RuleRepro

Your tests can stay green while your requirements change. RuleRepro connects a changed permission policy to human-reviewed rules and concrete, reproducible failures. The workflow is: compare requirement revisions; propose source-linked rules with Meta Muse Spark; review the rule and adapter; explicitly approve; check trusted local Python snapshots; save the counterexample; replay it without a model; export a pytest regression. Our team-documents example makes the problem tangible. An editor on team T1 can export a T2 document even though the existing test passes. RuleRepro reproduces that policy violation. The same approved rule checks the revised implementation, and the exported regression fails on the draft and passes on the fix. Frozen evidence preserves code, requirements, rules, approval metadata and the failing case. The project includes a local review UI, CLI, isolated per-project review/job data, a small TOML onboarding configuration, and a keyless judge quickstart. Local verification recorded 834 passing tests, plus two independent configured-project checks covering approval isolation, replay and regression export. Scope is deliberately explicit: trusted local Python, configured adapters and two supported Boolean permission domains. Execution is not sandboxed; clean results cover tested cases, not all possible behavior. Imports neither execute code nor install dependencies. Judges can bring a compatible configured project without core-source edits.