How can I use Agent Lightning for fine-tuning an agent’s system prompt using OpenAI models (gpt-4o)? Can VERL be used for this?

[rutvik-jaiswal-deeplearning]

I am trying to fine-tune an existing Agent Lightning agent—specifically its system prompt (agent behavior).
My constraints:

I need to use OpenAI models only (e.g., gpt-4o base).

I want to apply VERL for updating or optimizing the agent's behavior/prompt.

Preferably a simple working example (not split into many files).

My questions:

Does Agent Lightning support fine-tuning or behavioral optimization using VERL when using OpenAI LLMs?

Is VERL compatible with Agent Lightning for updating prompts or reward-based agent behavior?

Can you provide a validated minimal example showing how to use VERL to optimize an Agent Lightning agent using an OpenAI model?

Thank you!