Autoresearch Improvement Loop
Runs a bounded, evaluator-driven improvement loop on one mission and logs every iteration
- Repository
- Yeachan-Heo/oh-my-claudecode
- Stars
- 40,512
- Licence
- MIT
- Files
- 1 file
What it does
This skill tells an AI agent how to keep improving one defined task over repeated attempts, rather than stopping after one try. Each cycle the agent makes one change, runs an evaluator (a script or command that returns a pass/fail result and optional score), and saves the result as JSON plus a readable markdown log. It keeps going after failed checks and stops only at a set maximum runtime or another explicit stop condition.
Great for
- Developers who already have a mission and evaluator set up and want repeated, logged improvement attempts
- Teams running periodic improvement reruns on a single task via Claude Code's cron feature
- Anyone who wants a traceable history of each experiment and decision
Skip it if
- Users who have no evaluator yet: the skill expects one to exist and says evaluator generation belongs to /deep-interview --autoresearch
- Multi-task or multi-mission orchestration: the skill explicitly forbids it in v1
- Unattended long runs without a firm runtime cap: the loop keeps iterating through failures, so it can change files repeatedly until the limit is hit
Try asking
- “Use /deep-interview --autoresearch to define my mission and evaluator, then run autoresearch on it for 2 hours”
- “Continue the autoresearch run for the mission in .omc/autoresearch and show me the decision log”
- “Set up a nightly cron job that reruns the autoresearch mission and appends new run artifacts”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
autoresearch.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip autoresearch.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip autoresearch.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 3
Ask for the task
No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.
Get better results
- Set an explicit max-runtime before starting so the loop has a firm stop point
- Review the markdown decision logs under .omc/autoresearch periodically, since the agent changes code on each iteration
- Run it inside version control so every iteration's changes can be reviewed or reverted
What's inside
- SKILL.md
Redistributed unmodified from Yeachan-Heo/oh-my-claudecode under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.