sinapse-ai 7.4.4 → 7.4.5

package/.claude/rules/nsn-mode.md

@@ -0,0 +1,53 @@
+# NSN Mode — Never Say Never (Global Rule)
+
+> **NON-NEGOTIABLE**: This rule applies to ALL agents in the SINAPSE ecosystem.
+> No agent may say "I can't" without first exhausting alternatives.
+
+## Protocol
+
+When ANY agent encounters a barrier (technical limitation, permission denied, tool failure, unknown approach):
+
+1. **Classify** the barrier:
+   - `permission` — System permission missing
+   - `technical` — Tool or API limitation
+   - `knowledge` — Don't know how to do it
+   - `external` — External service blocks
+
+2. **Search** for solutions:
+   - Use WebSearch/WebFetch to find workarounds
+   - Query patterns: `"{tool} {error} workaround"`, `"{action} alternative approach"`
+   - Check documentation for the tool/service
+
+3. **Try 3+ alternatives** before escalating:
+   - If Tool A fails → try Tool B
+   - If API blocks → try browser automation
+   - If direct approach fails → try indirect approach
+   - If programmatic fails → try CLI tool
+
+4. **Configure and test** the solution found:
+   - Install dependencies if needed
+   - Adjust permissions
+   - Test in isolated context
+   - Validate result
+
+5. **Repeat** — If still blocked, return to step 2 with new context (max 5 cycles)
+
+6. **Escalate** — After 5 cycles, report to user with detailed log of what was tried
+
+## Anti-Patterns (FORBIDDEN)
+
+- Saying "I can't do this" without trying alternatives
+- Stopping at the first error
+- Assuming something is impossible without researching
+- Asking the user to do something the agent can automate
+- Giving up without documenting what was tried
+
+## Exceptions
+
+- Destructive actions affecting user data → ask confirmation first
+- Actions involving payment/purchase → ask confirmation first
+- Actions violating laws or terms of service → inform and stop
+
+## Applies To
+
+Every agent: @developer, @quality-gate, @architect, @sinapse-orqx, all squad orqx agents, and all specialist agents within squads. No exceptions.

package/.sinapse-ai/install-manifest.yaml

@@ -7,8 +7,8 @@
 # - SHA256 hashes for change detection
 # - File types for categorization
 #
-version: 7.4.4
-generated_at: "2026-03-27T17:44:30.530Z"
+version: 7.4.5
+generated_at: "2026-03-27T18:12:46.613Z"
 generator: scripts/generate-install-manifest.js
 file_count: 1101
 files:

package/README.en.md

@@ -196,55 +196,71 @@ Use `@squad-creator` or see the [Squads Guide](docs/guides/squads-guide.md).
 
 ## Claude Code vs Codex CLI — Full Comparison
 
-SINAPSE works on both IDEs. Claude Code delivers the full experience; Codex CLI supports the core with some differences.
+SINAPSE works on both IDEs. Both support most features with different approaches.
 
 ### Features
 
-| Feature | Claude Code | Codex CLI |
-|---------|:-----------:|:---------:|
-| **19 squads with 175 agents** | YES | YES |
-| **Knowledge bases per squad** | YES | YES |
-| **Tasks and workflows** | YES | YES |
-| **Story-Driven Development** | YES | YES |
-| **Agent commands (`*help`, `*task`)** | YES | YES |
-| **Squad awareness auto-routing** | YES | YES |
-| **Chrome Brain (browser automation)** | YES | NO |
-| **Hooks (PreToolUse / PostToolUse)** | YES | NO |
-| **MCP servers (Chrome DevTools, etc.)** | YES | NO |
-| **Agent handoff protocol** | YES | NO |
-| **SYNAPSE context engine** | YES | NO |
-| **Multi-model (Claude + others)** | NO | YES |
+| Feature | Claude Code | Codex CLI | Notes |
+|---------|:-----------:|:---------:|-------|
+| **19 squads with 175 agents** | YES | YES | Identical on both |
+| **Knowledge bases per squad** | YES | YES | Identical on both |
+| **Tasks and workflows** | YES | YES | Identical on both |
+| **Story-Driven Development** | YES | YES | Identical on both |
+| **Agent commands (`*help`, `*task`)** | YES | YES | Identical on both |
+| **Squad awareness auto-routing** | YES | YES | Identical on both |
+| **NSN Mode (Never Say Never)** | YES | YES | Global rule on both |
+| **Chrome Brain (browser automation)** | YES | YES | Claude: hooks auto-launch. Codex: via MCP or chrome-cdp-skill |
+| **MCP servers** | YES | YES | Claude: `~/.claude.json`. Codex: `config.toml` or `codex mcp add` |
+| **Hooks (PreToolUse / PostToolUse)** | YES | PARTIAL | Codex: Bash tool only, disabled on Windows |
+| **Agent handoff protocol** | YES | YES | Via instructions + skills on both |
+| **SYNAPSE context engine** | YES | PARTIAL | Codex: via instructions.md + skills (no dynamic rules) |
+| **Native skills system** | NO | YES | Codex has `$skill-name` with progressive disclosure |
+| **Multi-model (Claude + others)** | NO | YES | Codex supports any OpenAI model |
+| **Parallel subagents** | YES | YES | Codex: configurable max 6 threads |
+| **Non-interactive CI/CD** | NO | YES | Codex: `codex exec` with `--json` |
 
