script-driven-skill

Multi-phase AI workflows are orchestrated using a Python-based state machine that coordinates subagents via self-contained phase files, ensuring deterministic and auditable execution. The orchestrator outputs a progress.json file to track state and enable resumable runs.

To prevent skipping steps in an AI workflow, the orchestrator uses progress tracking and gating mechanisms. It enforces phase-by-phase execution by evaluating a progress.json file, ensuring all required phases complete sequentially before advancing.

Yes, you can resume complex AI tasks because the orchestrator supports resumable runs. It records execution state in progress.json and uses phase artifacts, allowing the state machine to pick up accurately from the last completed phase.

The best way to structure multi-step AI tasks is using self-contained phase markdown files under a phases/ directory. This approach separates phase logic from the orchestrator loop, allowing subagents to execute instructions deterministically.

Yes, you need Python to run this AI workflow state machine. The orchestrator relies on a Python script called next-step.py, which manages the control loop, spawns subagents, and updates the progress.json file.

You start the AI workflow orchestrator by creating a workspace directory under planning, then executing the next-step.py script with that workspace path. This initializes the state machine and generates progress tracking artifacts.