Agent Harness
Also called Harness (Agent engineering)
An agent harness runs interactions between a model and its tools. It supplies context, routes action requests, returns results, and manages when execution continues or stops.
[Anthropic][Hugging Face]In practice · hypothetical example
A model requests a file read. The harness dispatches the request and supplies the returned text to the next model call.
[Anthropic]A little deeper
The harness is execution software around the model. Its boundary varies across authors and products. [Anthropic][Hugging Face]
The model loop and tool routing layer. [Anthropic][Hugging Face]
May include configuration and infrastructure also described as scaffolding. Check the author’s scope. [Hugging Face]
A common mix-up
Harness and scaffold always mean different components.
Their boundaries overlap in some usage. An evaluation harness is a separate testing context. [Hugging Face]
Similar terms, different emphasis
Compare Agent Harness & Scaffold →The model requests a file read. What dispatches it in this example?
Sources & editorial notes
Evidence: supported. Primary-source support for this scoped entry; publication approved by the project owner.
- Harness, Scaffold, and the AI Agent Terms Worth Getting Right ↗ (opens in new tab)Hugging Face · 2026-05-25
Relevant section: Scaffolding; Harness - Scaling Managed Agents: Decoupling the brain from the hands ↗ (opens in new tab)Anthropic · 2026-04-08
Relevant section: Virtualized components; Decouple the brain from the hands
Last editorial review: 2026-09-13 by project-owner.
First observed in this corpus: Unknown.
Revision history
Revision 2 · Created 2026-09-12 · Updated 2026-09-13
Project owner approved the current content for publication. Existing evidence scope and limitations remain applicable.