` field. |
+**Protocolo:** Match keywords → Ler agent `.md` → Carregar skills do frontmatter → Anunciar ambos.
 
 ---
 
@@ -230,7 +211,7 @@ Before coding, answer: (1) Goal of agent/skill? (2) Principles to apply? (3) How
 |---|---|---|
 | **plan** | `project-planner` | 4-phase. NO CODE before Phase 4. |
 | **ask** | - | Focus on understanding. |
-| **edit** | `orchestrator` | Execute. Check {task-slug}.md first. |
+| **edit** | Keyword→Agent table | Route especialista. orchestrator só se 2+ domains. |
 
 ---