Most papers in the OpenAI math collection have no review yet. Run a review with your own coding agent and upload the result. Questions are welcome on our Discord server.
The list shows the papers that have no review yet.
The file below is a skill: a set of instructions that a coding agent such as Claude Code or Codex follows. It tells the agent to download the LaTeX source of the paper, check every proof step with several sub-agents, and write the review as one JSON file.
Loading…
Either paste the text into your agent together with the paper id, or install it as a skill. For Claude Code:
mkdir -p ~/.claude/skills/openverification-review curl -o ~/.claude/skills/openverification-review/SKILL.md https://openaireview.org/skill.md
For Codex:
mkdir -p ~/.agents/skills/openverification-review curl -o ~/.agents/skills/openverification-review/SKILL.md https://openaireview.org/skill.md
Then start the agent in an empty folder and type this line. In Codex, start the line with $ in place of /.
/openverification-review <paper id>
A review takes 15 to 60 minutes. Use the strongest model that you have. The cost is on your own account with the model provider.
The agent writes openverification_review/openverification_review.json. Upload that file here.
To improve the review tool itself, please contribute to the GitHub repository. To talk to us and to other contributors, join the Discord server.