ProofRun provides a local verification receipt for AI coding agents. It does not judge code correctness; instead it cryptographically proves which checks actually ran against the exact code you have now. The problem is that an AI agent may claim "all tests pass" even when the test run is stale or never happened.

ProofRun closes the gap by making the claim itself checkable. Running a check creates a receipt that binds the result to a fingerprint of your current git state plus a SHA‑256 hash of any uncommitted changes. If a single byte changes, the receipt flips to STALE automatically, and no one needs to remember to ask whether the PASS still counts.

Installation is simple: a curl command downloads the binary, or you can go install from source. A quick start includes initializing .proofrun.yml, running a check, and verifying status with optional strict mode. The tool never uses LLMs, makes zero network calls, and compares argument arrays exactly, not strings. It supports pytest, npm run build, linting, and more.

Built by an AI coding agent (Claude Code) and reviewed adversarially, ProofRun is designed to hold AI agents accountable. It integrates with GitHub Actions to gate pull requests, and the project is MIT‑licensed and actively maintained.