Running Claude Code in a loop to mirror human development practices
5 days ago
- #AI
- #Continuous Integration
- #GitHub
- Continuous Claude is a CLI tool designed to run Claude Code in a loop, maintaining persistent context across iterations.
- It was created to handle large-scale unit test coverage improvements, aiming for 80%+ coverage in weeks.
- Unlike one-shot AI coding tools, Continuous Claude allows for self-criticism and iterative improvement.
- The tool integrates with GitHub workflows, creating branches, pull requests, and monitoring CI checks.
- A shared markdown file serves as external memory, ensuring context continuity between iterations.
- The system can self-improve by tracking progress and adjusting strategies based on previous failures.
- GitHub Next's Continuous AI project explores running specialized agents simultaneously for different tasks.
- Continuous Claude can handle large refactoring tasks, dependency updates, and breaking changes.
- The approach is wasteful but effective, relying on the general direction of progress rather than individual successes.
- Humans remain in the loop through familiar mechanisms like PR reviews.