 ### How to Activate Agents
 
 | Action | Claude Code | Codex CLI |
 |--------|-------------|-----------|
-| Main orchestrator | `/SINAPSE:agents:sinapse-orqx` | `/skills` > `sinapse-orqx` |
-| Brand squad | `@brand-orqx` | `sinapse-brand` |
-| Copy squad | `@copy-orqx` | `sinapse-copy` |
-| Developer | `@developer` | `sinapse-dev` |
+| Main orchestrator | `/SINAPSE:agents:sinapse-orqx` | `@sinapse-orqx` or `$sinapse-orqx` |
+| Brand squad | `@brand-orqx` | `@brand-orqx` or `$sinapse-brand` |
+| Copy squad | `@copy-orqx` | `@copy-orqx` or `$sinapse-copy` |
+| Developer | `@developer` | `@developer` or `$sinapse-dev` |
 | List agents | `@sinapse-orqx *help` | `/skills` |
+| Chrome Brain | Auto-activates by prompt | `codex mcp add chrome-devtools` |
 
 ### Where Each IDE Shines
 
-**Claude Code** — Maximum potential:
-- Chrome Brain auto-activates when you request browser actions
-- Hooks automate pre/post-actions on each tool call
-- MCP servers connect to external tools (Chrome, Playwright, EXA)
-- Context engine injects rules automatically by domain
-- Handoff protocol optimizes agent switching
+**Claude Code** — Most integrated experience:
+- Chrome Brain auto-activates via hooks (zero manual config)
+- Context engine injects rules automatically by domain and file
+- Deny/allow rules protect files deterministically
+- Largest MCP server ecosystem (Figma, Supabase, Notion, etc.)
 
-**Codex CLI** — Solid core:
-- All squads and agents work perfectly
-- Identical knowledge bases and tasks
-- Supports multiple models (not just Claude)
-- Ideal for those who prefer GPT/other models with SINAPSE orchestration
+**Codex CLI** — Most flexible and extensible:
+- Native skills with `$skill-name` and progressive disclosure
+- Supports any OpenAI model (not just Claude)
+- `codex exec` for non-interactive CI/CD automation
+- Can run AS an MCP server (other agents can invoke Codex)
+- Profiles for quick config switching
+- Subagents with max 6 parallel threads
+
+### Chrome Brain Setup on Codex CLI
+
+```bash
+# Add Chrome DevTools MCP to Codex
+codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latest
+
+# OR install the chrome-cdp skill (direct connection, no Puppeteer)
+# pi install git:github.com/pasky/chrome-cdp-skill@v1.0.1
+```
 
 ### Recommendation
 
-> For **maximum potential**, use **Claude Code** — all advanced features (Chrome Brain, hooks, MCP, SYNAPSE engine) are only available on this IDE.
+> **Claude Code** for the most integrated and automated experience — Chrome Brain auto-activates, hooks run automatically, context engine injects rules.
 >
-> For **model flexibility**, use **Codex CLI** — works with any LLM and delivers the complete core of squads and agents.
+> **Codex CLI** for model flexibility, native skills, and CI/CD automation — with minimal manual setup for Chrome Brain.
 
