Autoresearch

This is a goal-directed iteration engine that modifies code, checks it against a metric, and keeps or discards changes in a loop. The core loop runs 25 times by default, but there are 12 subcommands for specific workflows: debugging hunts bugs through hypothesis testing, fix crushes errors until none remain, security runs STRIDE audits with red-team personas, and ship walks through an 8-phase deployment checklist. Everything logs to timestamped directories and you can chain subcommands together. It's bounded by default but you can override to unlimited if you're feeling brave. Use it when you need systematic improvement against a measurable goal rather than ad-hoc tweaking.

Install to Claude Code

npx -y skills add uditgoenka/autoresearch --skill autoresearch --agent claude-code