Add LIFE-HARNESS: Runtime Harness Adaptation for Deterministic LLM Agents #12

Hi, thank you for curating this great repository!

I would like to suggest adding our recent work on runtime harness adaptation for LLM agents:

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

• Paper: https://arxiv.org/abs/2605.22166
• Code: https://github.com/Tianshi-Xu/Life-Harness

LIFE-HARNESS focuses on improving frozen LLM agents by adapting the runtime interface around them, rather than updating model parameters. It introduces a four-layer harness design covering environment contracts, procedural skills, action realization, and trajectory regulation. The harness is evolved from training trajectories, fixed during evaluation, and reused across different model backbones in deterministic agent environments.

I think this work is relevant to the repository because it studies harness design and evolution beyond coding agents, while sharing the same broader idea that much of an agent’s capability can come from the runtime system surrounding the model.

Thanks again for maintaining the list, and I would be grateful if you could consider including it.

@article{xu2026adapting,
  title={Adapting the Interface, Not the Model: Runtime Harness Adaptation for Deterministic LLM Agents},
  author={Xu, Tianshi and others},
  journal={arXiv preprint arXiv:2605.22166},
  year={2026}
}