Agents System | AIOS Portal

📊 Visão Geral dos Agentes

O AIOS define 12 agentes especializados, cada um com persona, comandos, e responsabilidades específicas. Os agentes seguem o padrão Unified Activation Pipeline para inicialização.

Agent	Name	Archetype	Role	LOC
👑 aios-master	Orion	Orchestrator	Master + Framework Dev	17.8k
💻 dev	Dex	Builder	Full Stack Developer	22.9k
🏛️ architect	Aria	Visionary	System Architect	19.0k
✅ qa	Quinn	Guardian	Test Architect	17.4k
📋 pm	Morgan	Strategist	Product Manager	15.1k
📝 po	—	—	Product Owner	12.8k
🔄 sm	River	—	Scrum Master	11.1k
🔧 devops	Gage	—	DevOps Engineer	20.1k
📊 analyst	—	—	Business Analyst	10.2k
🗄️ data-engineer	—	—	Data Engineer	20.3k
🎨 ux-design-expert	—	—	UX Designer	18.4k
🎭 squad-creator	—	—	Squad Generator	12.1k

👑 AIOS Master (Orion)

Universal executor of all Synkra AIOS capabilities - creates framework components, orchestrates workflows, and executes any task directly.

Comandos Principais

*help *kb *status *create *modify *workflow *plan *validate-workflow *run-workflow *ids check *ids impact *ids register *ids health

IDS Pre-Action Hooks

pre_create: Query registry antes de criar (REUSE/ADAPT/CREATE)
pre_modify: Impact analysis antes de modificar
post_create: Auto-register após criação

Security

Check permissions antes de component creation
Require confirmation para manifest modifications
Log all operations com user identification
No eval() ou dynamic code execution em templates
Validate YAML syntax antes de salvar

🎯 Core Development Team

Expert Senior Software Engineer & Implementation Specialist. Implements stories with precision, testing, and minimal context overhead.

*develop *run-tests *build *gotcha *worktree-create *rollback

Git Restrictions

✅ git add, commit, status, diff, log, branch, checkout, merge
❌ git push, gh pr create (delegate to @github-devops)

Holistic System Architect & Full-Stack Technical Leader. Bridges frontend, backend, infrastructure, and everything in between.

*create-full-stack-architecture *create-backend-architecture *analyze-project-structure *create-plan

Responsibility Boundaries

✅ System architecture, tech stack, API design, security
→ @data-engineer: Schema design, query optimization
→ @github-devops: Git push, PR creation

Test Architect with Quality Advisory Authority. Provides thorough quality assessment and actionable recommendations.

*review *gate *code-review *security-check *create-fix-request

Gate Decisions

PASS - Approved
CONCERNS - Issues to address
FAIL - Blocking issues
WAIVED - Accepted risk

Investigative Product Strategist & Market-Savvy PM. Creates PRDs and orchestrates development via "Projeto Bob".

*create-prd *create-epic *execute-epic *research

Orchestration Constraints

NEVER emulate other agents dentro do context window
Spawn agents em SEPARATE terminals via TerminalSpawner
User profile "bob" = orchestration mode

🚀 Unified Activation Pipeline

Processo de Ativação

// UnifiedActivationPipeline.activate(agentId)

1. PARALLEL LOAD:
   - Config (core-config.yaml)
   - Session state
   - Project status
   - Git config
   - Permissions

2. SEQUENTIAL:
   - Detect session type
   - Detect workflow state

3. BUILD GREETING:
   - Filter commands by visibility (full/quick/key)
   - Suggest workflow next steps
   - Format adaptive greeting

4. DISPLAY & HALT:
   - Show greeting
   - Wait for user input

Greeting Levels

minimal: 💻 dev Agent ready
named: 💻 Dex (Builder) ready. Let's build something great!
archetypal: 💻 Dex the Builder ready to innovate!

Critical Rules

STAY IN CHARACTER! - Mantém persona até *exit
Não carregar external agent files
Carregar dependencies apenas quando user requests execution
agent.customization sempre tem precedência
Tasks com elicit=true REQUEREM user interaction

🐰 CodeRabbit Integration

Self-Healing Workflow (Story 6.3.3)

iteration = 0
max_iterations = 2

WHILE iteration < max_iterations:
  1. Run CodeRabbit: wsl bash -c '~/.local/bin/coderabbit --prompt-only -t uncommitted'
  2. Parse output for CRITICAL issues

  IF no CRITICAL issues:
    - Log: "✅ CodeRabbit passed"
    - BREAK (ready for review)

  IF CRITICAL issues found:
    - Attempt auto-fix
    - iteration++
    - CONTINUE

IF CRITICAL issues remain after 2 iterations:
  - HALT and report to user
  - DO NOT mark story complete

Severity Handling

CRITICAL: Auto-fix immediately
HIGH: Document in story Dev Notes
MEDIUM/LOW: Ignore

🤝 Agent Collaboration

Workflow típico

@pm (Morgan)
  └─ Creates PRD → Creates Epic
       └─ @sm (River) creates Stories
            └─ @architect (Aria) designs Architecture
                 └─ @dev (Dex) implements
                      └─ @qa (Quinn) reviews
                           └─ @github-devops (Gage) pushes

Delegation Rules

Task	Owner	Delegates To
Epic creation	@pm	@sm for stories
Schema design	@architect	@data-engineer
Git push/PR	@dev	@github-devops
Market research	@pm	@analyst
Test suite creation	@aios-master	@qa