OpenForgeRL: Train Harness-native Agents in Any Environment
AAIF 摘要
论文提出 OpenForgeRL，把 Claude Code、Codex、OpenClaw 这类真实 inference harness 放进训练闭环：代理层记录多轮模型调用和工具使用，Kubernetes 为 rollout 提供隔离容器，再把轨迹接入 veRL 等强化学习栈。它的价值在于把 agent 学习难度从抽象 benchmark 拉回真实环境、工具状态和 harness 行为差异。
English abstract
Modern AI agents rely on elaborate inference harnesses such as Claude Code, Codex, and OpenClaw to drive multi-turn reasoning, tool use, and access to external systems. While powerful, these complex harnesses also make agents hard to train end-to-end with open infrastructure, whose SFT/RL stacks cannot natively express stateful, multi-process harness inference. To address this, we present OpenForgeRL, an open-source framework for training harness-based agents end-to-end in diverse environments. OpenForgeRL achieves this with a lightweight proxy that serves the harness's model calls while recording them as training data for a standard RL codebase (e.g., veRL), and a Kubernetes orchestrator that runs each rollout in its own remote container, together enabling training on any harness in any environment at scale. By decoupling training and inference, OpenForgeRL allows researchers to easily train, study, and improve agents directly in the real harnesses and environments they are deployed with. We validate our framework across diverse, complex harnesses and environments, spanning tool/claw-based agents and multimodal GUI browser- and computer-use agents. Using only hundreds to a few thousand tasks, OpenForgeClaw reaches 31.7 pass^3 and 55.9 pass@3 on ClawEval and 33.7 on QwenClawBench. OpenForgeGUI reaches 37.7 on OSWorld-Verified, 63.0 on Online-Mind2Web, and 72.3 on WebVoyager. Both outperform open baselines of similar size on nearly all benchmarks, and in the GUI setting match or surpass models several times larger. Beyond benchmarks, we analyze how harness choice (e.g., ZeroClaw, OpenClaw, Codex) and RL shape agent behavior. We find that some harnesses are substantially harder to learn than others, and that RL improves agentic reliability, such as self-verification, tool coverage, and completing multi-step plans, though critical abilities such as error recovery remain weak.
Intake rationale
Agent 训练正在从离线任务转向真实 harness 内的轨迹、工具和容器化 rollout。
Obsidian evidence excerpt: 这篇把 agent 训练放回真实 harness：代理层记录模型调用，Kubernetes 为每个 rollout 开远端容器，再把数据接到 veRL 一类 RL 栈。摘要里的覆盖面很适合跟工程里的 Codex/Claude Code/OpenClaw 类工具对照：同一个模型，换 harness 后学习难度和行为会变。
Metadata
- Authors: Xiao Yu, Baolin Peng, Ruize Xu, Hao Zou, Qianhui Wu, Hao Cheng, Wenlin Yao, Nikhil Singh, Zhou Yu, Jianfeng Gao
- Original date: 2026-07-23
- Source: https://arxiv.org/abs/2607.21557
- Local intake source:
OpenClaw定时任务/论文流水线/2026-07-25-论文流水线.md