What problem does it solve?
Orchestrating many parallel AI agents and preserving progress across crashes is complex and error-prone; this Skill provides tested patterns and helpers to manage state, resume partial runs, and collect progressive results so long-running pipelines complete reliably.
Core Features & Use Cases
- State management & crash resume: Canonical state file convention, atomic writes, phase- and node-level status, and resume detection by git SHA.
- Agent dispatch & fan-out/fan-in: Helpers and reference patterns to start nodes before dispatch, spawn parallel agents, and update node-level outcomes without re-running completed work.
- Progressive collection & synthesis: Stream incremental results as agent groups finish, deduplicate across batches, and run downstream synthesis early.
- Telemetry, timing & notifications: Unified telemetry logging with rotation, per-phase timing reports, pipeline UUIDs, and optional TTS notifications for pipeline events.
- Use cases: Implement robust commands like audit-all, apply-waves, and domain audit tasks that require parallel agents, fault tolerance, and end-to-end timing and logging.
Quick Start
Source the orchestrator helpers, derive a state file for your command, and run orch_check_resume to initialize or resume the pipeline.
