long-running-orchestrator

Autonomous AI workflows require separating initialization from execution and persisting task state. This skill provides a structured prompt blueprint for deterministic task orchestration, using checkpoints and self-correcting loops to prevent deadlocks and ensure stable operation across days.

State recovery after an interruption is handled through state persistence and checkpoints. This skill preserves progress in a task workspace containing todo.md, progress.log, and config.json, enabling automated recovery and resume for multi-step workflows.

To set up a long-running task orchestrator, tell Claude Code your objective. It generates a task workspace with todo.md, progress.log, and config.json, then scaffolds a worker workflow to process tasks until completion with minimal human supervision.

No external tool dependencies are required for long-running agent orchestration. This skill operates entirely within the Claude Code Agent Skills framework, providing a structured approach to iterative data collection and multi-step workflows without external packages.

Error handling in days-long autonomous tasks uses self-correcting loops to avoid deadlocks and enable automated recovery. The orchestrator separates initialization from execution, ensuring error states are captured and workflow execution resumes persistently.

State persistence for multi-step AI workflows is needed when tasks require continuous monitoring or autonomous operation with minimal supervision. It supports resume after interruptions by preserving progress in structured files like progress.log and todo.md.