Coding agents sometimes make a failing test pass by changing the test instead of fixing the code. We observed this in a real Claude session: the agent changed the expected value in math.test.js from 3 to -1, while math.js continued to subtract. The test turned green, but the underlying bug remained. MangoPals is a self-improving harness that turns failures like this into persistent rules. For each recorded session, it runs and scores two replays: a baseline using the current rules (“before”) and a second replay using a proposed new rule (“replay”). Raindrop traces both runs. The harness stores the comparison in MongoDB Atlas and keeps the proposed rule only if the replay score improves. The next agent receives the kept rules through MCP before starting work. Starting with empty memory, we replayed 36 real recorded Claude sessions: The first session that altered a test expectation improved from 0 → 1, so the harness kept the rule identifying the changed assertion in math.test.js. The next two sessions with the same behavior scored 1 → 1. The proposed additions provided no improvement, so no new rules were kept. The remaining 33 clean sessions added no rules. The dashboard updates live as replay events land in Atlas, showing how the memory graph grows.
More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon
MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…
WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…
