I built a lie detector for AI coding agents. Then I read the rules.

Developer Dann Waneri has created 'claimcheck,' a verification system designed to ensure AI coding agents are truthful about their code modifications. The tool requires agents to submit a JSON claim detailing their changes, which is then verified against the actual diff. The system enforces rules such as scope validation, protected path restrictions, and description accuracy, utilizing an LLM to confirm that the agent's stated actions match the code changes. If a claim is found to be false or inconsistent, the system rejects the merge. Built on Cloudflare using Workers and Durable Objects, the project successfully demonstrated its ability to handle multiple concurrent agents, catching both intentional and accidental misreports. While originally intended for a competition, the project remains a functional open-source prototype that highlights the importance of verifying agent claims in multi-agent development environments, ensuring that AI-driven code contributions are transparent and accurate.