MemoHarness: Agent Harnesses That Learn from Experience
Quick Answer
MemoHarness is an adaptive optimization framework for agent harnesses that learns from execution experiences, improving performance over static configurations in benchmarks like shell-agent and code generation.
Quick Take
MemoHarness is an adaptive optimization framework for agent harnesses that learns from execution experiences, improving performance over static configurations in benchmarks like shell-agent and code generation. It utilizes a dual-layer experience bank to adapt harnesses without needing test-time labels, showing selective transfer to unseen tasks while remaining cost-effective.
Key Points
- MemoHarness decomposes harnesses into six editable control dimensions.
- It stores diagnoses and patterns in a dual-layer experience bank.
- The framework adapts harnesses without requiring test-time labels or feedback.
- MemoHarness shows improved performance in analytical reasoning benchmarks.
- Its context remains cost-competitive with cacheable retrieved experiences.
Paper Resources
📖 Reader Mode
~2 min readAbstract:An agent harness is the external control layer that turns a base LLM into an executable agent by managing context, tools, orchestration, memory, decoding, and output handling. While harness design strongly affects agent behavior, most automatic improvement methods optimize narrower artifacts such as prompts, pipelines, or workflows, and deployed agents usually reuse a single global harness for all cases. We introduce MemoHarness, an adaptive harness optimization framework that learns from its own executions. MemoHarness decomposes the harness into six editable control dimensions, stores per-case diagnoses and distilled global patterns in a dual-layer experience bank, and adapts the learned harness to each test case using retrieved experience without test-time labels, feedback, or additional search. In our evaluation across shell-agent, code-generation, and analytical-reasoning benchmarks, MemoHarness improves over the fixed harnesses we compare against and shows selective transfer to unseen suites and base models. Its additional context can also remain cost-competitive when much of the retrieved experience is cacheable. These results provide evidence that execution experience is a practical substrate for building agent harnesses that are more adaptive than a single static configuration, while leaving broader claims about statistical robustness and component attribution to future work.
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| Cite as: | arXiv:2607.14159 [cs.AI] |
| (or arXiv:2607.14159v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2607.14159 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yue Huang [view email]
[v1]
Tue, 14 Jul 2026 21:22:18 UTC (1,323 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.AI
See more →Automatic Ordinary Differential Equations Discovery For Biological Systems Using Large Language Model Powered Agentic System
The MEDA system utilizes large language models and symbolic regression to autonomously discover ordinary differential equations for biological systems, achieving strong structural recovery and biologically plausible models. It outperforms existing methods by integrating domain knowledge and mechanistic constraints, demonstrating effective retrieval and extrapolation capabilities.