feat: father-bot — agente del sistema que crea otros agentes via Matrix

Primer agente privilegiado en agents/_specials/. Usa provider claude-code
con acceso completo al repositorio para ejecutar el pipeline de creacion
de agentes de forma autonoma (scaffold, build, register, verify, restart).

Archivos:
- agent.go: reglas puras (DM/mention → LLM)
- config.yaml: claude-code provider, bypassPermissions, E2EE, audit
- prompts/system.md: guia completa de creacion con decision tree,
  convencion de IDs, validaciones y seccion anti-injection

Config: claude-code con working_dir al repo, allowed_tools restringido,
model sonnet, timeout 10m, add_dirs con templates y referencias.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

agents/_specials/father-bot/config.yaml

@@ -0,0 +1,231 @@
+# ============================================
+# FATHER BOT — Agente privilegiado del sistema
+# ============================================
+# Crea otros agentes y robots via Matrix usando claude-code.
+# Ubicado en _specials/ por su rol de sistema. ACL admin-only.
+
+agent:
+  id: father-bot
+  name: "Father Bot"
+  version: "1.0.0"
+  enabled: true
+  description: "Agente del sistema que crea otros agentes y robots via Matrix. Acceso completo al repositorio."
+  tags: [system, privileged, creator]
+
+# ============================================
+# PERSONALIDAD Y COMPORTAMIENTO
+# ============================================
+personality:
+  tone: technical
+  verbosity: concise
+  language: es
+  languages_supported: [es, en]
+  emoji_style: minimal
+  prefix: ""
+  error_style: detailed
+
+  role: "Arquitecto de agentes — crea, configura y despliega nuevos bots Matrix"
+  backstory: "Soy el agente padre del sistema. Conozco la arquitectura completa del proyecto y puedo crear nuevos agentes o robots bajo demanda."
+  expertise: [go, matrix, agent-architecture, devops, shell-scripting]
+  limitations: ["No modifico agentes existentes sin confirmacion explicita", "No elimino agentes"]
+
+  communication:
+    formality: semiformal
+    humor: none
+    personality: pragmatic
+    response_style: structured
+    quirks: []
+    avoid_topics: []
+    catchphrases: []
+
+  custom_directives:
+    - "Siempre confirma el tipo (agent/robot) y el nombre antes de crear"
+    - "Reporta cada paso del pipeline con resultado (exito/fallo)"
+    - "Si algo falla, muestra el error y sugiere recovery"
+
+  templates:
+    greeting: "Soy Father Bot. Puedo crear agentes y robots para este sistema. Describeme lo que necesitas."
+    unknown_command: "Comando desconocido. Usa !help o describeme que agente necesitas crear."
+    permission_denied: "Solo administradores pueden interactuar conmigo."
+    error: "Error en la operacion: {{.Error}}"
+    success: "{{.Summary}}"
+    busy: "Estoy creando un agente, espera a que termine..."
+
+  behavior:
+    proactive: false
+    ask_confirmation: true
+    show_reasoning: true
+    thread_replies: true
+    typing_indicator: true
+    acknowledge_receipt: true
+
+# ============================================
+# LLM — claude-code provider
+# ============================================
+llm:
+  primary:
+    provider: claude-code
+    model: ""
+    api_key_env: ""
+    base_url: ""
+    max_tokens: 16384
+    temperature: 0.3
+
+    claude_code:
+      binary: "claude"
+      timeout: 10m
+      disable_tools: false
+      allowed_tools: [Bash, Read, Edit, Write, Glob, Grep]
+      disallowed_tools: []
+      working_dir: "/home/ubuntu/CodeProyects/agents_and_robots"
+      permission_mode: "bypassPermissions"
+      model: "sonnet"
+      fallback_model: "haiku"
+      session_id: ""
+      add_dirs:
+        - ".claude/rules"
+        - "agents/_template"
+        - "agents/_template_robot"
+        - "agents/assistant-bot"
+        - "agents/asistente-2"
+        - "internal/config"
+        - "dev-scripts/agent"
+
+  fallback:
+    provider: ""
+    model: ""
+    api_key_env: ""
+
+  reasoning:
+    system_prompt_file: "prompts/system.md"
+    context_window: 16384
+    memory_messages: 30
+
+  tool_use:
+    enabled: false
+    max_iterations: 5
+    parallel_calls: false
+
+  rate_limit:
+    requests_per_minute: 20
+    tokens_per_minute: 200000
+    concurrent_requests: 2
+
+# ============================================
+# TOOLS — deshabilitadas (claude-code maneja todo)
+# ============================================
+tools:
+  ssh:
+    enabled: false
+  http:
+    enabled: false
+  scripts:
+    enabled: false
+  file_ops:
+    enabled: false
+  matrix_send:
+    allowed_rooms: []
+  mcp:
+    enabled: false
+  memory:
+    enabled: false
+  knowledge:
+    enabled: false
+  shared_knowledge:
+    enabled: false
+  skills:
+    allowed_interpreters: []
+
+# ============================================
+# SKILLS — deshabilitadas
+# ============================================
+skills:
+  enabled: false
+
+# ============================================
+# MEMORIA — habilitada para contexto de conversacion
+# ============================================
+memory:
+  enabled: true
+  window_size: 30
+  db_path: ""
+
+# ============================================
+# MATRIX
+# ============================================
+matrix:
+  homeserver: "${MATRIX_HOMESERVER}"
+  user_id: "@father-bot:${MATRIX_SERVER_NAME}"
+  access_token_env: MATRIX_TOKEN_FATHER_BOT
+  device_id: "ZMLLZOHAXM"
+
+  encryption:
+    enabled: true
+    store_path: "./agents/_specials/father-bot/data/crypto/"
+    pickle_key_env: PICKLE_KEY_FATHER_BOT
+    trust_mode: tofu
+    recovery_key_env: SSSS_RECOVERY_KEY_FATHER_BOT
+
+  rooms:
+    listen: []
+    respond: []
+    admin: []
+
+  filters:
+    command_prefix: "!"
+    mention_respond: true
+    dm_respond: true
+    ignore_bots: true
+    ignore_users: []
+    unauthorized_response: explicit
+    min_power_level: 0
+
+  threads:
+    enabled: true
+    auto_thread: false
+
+# ============================================
+# SSH INVENTORY — disponible para el subprocess claude-code
+# ============================================
+ssh:
+  defaults:
+    user: "root"
+    port: 22
+    key_file_env: SSH_KEY_FILE
+    known_hosts: "~/.ssh/known_hosts"
+    keepalive_interval: 30s
+    timeout: 60s
+  targets: {}
+
+# ============================================
+# SEGURIDAD
+# ============================================
+security:
+  audit:
+    enabled: true
+    log_file: ""
+    log_to_room: ""
+    include: [command, llm_request, llm_response]
+
+  secrets:
+    provider: env
+
+  sanitize:
+    enabled: true
+    mode: warn
+    min_severity: medium
+    disabled_patterns: []
+
+  tool_rate_limit:
+    enabled: false
+
+# ============================================
+# SCHEDULING
+# ============================================
+schedules: []
+
+# ============================================
+# STORAGE
+# ============================================
+storage:
+  base_path: ""