Will Wild

An AI engineer that never repeats a failed fix. What it is Repo Rescue is an evidence-driven repair agent. Instead of keeping only chat history, it stores verified repair episodes — repo revision, issue fingerprint, attempted patch, test command, verifier result, and applicability constraints — and reuses them in later, fresh sessions. The problem AI coding assistants repeat failed fixes across sessions because a failed attempt is thrown away once the chat ends. Repo Rescue keeps both failures and successes, with the evidence that decided each one. How it works A bounded agent retrieves a stored lesson, proposes a patch, and an independent verifier — separate from the proposer — checks it in a fresh temporary copy of the fixture by running the test as a subprocess. Nothing merges or deploys without human review.