Summary I researched your question about whether OpenHands has documentation for connecting trajectory generation with existing RL frameworks like VERL. Answer: No, there is no official OpenHands documentation about connecting trajectory generation with existing RL frameworks like VERL. Key Findings: What OpenHands Provides: • Built-in trajectory saving via environment variables (

SAVE_TRAJECTORY_PATH

,

REPLAY_TRAJECTORY_PATH

) • API endpoint for exporting trajectories:

/conversations/{conversation_id}/download

• Trajectory export format: ZIP file containing individual event JSON files and metadata • SDK persistence with

base_state.json

and

events/

directory structure External Integration: • SkyRL is a third-party project that bridges OpenHands with VERL for RL training • Built specifically for long-horizon LLM agent training on environments like SWE-Bench • Documentation available at https://skyrl.readthedocs.io/ Community Activity: • GitHub issues show users asking about converting OpenHands trajectory data for RL training • Indicates community interest but confirms lack of official documentation This was a research and information-gathering task with no code changes made.