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

TL;DR

This paper introduces Life-Harness, a runtime adaptation method that improves deterministic LLM agents by modifying the interaction interface without changing the underlying model weights, leading to significant performance gains.

Contribution

Life-Harness is a novel, lifecycle-aware runtime harness that adapts the model-environment interface, enhancing agent performance without retraining models or altering evaluation setups.

Findings

Improves 116 out of 126 settings across 7 environments and 18 models.

Achieves an average relative improvement of 88.5%.

Transfers across models, indicating reusable environment-side adaptations.

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 during held-out evaluation. On seven deterministic environments from $\tau$-bench, $\tau^2$-bench, and AgentBench, Life-Harness improves 116 out of 126 model--environment settings across 18 model backbones, with an average relative improvement of 88.5%. Harnesses evolved only from Qwen3-4B-Instruct trajectories transfer to 17 other models, showing that Life-Harness captures reusable environment-side structure rather than model-specific behavior. These results position runtime interface adaptation as a complementary alternative to model-centric agent training. Code is available at GitHub.