skill-forge

Automate AI skill improvement by running an autoresearch loop that mutates target files, evaluates changes, and preserves only beneficial mutations. Skill Forge supports scheduled tasks to run unattended overnight and deliver a morning report.

Autonomous codebase optimization mutates target files and scores them against a numeric shell metric, like test coverage or bundle size. The evaluation mechanism keeps high-impact mutations and automatically reverts harmful regressions.

Yes, you can optimize arbitrary shell metrics like Docker image size or test coverage using Generic Mode. It applies the autoresearch paradigm to mutate any target file against a numeric shell metric to reduce size or improve quality.

Set up and validate skill mutations using the built-in setup wizard and dry-run validation. You can use the optional Guided mode for interactive control to validate changes before scheduling unattended overnight optimization runs.

If an automated code mutation crashes or causes a regression, crash handling mechanisms recover the process and harmful mutations are automatically revoked by reverting the files. TSV experiment logs track all changes for safety.