ck:autoresearch

Automated iterative code optimization executes a sequence of atomic code changes and measures a defined metric per iteration. It uses a git-tracked history to enable rollback and auto-keeps or discards changes to drive measurable improvements.

Git-backed experimentation prevents regressions by recording each atomic change as a commit and using an optional Guard step. This pairs with a verify command to ensure genuine progress and enable rollback if a metric degrades.

You can improve mechanical metrics like test coverage, bundle size, lint errors, or performance. The loop requires a measurable Verify command that outputs a single numeric metric to evaluate each iteration.

Yes, you need a clean git repository to run automated iterative optimization. A clean state allows the loop to track atomic changes, base decisions on commit history, and safely roll back discarded iterations.

You configure iteration limits and noise control using optional Iterations, Noise, and Min-Delta configurations. These settings control the loop length and prevent marginal metric fluctuations from being kept as false progress.

The best way to set up guarded verification is to pair a Verify command that outputs a numeric metric with an optional Guard step. This ensures regression prevention by validating genuine progress across build pipelines.