An API testing tool built specifically for AI agent loops

I was working on a small API for an internal tool. I wanted my coding agent — Claude Code, in this case, but Cursor or opencode would have done — to take the boring part off my plate: write a happy-path test for each endpoint I added, run it, and fix it when something broke. The "write" part was great. Claude generated reasonable tests on the first shot. The "run" part was fine. The "fix" part is where it fell apart. Here's a typical failure as Jest spits it out: AssertionError: expected 2...