What is an agent harness?
Answer out loud or in your head first, then compare.
Answer
It's the software around a model that lets it work as an agent. Among other things, it passes context, exposes tools, runs the allowed calls, records results and stops the work. The model itself proposes actions, but it needs something to carry them out.
You can compare it to a workstation: the model has skills, but it needs tools and rules for using them. In a coding agent those capabilities are reading files, editing and running tests. The harness should also limit time and access. The instruction “don't do anything dangerous” doesn't replace restricting permissions in the environment itself.
Why it matters: One of the most important terms in this offer.
Possible follow-ups: What must code enforce, and what can be described in the prompt?
How did it go? Be honest — “Not yet” puts the question on your review list.
Practise in code
- A minimal agent loop — runs tests in your browser
Further reading
- Anthropic: Effective harnesses for long-running agents(opens in a new tab)26.11.2025. Working in stages, progress artifacts and resuming between sessions.