Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI
Frozen Builder models learn reusable Meta-Skills that improve agent harnesses by 8.95 points on Harness-Bench and NewtonBench.
The paper studies test-time AI-for-AI, where a Builder with frozen weights learns to construct better execution environments for a frozen Target model. Reusable Meta-Skills specify when support is needed and which resources to provide, learned from development-set feedback and then applied to unseen tasks. On Harness-Bench and NewtonBench, the full skill bank raises macro-average performance by 8.95 points over no-skill construction and 12.02 points over handing the same bank directly to the Target.
- A Builder learns reusable Meta-Skills from Target execution feedback.
- Model weights stay fixed while skills shape harnesses for unseen tasks.
- The full skill bank improves macro-average performance by 8.95 points.
- It beats directly giving the same bank to the Target by 12.02 points.
Full article145 words · extracted from huggingface.co · click to collapse
Agent performance depends on both reasoning ability and the environment in which it acts. We study test-time AI-for-AI, asking how a Builder can learn to construct better execution environments for a Target while both models' weights remain fixed. To make the Builder's experience reusable, we introduce Meta-Skill: principles specifying when support is needed and what resources to provide. The Builder learns these principles from Target's execution feedback on the development set, then uses the frozen skill bank to construct harnesses for unseen tasks. Across Harness-Bench and NewtonBench, full-bank meta-skills improve macro-average performance by 8.95 percentage points over no-skill construction, and 12.02 points over direct delivery of the same bank to the Target. These results highlight the value of translating experience into executable support. Gains when the same model serves both roles further suggest a path to system level self-improvement through learning to build better environments.
Text extracted automatically; images, tables and formatting may be missing. Original: https://huggingface.co/papers/2609.38143