HarnessPro: Recursive Harness Self-Improvement with Pareto Optimization for Performance-Cost Trade-offs
What this work does
HarnessPro recursively improves agent instructions, tools, and runtime policies while retaining several performance-cost operating points. It replaces direct trajectory judging with fixed executable tests synthesized from public task evidence.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.