arXiv AI

What Does Multi-Harness RL Learn? Credit Assignment and Portability in Coding Agents

The study investigates how multi‑harness reinforcement learning (RL) affects coding agents by comparing two grouping strategies—Within (one group per task‑harness pair) and Cross (harnesses pooled within a task)—using a Qwen3‑8B policy trained on frozen task‑harness records from Aider, OpenHands, Qwen Code, and SWE‑agent. Across 24,000 sealed evaluations, the choice of evaluation harness dramatically increases solve rates (from 2.14 % to 9.27 %), while the grouping rule has a negligible effect. Both grouping rules yield similar gains on the same source harness, and Cross‑harness credit does not improve portability beyond Within‑harness credit, suggesting that multi‑harness RL reports should specify grouping boundaries and test on unseen harnesses.

arXiv AI
Oct 12

Multi-Harness Reinforcement Learning for Coding Agents: Training Gains and Harness Dependence

The study evaluates multi-harness reinforcement learning (RL) for coding agents by training two Qwen3-8B RL variants—one with rewards normalized within each harness and one across harnesses—using identical replay data. Results show modest gains over the supervised fine-tuning baseline (+0.48 and +0.77 percentage points for within and cross normalization, respectively), but no consistent advantage of cross normalization on unseen harnesses. Performance differences across harnesses are much larger than the training gains, indicating that harness dependence remains a significant challenge.

By Chenqian Le, Jiayi Cheng, Qijia He, Runhao Li, Yinghao Li, Xupeng Chen
arXiv AI
Oct 2

Cross-Benchmark Transfer from RL on Agentic Coding Tasks

The paper reports that applying reinforcement learning (RL) to the Kimi K2.7 Code model on 1,700 agentic coding tasks improves its performance on six external benchmarks. After a single epoch of GSPO training on a rank‑32 LoRA adapter, pass‑@1 scores increased across all benchmarks, with significant gains even on data released after training. The trained model also reduces agent steps and avoids common failure modes such as dropping requirements or breaking existing behavior.

By Sushant Mehta, Logan Ritchie, Edwin Chen
arXiv AI
Sep 15

HarnessBandit: Joint Learnability-Transferability Scheduling for Multi-Harness Agentic Reinforcement Learning

arXiv:2609.13739v1 Announce Type: cross Abstract: Language-model agents are increasingly deployed through diverse harnesses that differ in system prompts, tool schemas, control loops, and trajectory...

By Hongliang Wei (Harbin Institute of Technology, Alibaba Cloud), Xiaobing Tu (Alibaba Cloud), Yinggui Wang (Alibaba Cloud), Zhengxi Liu (Alibaba Cloud), Rongkun Xue (Alibaba Cloud), Jinkui Ren (Alibaba Cloud), Xiantao Zhang (Alibaba Cloud), Debin Zhao (Harbin Institute of Technology), Xiaopeng Fan (Harbin Institute of Technology)
arXiv Machine Learning
Sep 22

CHART: A Harness-Rotation Curriculum for Harness-Robust Search Agents

The paper introduces CHART, a curriculum that rotates harnesses during training to teach search agents parallel search strategies robustly across different harness configurations. Unlike static harness augmentation, CHART gradually consolidates behavior by graduating learned harnesses and replacing them, maintaining a reward gap that drives learning. Experiments show CHART enables agents to parallelize on 89% of held‑out harnesses, improves performance on a new QA task by 5.6pp, and benefits more from meta‑harness search than baselines.

By Xinlu Zhang, Ying-Chun Lin, Zhihan Zhang, Besnik Fetahu, Xi Chen
arXiv Computation and Language
Oct 8

CoTrace: Data Recipes for Training Terminal Agents with Harness-Model Co-Evolution

arXiv:2610.10426v1 Announce Type: new Abstract: Terminal-agent capability depends jointly on model weights and the runtime harness that formats prompts, binds tools, and handles error recovery. Exist...

By Jixuan Chen, Jiaxin Zhang, Qinyuan Ye, Yada Pruksachatkun, Haoxiang Zhang, Jingming Zhuo, Yifan Zhang, Yutong Dai, Juntao Tan, Xiangyu Peng, Silvio Savarese, Zeyuan Chen, Lianhui Qin, Chien-Sheng Wu
arXiv AI
Oct 9

Harness Evolution Hits a Ceiling: When Weight Training Should Begin

The paper investigates when to evolve a harness versus training model weights for long‑horizon LLM agents. By labeling failed trajectories with the first signal that fires, the authors separate process failures from content failures, showing that harness evolution fixes the former while weight training addresses the latter. Experiments on DeepPlanning and WebArena‑Lite demonstrate significant performance gains, with LoRA adapters trained on evolved harness trajectories further improving held‑out scores and reducing content failures.

By Yuan Tian, Bing Hu, Hao Wang, Binghang Lu, Fang Wu
arXiv AI
Nov 6

Measuring the Marginal Value of Verifier-Based RL Beyond Supervised Fine-Tuning

The paper evaluates the true benefit of verifier‑based reinforcement learning (RLVR) over supervised fine‑tuning (SFT) using a deterministic benchmark of 1,877 synthetic GPU and LLM training failure diagnosis tasks. In controlled experiments, SFT alone achieved high performance (0.785 on the hard split), while RLVR added no detectable improvement in‑distribution and only a modest gain under distribution shift (+0.33 points). The study concludes that most performance gains come from SFT, with verifier‑based RL contributing only small additional benefits.

By Amit Sagtani, Saloni Garg
arXiv Computation and Language
Oct 28

Your Policy Is Secretly a Harness: The Decision Geometry of Agent Interfaces

The paper investigates how language‑model agents learn from the harnesses—prompts, tools, and control logic—through which they are deployed. Using KL‑regularized control, it shows that a trained policy’s log odds reflect the harness’s action preferences, and that the agent’s deployed return equals the harness’s best possible return minus a mismatch term caused by the policy’s suboptimal choices. Experiments on MiniWoB++ and Terminal‑Bench Lite demonstrate that changing harnesses can drastically alter performance, and that the ceiling‑minus‑mismatch metric predicts gains better than either component alone.

By Jian Mu, Ruibo Shen, Yao Shu