workflow:harness

Automate iterative development cycles by running Claude Code in an external bash loop that orchestrates task breakdown, tracks progress via acceptance criteria in PLAN.md, and generates detailed iteration logs for autonomous execution.

Spec-based progress tracking monitors task completion by evaluating defined acceptance criteria in a PLAN.md file, allowing an external bash loop to autonomously determine if iterative feature implementation and refactoring steps meet the required specifications.

Yes, you can use an external bash loop to manage multi-step feature implementation, enabling AI agents to autonomously work through complex tasks and refactoring by orchestrating execution and tracking progress against defined acceptance criteria.

Initialize a new feature folder using the harness init command, generate detailed specs, define tasks with acceptance criteria in PLAN.md, and let the external loop autonomously build and refine the feature until all criteria are met.

Yes, predefined acceptance criteria are required to monitor task completion and guide the external bash loop, ensuring the autonomous development workflow builds and refines features accurately until all specified conditions are met.

Autonomous AI agents rely entirely on the accuracy of the initial specs and acceptance criteria in PLAN.md; poorly defined criteria or ambiguous task breakdowns can cause the external bash loop to iterate inefficiently or fail to meet feature requirements.