Adapting the Interface, Not the Model: Runtime Harness Adaptation for Deterministic LLM Agents

arXiv:2605.22166v3 Announce Type: replace Abstract: LLM agents are shaped not only by their language models, but also by the runtime harness that mediates observation, tool use, action execution, feedback interpretation, and trajectory control. While existing agent adaptation methods mainly update model parameters, many failures in deterministic, rule-governed domains stem from mismatches at the model--environment interface. We propose Life-Harness, a lifecycle-aware runtime harness that improves frozen LLM agents without changing model weights or evaluation environments. Life-Harness evolves from training trajectories by converting recurring interaction failures into reusable interventions across environment contracts, procedural skills, action realization, and trajectory regulation, and remains fixed for evaluation on unseen tasks. We evaluate Life-Harness across eight environments from AgentBench, $\tau$-bench, $\tau^2$-bench, and AppWorld using 18 open-weight backbones, and further test three frontier models on seven environments. With Qwen3-4B-Instruct, Life-Harness exceeds Meta-Harness and AHE by 28.7 and 7.5 percentage points in mean accuracy across the seven shared environments, respectively, while using 8.5\% and 68.4\% fewer optimization tokens. The evolved harnesses also benefit GPT 5.5, Opus 4.8, and Gemini 3.1 Pro, yielding an average accuracy gain of 3.5 percentage points. Code is available at https://github.com/Tianshi-Xu/Life-Harness.

Abstract:LLM agents are shaped not only by their language models, but also by the runtime harness that mediates observation, tool use, action execution, feedback interpretation, and trajectory control. While existing agent adaptation methods mainly update model parameters, many failures in deterministic, rule-governed domains stem from mismatches at the model--environment interface. We propose Life-Harness, a lifecycle-aware runtime harness that improves frozen LLM agents without changing model weights or evaluation environments. Life-Harness evolves from training trajectories by converting recurring interaction failures into reusable interventions across environment contracts, procedural skills, action realization, and trajectory regulation, and remains fixed for evaluation on unseen tasks. We evaluate Life-Harness across eight environments from AgentBench, $\tau$-bench, $\tau^2$-bench, and AppWorld using 18 open-weight backbones, and further test three frontier models on seven environments. With Qwen3-4B-Instruct, Life-Harness exceeds Meta-Harness and AHE by 28.7 and 7.5 percentage points in mean accuracy across the seven shared environments, respectively, while using 8.5\% and 68.4\% fewer optimization tokens. The evolved harnesses also benefit GPT 5.5, Opus 4.8, and Gemini 3.1 Pro, yielding an average accuracy gain of 3.5 percentage points. Code is available at this https URL.

Artificial Intelligence (cs.AI)

arXiv:2605.22166 [cs.AI]

(or arXiv:2605.22166v3 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2605.22166 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

aiqbench analysis

Adapting the Interface, Not the Model: Runtime Harness Adaptation for Deterministic LLM Agents

Related coverage