 ---
 

package/README.md

@@ -221,55 +221,71 @@ Use `@squad-creator` ou veja o [Guia de Squads](docs/guides/squads-guide.md).
 
 ## Claude Code vs Codex CLI — Comparativo Completo
 
-O SINAPSE funciona em ambas as IDEs. Claude Code oferece a experiencia completa; Codex CLI suporta o core com algumas diferencas.
+O SINAPSE funciona em ambas as IDEs. Ambas suportam a maioria das funcionalidades com abordagens diferentes.
 
 ### Funcionalidades
 
-| Funcionalidade | Claude Code | Codex CLI |
-|---------------|:-----------:|:---------:|
-| **19 squads com 175 agentes** | SIM | SIM |
-| **Knowledge bases por squad** | SIM | SIM |
-| **Tasks e workflows** | SIM | SIM |
-| **Story-Driven Development** | SIM | SIM |
-| **Agent commands (`*help`, `*task`)** | SIM | SIM |
-| **Squad awareness auto-routing** | SIM | SIM |
-| **Chrome Brain (browser automation)** | SIM | NAO |
-| **Hooks (PreToolUse / PostToolUse)** | SIM | NAO |
-| **MCP servers (Chrome DevTools, etc.)** | SIM | NAO |
-| **Agent handoff protocol** | SIM | NAO |
-| **SYNAPSE context engine** | SIM | NAO |
-| **Multi-model (Claude + outros)** | NAO | SIM |
+| Funcionalidade | Claude Code | Codex CLI | Notas |
+|---------------|:-----------:|:---------:|-------|
+| **19 squads com 175 agentes** | SIM | SIM | Identico em ambas |
+| **Knowledge bases por squad** | SIM | SIM | Identico em ambas |
+| **Tasks e workflows** | SIM | SIM | Identico em ambas |
+| **Story-Driven Development** | SIM | SIM | Identico em ambas |
+| **Agent commands (`*help`, `*task`)** | SIM | SIM | Identico em ambas |
+| **Squad awareness auto-routing** | SIM | SIM | Identico em ambas |
+| **NSN Mode (Never Say Never)** | SIM | SIM | Regra global em ambas |
+| **Chrome Brain (browser automation)** | SIM | SIM | Claude: hooks auto-launch. Codex: via MCP ou chrome-cdp-skill |
+| **MCP servers** | SIM | SIM | Claude: `~/.claude.json`. Codex: `config.toml` ou `codex mcp add` |
+| **Hooks (PreToolUse / PostToolUse)** | SIM | PARCIAL | Codex: so Bash tool, desabilitado no Windows |
+| **Agent handoff protocol** | SIM | SIM | Via instructions + skills em ambas |
+| **SYNAPSE context engine** | SIM | PARCIAL | Codex: via instructions.md + skills (sem regras dinamicas) |
+| **Skills system nativo** | NAO | SIM | Codex tem `$skill-name` com progressive disclosure |
+| **Multi-model (Claude + outros)** | NAO | SIM | Codex suporta qualquer modelo OpenAI |
+| **Subagents paralelos** | SIM | SIM | Codex: max 6 threads configuraveis |
+| **CI/CD non-interactive** | NAO | SIM | Codex: `codex exec` com `--json` |
 
 ### Como Ativar Agentes
 
 | Acao | Claude Code | Codex CLI |
 |------|-------------|-----------|
-| Orquestrador principal | `/SINAPSE:agents:sinapse-orqx` | `/skills` > `sinapse-orqx` |
-| Squad de brand | `@brand-orqx` | `sinapse-brand` |
-| Squad de copy | `@copy-orqx` | `sinapse-copy` |
-| Developer | `@developer` | `sinapse-dev` |
+| Orquestrador principal | `/SINAPSE:agents:sinapse-orqx` | `@sinapse-orqx` ou `$sinapse-orqx` |
+| Squad de brand | `@brand-orqx` | `@brand-orqx` ou `$sinapse-brand` |
+| Squad de copy | `@copy-orqx` | `@copy-orqx` ou `$sinapse-copy` |
+| Developer | `@developer` | `@developer` ou `$sinapse-dev` |
 | Listar agentes | `@sinapse-orqx *help` | `/skills` |
+| Chrome Brain | Auto-ativa por prompt | `codex mcp add chrome-devtools` |
 
