🤓 Latest Submissions
BugProof: From API Failure to Bob Fix
A developer can see that an API response is wrong but still lack a useful handoff for the person or AI assistant who will fix it. The request, expected contract, observed result, failing test, and repair instructions often live in separate messages. That slows investigation and makes it easy to declare a fix before the original failure has been captured. BugProof turns these fragments into a concrete developer task. The workbench accepts a method, endpoint path, request JSON, and expected and observed responses. It compares HTTP status and response values, then produces an editable Vitest regression test, an issue summary, and a focused task for IBM Bob IDE. The developer can run the test before a change, hand the case to Bob in an authorized repository, review the patch, and rerun the same assertion. The pasted case stays in the browser tab until the developer copies or downloads it. A built-in example loads an actual failing call from BugProof's inventory service. Seven requested units exceed five in stock, but the original code returns success and negative stock. The guided lab lets a reviewer step through that failure, inspect Bob's completed validation and capacity repair, and run twelve live comparisons. The evidence vault connects the code to genuine Bob session summaries and recorded tests. This is a working prototype with a reusable front-end handoff and a complete repair demonstration. The repository has 78 passing tests. Across eleven independent recorded scenarios, the original implementation fails nine and the repaired implementation passes all eleven. These are functional results, not a measured time-saving percentage. The workbench does not call arbitrary external endpoints or run IBM Bob from the website. The generated test and Bob task are meant for a developer to review and use in their own repository. The inventory example resets on each request and does not model persistent or concurrent stock. That scope is shown clearly so the output remains useful without overstating what was verified.
27 Sep 2026
