autoresearch-agent

Automate code performance tuning by running repeated edit, evaluate, and compare cycles on a target file. The process keeps only changes that improve your measurable metric and resets failed attempts using git rollback automation.

The best way to reduce bundle size iteratively is to run an optimization loop that modifies the target file, measures the size output via a fixed evaluation command, and commits only winning changes while rolling back failures in git.

Yes, you can use an automated experiment loop for prompt optimization by targeting your prompt file and evaluating it against a measurable quality metric. The loop edits the prompt, runs the evaluation command, and retains only improvements.

Yes, a git repository is required for automated file optimization loops. The process relies on git rollback automation to discard failed experiment attempts safely and commit only the changes that yield measurable gains.

Git rollback during benchmarking works by discarding failed edit attempts that do not improve the measurable metric. It automatically resets the target file to the last winning state and commits only changes that produce measurable gains.

A key limitation is that the experiment loop requires a fixed evaluation command that reliably prints a metric. It optimizes a single target file at a time, making it unsuitable for multi-file refactoring or tasks lacking a quantifiable evaluation metric.