Core agent capabilities including governance, graduation, and orchestration.
- Agent Governance - Maturity levels, permissions, and lifecycle management
- Agent Graduation Guide - 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 - Real-time supervision system
- Agent Guidance Implementation - Real-time agent operation tracking
- Governance Learning Integration - Learning from governance events
- Unstructured Complex Tasks - Intent classification and dynamic routing
- Fleet Admiral - Multi-agent orchestration for complex tasks
| 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 |
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:
- Each capability use is tracked via
CapabilityGraduationService.record_usage() - Success counts accumulate in
agent.properties["capability_stats"] - Automatic promotion when thresholds are reached
- Can be reset to STUDENT if issues occur
See: Capability Graduation Details
- 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
- Testing Index - Agent testing documentation
- Development Guide - Building agents
- API Documentation - Agent API endpoints
Last Updated: April 5, 2026