Auto Review Loop
Makes an agent repeatedly get Claude to review its research work, fix issues, and re-review until it passes.
- Stars
- 18,482
- Licence
- MIT
- Files
- 1 file
What it does
This skill makes an AI agent run an automatic review cycle on a research project. The agent sends its work to a separate Claude model acting as a critical reviewer, reads the score and weaknesses it returns, makes fixes, and asks again. It stops when the reviewer gives a positive verdict or after a fixed maximum of 4 rounds. Progress is logged to files in a review-stage folder so the loop can resume after the conversation is compacted.
Great for
- Researchers who want a second model to stress-test a paper or experiment before submission
- Codex users who want Claude Code to act as the reviewer rather than another Codex agent
- People who want an unattended improvement loop and can check the final log afterwards
Skip it if
- Final approval decisions: the skill's own acceptance markers are generated by the agent, so a human should still judge the result (risk: false confidence)
- Projects where the content is confidential and must not be sent to an external model service (reason: reviews are routed through a Claude reviewer bridge)
- Users without the claude-review MCP bridge set up (reason: the loop cannot run without it)
Try asking
- “Run an auto review loop on my draft paper until it passes review”
- “Review my experiment results and implement fixes until the reviewer is satisfied, max 4 rounds”
- “Use hard difficulty with reviewer memory to challenge my research claims”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 1 original files, the MIT licence and an attribution note.
auto-review-loop.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip auto-review-loop.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip auto-review-loop.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
- Start with the default medium difficulty; use hard or nightmare only when you want deeper adversarial checks, as they take more time and calls.
- Check review-stage/AUTO_REVIEW.md and REVIEW_STATE.json after each run rather than trusting the final verdict alone.
- Keep the HUMAN_CHECKPOINT option on if you want to approve each round's suggested fixes before the agent edits your work.
What's inside
- SKILL.md
Redistributed unmodified from wanshuiyin/Auto-claude-code-research-in-sleep 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.