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.
arXiv:2608. 07545v1 Announce Type: cross Abstract: An LLM agent's capability depends not only on model weights but on its harness: prompts, tools, skills, and control flow.
arXiv:2610.15055v1 Announce Type: new Abstract: Self-evolving agents are usually scored by endpoint Gain: held-out success after an update minus success before. That difference can credit a frozen re...
arXiv:2610.17609v1 Announce Type: new Abstract: Training-free self-improvement lets an agent use feedback to revise the response it is about to return and to update the harness state that shapes how...
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.
arXiv:2608. 08466v1 Announce Type: new Abstract: Modern LLM agents are often improved by modifying prompts, tools, or workflows manually, while the executable scaffold surrounding the model---the \emph{harness}---is typically treated as a fixed artifact after deployment.
arXiv:2609.24974v1 Announce Type: cross Abstract: Agent harnesses, the external systems that mediate model-environment interaction, can substantially improve agent performance, but their gains remain...
The paper introduces WHALE, a method that alternates between updating a language model’s weights and searching for a better harness (the code that manages context and control flow). By iteratively fine‑tuning the model under the current harness and then optimizing the harness under the updated model, WHALE improves performance across search QA, math reasoning, and chess puzzles, outperforming weight‑only, harness‑only, and Fast‑Slow Training by 4.15–24.38 percentage points in mean@8 accuracy. The approach uses either fixed phase lengths or an adaptive patience rule to decide when to switch phases, and the authors provide code on GitHub.
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.
The paper introduces HARNESSEVO, a method that decomposes a large language model’s harness into four independently evolvable components—role, task‑strategy, tool/format‑rules, and reflection/control. Experiments on ALFWorld show that overall success rates are similar to flat‑string evolution, but the reflection/control component alone accounts for most of the performance gains. The study also finds that evenly distributing optimization budget across all slots can be detrimental; concentrating resources on the high‑credit control slot recovers lost performance, while on WebShop all slots remain ineffective, suggesting task‑specific differences in harness value.
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.
The paper introduces a method for co-evolving inspectable verifiers alongside self-improving agents. By synthesizing verifiers from clustered failures and selecting them based on agreement with a reference set, the authors demonstrate improved held‑out agreement on MBPP+ and outperform a bare LLM judge. They also show that removing anchor guards collapses the verifier into a vacuous grader, yet the collapsed verifier still trains skills effectively, indicating that downstream task scores cannot certify a self‑evolved verifier.
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.