RUC's ClawGym II Trains Agents via Black-Box RL on Real Harnesses

Renmin University's ClawGym II paper introduces a black-box RL framework that decouples execution from optimization, capturing tool calls from real agent harnesses like Claude Code and OpenClaw to stably train general agents across mixed frameworks.

2026-08-18 ~ 2026-08-19 · 3 related posts

• ClawGym II: Black-Box RL Stably Optimizes Agents Through Complex Harnesses — RUC · 2026-08-18
• ClawGymII: Training Models Directly via Black-Box Agent Harnesses — 智东西 · 2026-08-18
• ClawGym II paper: Improving agents via mixed-harness training — omarsar0 · 2026-08-19