# Agent System Documentation

Core agent capabilities including governance, graduation, and orchestration.

## Core Components

### Governance & Maturity ✨ Enhanced 2026
- **Agent Governance** - Maturity levels, permissions, and lifecycle management
  - **Three-Layer Governance** - OPERATIONAL, TACTICAL, STRATEGIC decision layers
  - **Policy Engine** - Context-aware policy evaluation with priority resolution
  - **Governance-as-a-Service** - Multi-tenant governance API
- **[Agent Graduation Guide](graduation.md)** - How agents learn and get promoted
- **Capability Graduation** ⚡ **Individual capability progression (5/20/50 rule)**
- **Student Agent Training** - Training workflow for student agents
- **[Supervision & Learning](../archive/implementation/SUPERVISION_LEARNING_INTEGRATION_COMPLETE.md)** - Real-time supervision system

### Guidance & Orchestration
- **[Agent Guidance Implementation](guidance-system.md)** - Real-time agent operation tracking
- **Governance Learning Integration** - Learning from governance events

### Multi-Agent Coordination ✨ Enhanced 2026
- **[Unstructured Complex Tasks](unstructured-tasks.md)** - Intent classification and dynamic routing
- **Fleet Admiral** - Multi-agent orchestration for complex tasks
  - **Conductor Agent** - 5 execution strategies (SEQUENTIAL, PARALLEL, HYBRID, ADAPTIVE, ROLLBACK_SAFE)
  - **Workflow State Machine** - Validated state transitions with rollback support
  - **Event Bus** - Event-driven workflow triggering with pub/sub
  - **Workflow Templates** - Pre-built enterprise patterns
- **Queen Agent** - Structured workflow automation for repeatable processes

## Key Concepts

### Agent Maturity Levels
| Level | Confidence | Automated Triggers | Capabilities |
|-------|-----------|-------------------|--------------|
| **STUDENT** | <0.5 | Blocked (read-only) | Charts, markdown presentation |
| **INTERN** | 0.5-0.7 | Proposal required | Streaming, form presentation |
| **SUPERVISED** | 0.7-0.9 | Supervised execution | Form submissions, state changes |
| **AUTONOMOUS** | >0.9 | Full execution | All capabilities |

### ⚡ Capability Graduation Logic

Individual capabilities (skills, tools, actions) graduate independently based on usage:

| Capability Level | Promotion Requirement | Added Permissions |
|-----------------|---------------------|-------------------|
| **STUDENT → INTERN** | 5 successful uses | Basic tool access |
| **INTERN → SUPERVISED** | 20 successful uses | Advanced integrations |
| **SUPERVISED → AUTONOMOUS** | 50 successful uses | Full autonomy, critical actions |

**How It Works**:
1. Each capability use is tracked via `CapabilityGraduationService.record_usage()`
2. Success counts accumulate in `agent.properties["capability_stats"]`
3. Automatic promotion when thresholds are reached
4. Can be reset to STUDENT if issues occur

**See**: Capability Graduation Details

### Intent Classification
- **CHAT** - Simple queries → LLMService
- **WORKFLOW** - Structured tasks → **Queen Agent** (Queen Hive) → **Workflow Automation**
- **TASK** - Unstructured complex long-horizon tasks → **Fleet Admiral** → **Dynamic Fleet Coordination**

**Key Distinction**:
- **Queen Agent** handles **structured complex tasks** with predefined steps (repeatable business processes)
- **Fleet Admiral** handles **unstructured complex long-horizon tasks** requiring adaptive planning and dynamic agent recruitment

## Quick Links

- **Testing Index** - Agent testing documentation
- **Development Guide** - Building agents
- **API Documentation** - Agent API endpoints

---

*Last Updated: April 5, 2026*