Harneloop

Harneloop is an open-source, agent-first framework for building and evolving task-specific AI agent harnesses through artifact-aware testing, trace-backed diagnosis, and evidence-gated promotion.

The agent attempts a real task, captures the resulting artifacts, logs, traces, and state, compares the result with the desired outcome, and traces likely mistakes back through the recorded run. It can then propose a candidate change to its instructions, context, tools, retrieval, validators, or environment and test again. The working harness stays unchanged until evidence shows that the candidate improves the target without unacceptable regressions.