ralph-loop

Autonomous iterative task execution uses a stateful loop that preserves context across iterations via a local state file, enabling Claude to progress tasks without starting over. It targets well-defined tasks with clear completion criteria like refactoring or code migrations.

Batch processing in a loop is handled by running Claude with a persistent local state file and a TODO.md file. This structured activation protocol maintains state between iterations, reducing manual check-ins until the batch task is verified complete.

State management is configured by creating a frontmatter-driven state file at .claude/ralph-loop.local.md. This file preserves context across iterations, allowing the autonomous loop to resume task progress reliably without manual state tracking.

The loop terminates safely using explicit completion criteria, specifically a completed TODO.md file or a defined completion promise. These signals ensure the autonomous iteration stops precisely when the task's completion criteria are met.

Code migrations and refactoring are ideal use cases for this autonomous loop because they are well-defined tasks with verifiable completion criteria. The persistent state file ensures progress is maintained across iterations without restarting the migration.

The autonomous loop provides safe manual interruption options alongside its explicit completion signals. You can monitor checkpoints and progress at each iteration, allowing you to manually stop the loop if the task deviates from expected behavior.