ralph-wiggum

Iterative development loops are managed by orchestrating an agent across multiple iterations using a persistent task file and state. Progress is checkpointed after each turn, ensuring work is verified frequently rather than completed in a single shot.

You can control pacing during refactoring by setting items-per-iteration and max-iterations options. These parameters guide effort distribution across turns, preventing the agent from attempting too much work in a single iteration.

Periodic reflection triggers structured self-reflection checkpoints to reduce unproductive grinding during complex tasks. It is needed when progress stalls, allowing the agent to reassess the plan, update evidence, and refocus on remaining checklist items.

To start an iterative loop for feature development, create a writable task markdown file containing Goals, a Checklist, and a Verification section. The loop persists its state in a local directory and runs until a completion marker is enforced.

No external dependencies or components are required to run paced iterative dev loops. You only need a writable task file to track goals and checklists, while loop state is persisted locally as JSON.

State across agent iterations is maintained using dedicated start and done control markers alongside a persistent state file. If these markers are missing or the state directory is not writable, the loop fails to checkpoint progress and loses continuity.