The question
A code reviewer can point to a line, describe a failure, and sound entirely convincing. Someone still has to establish whether the failure actually happens.
Habeas gives that job to the auditor. Its name borrows from habeas corpus. A claim comes with a demand to produce the evidence.
What happens in an audit?
Habeas takes one small part of a Git checkout at a time. One agent looks for problems. Another checks the surrounding code. A fresh critic sees the claim and tries to find a reason to dismiss it.
A surviving bug claim goes to a prover, which writes a test. The test has to fail twice on the current code for the stated reason. A final reviewer checks that the failure demonstrates the original claim.
What comes out?
A confirmed bug gets an issue folder with the test and its evidence. When a claim runs out of proof attempts, it goes into quarantine. Suggestions such as missing test coverage can still be recorded as advisory findings.
The repository contains the audit workflow, proof scripts, and setup instructions for running it on a checkout of your own.
Read the Habeas README