Blog

This article provides a systematic overview of rubrics in the context of large language models. Rubrics are structured, multi-dimensional scoring criteria that break down vague quality judgments into explicit, individually assessable standards. They are particularly valuable for evaluating open-ended tasks like report writing and agent workflows, where no single correct answer exists.

This article starts from the basic concept of harness and uses Claude Code as an example to give a full introduction to how harness works in a real product. Then, based on my recently open-sourced project SearchClaw, I will further discuss how harness design can be applied to deep search and deep research agents. Finally, I will share some of my personal thoughts and understanding of the concept of harness.

This article starts from the basic concepts of AI agents, then introduces OpenClaw's background, architecture, workflow, security risks, and some of my personal thoughts. This is a beginner-friendly article, so no prior knowledge is needed.