[Question] Missing agent_selection.json for LLMPolicy (Puppeteer branch)

[StubbornKitten]

Great work on Multi-Agent Collaboration via Evolving Orchestration! Your puppeteer-style framework is really insightful for LLM-based multi-agent systems, and the open source code is super helpful.
When exploring the puppeteer branch, I found that the LLMPolicy in inference/policy/base_policy.py references the prompt file prompts/general/agent_selection.json, but this file is missing in the repo.
Could you kindly provide this missing file, or share the design ideas of this agent selection prompt?
Thanks a lot for your time and help!

[NA-Wen]

Thanks for pointing this out!

For the agent selection in LLMPolicy, the general idea is to let the LLM decide which agent to invoke based on the current execution context. Specifically, the prompt includes:

• The current task context (e.g., progress of the task or intermediate state)
• The full list of available agents
• The sequence of previously executed agents

Given this information, the LLM is asked to select the most appropriate next agent.

That said, this part is not a key method in our framework, and the prompt file was not included in the repository—sorry for the inconvenience caused by the missing file!