LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents

• URL: http://arxiv.org/abs/2608.17393v1
• Date: Tue, 18 Aug 2026 05:34:35 GMT
• Title: LEGO-RL: Harness-Native Reinforcement Learning for Coding Agents
• Abstract summary: We present LEGO-RL, a framework that bridges native coding-agent harnesses with scalable policy-gradient optimization.<n>We evaluate LEGO-RL by training the sparse MoE model Qwen3.5-35B-A3B with GSPO across three native coding-agent harnesses.
• Score: 56.902747731541695
• License: http://creativecommons.org/publicdomain/zero/1.0/
• Abstract: Reinforcement learning for coding agents increasingly relies on long-running agent harnesses to manage tool integration, repository contexts, and execution feedback. However, the native execution environments of these harnesses are inherently misaligned with policy-gradient training: environmental crashes and reward hacking corrupt outcome signals, while train-inference discrepancies decouple rollout behavior from policy updates. To address this, we present LEGO-RL, a framework that bridges native coding-agent harnesses with scalable policy-gradient optimization without modifying their internal control flow. LEGO-RL is built upon three pillars: (1) faithful optimization via in-process LLM proxying that captures raw generation streams for token-level alignment and robust trainer-side log-probability recomputation, even under harness-side compaction or re-serialization; (2) reliable execution via scalable sandbox orchestration featuring image caching and stage-wise defenses to mitigate reward hacking; and (3) observable training through an integrated plugin that automates validation and monitoring, paired with a Live UI for granular trajectory diagnostics. We evaluate LEGO-RL by training the sparse MoE model Qwen3.5-35B-A3B with GSPO across three native coding-agent harnesses. LEGO-RL improves Qwen3.5-35B-A3B across OpenHands SDK (64.0% to 70.4%), Claude Code (62.4% to 68.2%), and OpenCode (57.2% to 66.6%) on SWE-bench Verified, while maintaining a rollout-training probability correlation above 0.99.

Related papers

• Agent Lightning v1.0: Towards Harnessed Agentic RL [28.087691629023038]
  We present Agent Lightning v1.0, a lightweight framework for harnessed agentic RL.<n>It supports arbitrary agent harnesses and serves as a practical testbed for studying these challenges.<n>We evaluate it on instruction-following, search, and coding agents, and provide a complete reproducible pipeline for coding-agent RL.
  arXiv  Detail & Related papers  (2026-08-18T08:50:13Z)
• Efficient Agentic Reinforcement Learning with On-Policy Intrinsic Knowledge Boundary Enhancement [16.192937389387982]
  We propose AKBE (Agentic Knowledge Boundary Enhancement), an on-policy method that dynamically probes the model's intrinsic knowledge boundary.<n> Experiments on seven QA benchmarks demonstrate that AKBE improves task accuracy by +1.85 on average and reduces tool calls by 18% over standard agentic RL.
  arXiv  Detail & Related papers  (2026-05-26T12:42:23Z)
• ChipMATE: Multi-Agent Training via Reinforcement Learning for Enhanced RTL Generation [55.947962672433675]
  ChipMATE is the first self-trained multi-agent framework for RTL generation.<n>Inspired by industrial practice, ChipMATE pairs a Verilog agent with a Python reference-model agent that mutually verify each other's outputs.<n>ChipMATE achieves 75.0% and 80.1% pass@1 on VerilogEval V2 with 4B and 9B base models.
  arXiv  Detail & Related papers  (2026-05-13T01:04:21Z)
• Agent^2 RL-Bench: Can LLM Agents Engineer Agentic RL Post-Training? [23.212983812736343]
  Agent2 RL-Bench is a benchmark for evaluating agentic RL post-training.<n>We find that supervised pipelines dominate agent-driven post-training under fixed budgets.
  arXiv  Detail & Related papers  (2026-04-12T09:35:27Z)
• RLVE: Scaling Up Reinforcement Learning for Language Models with Adaptive Verifiable Environments [111.87296453908199]
  We introduce Reinforcement Learning with Adaptive Verifiable Environments (RLVE)<n>RLVE enables each verifiable environment to dynamically adapt its problem difficulty distribution to the policy model's capabilities as training progresses.<n>We show that environment scaling, i.e., expanding the collection of training environments, consistently improves reasoning capabilities.
  arXiv  Detail & Related papers  (2025-11-10T17:18:35Z)
• Efficient Multi-turn RL for GUI Agents via Decoupled Training and Adaptive Data Curation [65.3648667980258]
  Vision-language model (VLM) based GUI agents show promise for automating complex tasks, but face significant challenges in applying reinforcement learning (RL)<n>We propose DART, a Decoupled Agentic RL Training framework for GUI agents, which coordinates heterogeneous modules in a highly decoupled manner.<n>On the OSWorld benchmark, DART-GUI-7B achieves a 42.13% task success rate, a 14.61% absolute gain over the base model, and 7.34% higher than open-source SOTA.
  arXiv  Detail & Related papers  (2025-09-28T13:19:20Z)
• rStar2-Agent: Agentic Reasoning Technical Report [25.266747156205266]
  We introduce rStar2-Agent, a 14B math reasoning model trained with agentic reinforcement learning to achieve frontier-level performance.<n>To this end, rStar2-Agent boosts a pre-trained 14B model to state of the art in only 510 RL steps within one week, achieving average pass@1 scores of 80.6% on AIME24 and 69.8% on AIME25.
  arXiv  Detail & Related papers  (2025-08-28T12:45:25Z)
• Ring-lite: Scalable Reasoning via C3PO-Stabilized Reinforcement Learning for LLMs [51.21041884010009]
  Ring-lite is a Mixture-of-Experts (MoE)-based large language model optimized via reinforcement learning (RL)<n>Our approach matches the performance of state-of-the-art (SOTA) small-scale reasoning models on challenging benchmarks.
  arXiv  Detail & Related papers  (2025-06-17T17:12:34Z)
• DigiRL: Training In-The-Wild Device-Control Agents with Autonomous Reinforcement Learning [61.10299147201369]
  This paper introduces a novel autonomous RL approach, called DigiRL, for training in-the-wild device control agents. We build a scalable and parallelizable Android learning environment equipped with a VLM-based evaluator. We demonstrate the effectiveness of DigiRL using the Android-in-the-Wild dataset, where our 1.3B VLM trained with RL achieves a 49.5% absolute improvement.
  arXiv  Detail & Related papers  (2024-06-14T17:49:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.