Show HN: What 1k Harness Experiments Taught Me About Self-Improving Agents

Category: ai-ml

Tags: self-improving-agent, llm-agent, experiment

Score: 5.5/10 (Innovation: 6, Technical: 6, Documentation: 6, Utility: 4)

This project implements a self-improving loop where an outer coding agent proposes patches to an LLM-driven shell agent 'harness' that solves Terminal-Bench tasks, then evaluates and commits or reverts changes based on binomial tests. It is interesting for exploring meta-learning and automated agent optimization in a controlled experimental setup.

Target audience: ai researchers, ml engineers

Repository: https://www.henrypan.com/blog/2026-05-25-self-improvement-harness/ · Python · MIT · 1 stars

View on Hacker News