Harneloop
Harneloop is an open-source framework for building and evolving task-specific AI agent harnesses through real attempts, artifact inspection, traced evidence, isolated candidates, and evidence-gated promotion.
Agents map the real environment, test work on the target task, inspect the produced result, trace likely failures, and preserve only changes that demonstrate improvement without unacceptable regressions.
View Harneloop on GitHub or read the ViperMesh case study.