orchestrator
Use when managing agent state transitions (START/INIT/IMPLEMENT/TEST/COMPLETE), triggering context compression at 80% capacity, or handling session lifecycle. Load at session start, on state change, or when context exceeds threshold. Core skill for single-orchestrator architecture.
When & Why to Use This Skill
The Orchestrator skill provides a robust state-machine framework designed for managing complex AI agent workflows within a single-orchestrator architecture. It automates the lifecycle of agent tasks—ranging from initialization to final implementation—while optimizing performance through intelligent context compression at 80% capacity. By enforcing strict state transitions and managing session persistence, this skill ensures that AI agents remain reliable, efficient, and context-aware during long-running or multi-stage operations.
Use Cases
- Software Development Lifecycle: Managing an agent's progression through INIT, IMPLEMENT, and TEST phases to ensure code is verified before completion.
- Context Limit Management: Automatically triggering context compression and summarization when token usage reaches the 80% threshold, preventing session crashes or performance degradation.
- Session Recovery and Resumption: Utilizing state-machine logs to resume interrupted tasks with a fresh context and summary, ensuring no progress is lost in long-running workflows.
- Workflow Validation: Enforcing logical state transitions (e.g., preventing an agent from moving to 'COMPLETE' without passing 'TEST') to maintain high-quality output and operational guardrails.
| name | orchestrator |
|---|---|
| description | "Use when managing agent state transitions (START/INIT/IMPLEMENT/TEST/COMPLETE), triggering context compression at 80% capacity, or handling session lifecycle. Load at session start, on state change, or when context exceeds threshold. Core skill for single-orchestrator architecture." |
| keywords | state-machine, compression, session, transitions, context-management |
Orchestrator
State machine management for single-orchestrator architecture.
Instructions
- Check current state:
scripts/check-state.sh - Validate transition allowed:
scripts/validate-transition.sh FROM TO - Load skill for current state
- Execute until exit conditions (code verified, not judged)
- Compress when
scripts/check-context.shreturns threshold exceeded
State → Skill Mapping
| State | Load Skill |
|---|---|
| INIT | initialization/ |
| IMPLEMENT | implementation/ |
| TEST | testing/ |
| COMPLETE | context-graph/ |
References
| File | Load When |
|---|---|
| references/state-machine.md | Designing state transitions |
| references/compression.md | Context exceeds 50% (progressive checkpoints) |
| references/session-management.md | Session start/recovery |
| references/session-resumption.md | Resuming with fresh context + summary |
Scripts
| Script | Purpose | Output |
|---|---|---|
| scripts/check-state.sh | Get current state | JSON state |
| scripts/validate-transition.sh | Check if transition valid | exit 0/1 |
| scripts/check-context.sh | Check context usage | threshold level |
| scripts/enter-state.sh | Transition to new state | Updated state.json |