Learning Meta-Skills for Agent Harness Design in Test-Time AI4AI
Abstract: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.
| Comments: | 22 Pages, 4 Figures, 5 Tables |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.38143 [cs.AI] |
| (or arXiv:2609.38143v2 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2609.38143 arXiv-issued DOI via DataCite |
Submission history
From: Cheng Qian [view email]
[v1]
Tue, 29 Sep 2026 17:55:56 UTC (1,237 KB)
[v2]
Thu, 1 Oct 2026 05:12:18 UTC (1,237 KB)