Community · by leo-lilinxiao
Codex autoresearch Claude skill
The Codex autoresearch skill helps an AI assistant run repeated Git experiments toward a numeric target, keeping gains and undoing failures.
What the Codex autoresearch skill does
The assistant runs a loop of forming a hypothesis, changing code, measuring the result and learning from it. It keeps changes that help and reverts those that do not. It suits goals a command can measure, such as test failures, coverage, type errors, latency or binary size, and it can also report status or results without starting new experiments.
What it covers
- Reversible experiment loop that keeps or reverts changes
- Generate an HTML report of a run
- Run foreground or background experiments
- Target test failures, coverage, type errors, latency, binary size and other measurable outcomes
How to install it
In Codex
$skill-installer install https://github.com/leo-lilinxiao/codex-autoresearchWhat you need
- “Open a clean Git repository with Full Access”