 ### Onde Cada IDE Brilha
 
-**Claude Code** — Experiencia maxima:
-- Chrome Brain auto-ativa quando voce pede algo no browser
-- Hooks automatizam pre/pos-acoes em cada tool call
-- MCP servers conectam a ferramentas externas (Chrome, Playwright, EXA)
-- Context engine injeta regras automaticamente por dominio
-- Handoff protocol otimiza troca entre agentes
+**Claude Code** — Experiencia mais integrada:
+- Chrome Brain auto-ativa por hooks (zero configuracao manual)
+- Context engine injeta regras automaticamente por dominio e arquivo
+- Deny/allow rules protegem arquivos deterministicamente
+- Maior ecossistema de MCP servers (Figma, Supabase, Notion, etc.)
 
-**Codex CLI** — Core solido:
-- Todos os squads e agentes funcionam perfeitamente
-- Knowledge bases e tasks identicas
-- Suporta multiplos modelos (nao apenas Claude)
-- Ideal para quem prefere GPT/outros modelos com a orquestracao SINAPSE
+**Codex CLI** — Mais flexivel e extensivel:
+- Skills nativas com `$skill-name` e progressive disclosure
+- Suporta qualquer modelo OpenAI (nao apenas Claude)
+- `codex exec` para automacao CI/CD sem interacao
+- Pode rodar como MCP server (outros agentes invocam Codex)
+- Profiles para alternar configuracoes rapidamente
+- Subagents com max 6 threads paralelos
+
+### Setup Chrome Brain no Codex CLI
+
+```bash
+# Adicionar Chrome DevTools MCP ao Codex
+codex mcp add chrome-devtools -- npx chrome-devtools-mcp@latest
+
+# OU instalar a skill chrome-cdp (conexao direta, sem Puppeteer)
+# pi install git:github.com/pasky/chrome-cdp-skill@v1.0.1
+```
 
 ### Recomendacao
 
-> Para **maximo potencial**, use **Claude Code** — todas as funcionalidades avancadas (Chrome Brain, hooks, MCP, SYNAPSE engine) so estao disponiveis nessa IDE.
+> **Claude Code** para quem quer a experiencia mais integrada e automatizada — Chrome Brain auto-ativa, hooks rodam sozinhos, context engine injeta regras.
 >
-> Para **flexibilidade de modelo**, use **Codex CLI** — funciona com qualquer LLM e entrega o core completo de squads e agentes.
+> **Codex CLI** para quem quer flexibilidade de modelo, skills nativas, e automacao CI/CD — com setup manual minimo para Chrome Brain.
 
 ---
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sinapse-ai",
-  "version": "7.4.4",
+  "version": "7.4.5",
   "description": "SINAPSE AI: Framework de orquestracao de IA — 18 squads, 174 agentes especializados",
   "bin": {
     "sinapse": "bin/sinapse.js",

package/sinapse/agents/sinapse-orqx.md

@@ -71,6 +71,32 @@ When the user provides ANY briefing, request, or initiative (regardless of compl
 
 **The user should NEVER have to request orchestration or handoffs. This is Imperator's core function and must happen automatically on every briefing.**
 
+## NON-NEGOTIABLE: NSN MODE (NEVER SAY NEVER)
+
+> **Every agent under Imperator's command operates in NSN Mode.**
+
+Before ANY agent says "I can't", "this is not possible", or "I don't know how":
+1. Search the web for workarounds (WebSearch/WebFetch)
+2. Try 3+ alternative approaches
+3. Configure, test, repeat (max 5 cycles)
+4. Only escalate after exhausting all alternatives with a detailed report
+
+**Imperator enforces NSN across all squads.** If a squad agent gives up too early, Imperator intervenes with alternative routing or cross-squad coordination.
+
+See `.claude/rules/nsn-mode.md` for the full protocol.
+
+## NON-NEGOTIABLE: DOCUMENTATION HANDOFFS
+
+> **Every orchestration plan MUST include documentation handoffs.**
+
+No project continues without proper documentation. For every execution:
+1. **Story file** must be created/updated with progress checkboxes
+2. **File List** must be maintained in the story
+3. **Architectural decisions** must be documented in the handoff artifact
+4. **README/docs** must be updated when user-facing changes are made
+
+Documentation is NOT optional. It is a handoff deliverable like code.
+
 ## COMPLETE AGENT DEFINITION
 
 ```yaml