maxwellccc/autonomous-qa-loop

Autonomous QA Loop is a small repository providing a prompt pattern for running repeated quality assurance passes on software projects using AI coding agents. It is released under CC0, meaning anyone can use, modify, or redistribute it freely for any purpose. The project addresses a specific limitation of AI-assisted code review: agents tend to become less effective after the first review pass. Once an agent has seen prior assumptions, suspected fixes, and earlier conclusions, it gravitates toward the same areas and misses other categories of bugs. The solution the project proposes is starting each review round with a completely fresh agent that has no conversation history. Each fresh agent receives only the original requirement, the code or artifacts to review, and authoritative context documents. Nothing about suspected problems, intended fixes, or prior opinions is passed along. The repository provides a file called PROMPT.md that defines the structure any generated QA prompt must follow. The prompt must contain exactly four sections: Background, Goal (Original Request), Review Target, and Relevant Context Documents. After each fresh pass, findings are collected and triaged in the main conversation thread, confirmed bugs are fixed, and then another fresh neutral pass begins. The loop continues until independent passes stop surfacing meaningful new issues. For large codebases, the README recommends splitting the review scope by module, running fresh agents on each part in parallel, then consolidating the findings together before starting another round. The tool can also be installed as a skill in Codex, OpenAI's coding agent, by copying the included folder to the Codex skills directory. The repository includes a YAML configuration file for OpenAI agent integration. A complete Chinese translation of the README is also included.