[Feature Request] Add WFGY multi-agent failure map (ProblemMap No.13) to LightAgent docs

[onestardao]

Hi, thank you for publishing LightAgent — the framework is very helpful for people trying to implement real multi-agent systems without a lot of boilerplate.

I maintain WFGY ProblemMap (MIT), a 16-mode failure taxonomy for RAG + agents:

https://github.com/onestardao/WFGY/blob/main/ProblemMap/README.md

Besides “classic” RAG issues, it has a dedicated slot:

• No.13 – multi-agent chaos {OBS}
  • Sub-pages on role drift, cross-agent memory overwrite, and black-box interactions.

This ProblemMap has already been integrated or referenced by:

• Harvard MIMS Lab – ToolUniverse (LLM tools benchmark; WFGY used for robustness / RAG debugging).
• QCRI LLM Lab – Multimodal-RAG-Survey.
• Univ. of Innsbruck Data Science Group – Rankify.

From my experience, LightAgent-style setups hit a very specific cluster of problems:

• Agents gradually overwrite or corrupt each other’s memory.
• Roles drift over time (planner becomes worker, critic starts acting like another planner, etc.).
• Logs are hard to interpret when something goes wrong (“black box chaos”).

Proposal

I would like to contribute a multi-agent troubleshooting section for LightAgent:

1. A short doc (e.g. docs/troubleshooting/multi_agent_problemmap.md) that:

   • Summarizes ProblemMap No.13 and its sub-modes (role drift, memory overwrite, hidden feedback loops).
   • Shows how they manifest inside LightAgent (chat history growth, tool calls, shared vs. local memory, etc.).
   • Gives concrete logging / instrumentation tips to make these issues visible.

2. A small “debugging checklist” table for LightAgent users, mapping:

   • Symptom → ProblemMap mode → what to inspect in LightAgent.
   • For example:
     • “Agent A suddenly starts talking like Agent B” → multi-agent chaos / role drift → check prompt templates + naming + routing.
     • “All agents forget instructions after a few turns” → memory overwrite → inspect global memory append rules.

3. Optionally, a minimal example script where I intentionally create a multi-agent failure, then show how to diagnose it using the ProblemMap lens.

Everything would live in docs only, no breaking changes to the framework.
If this is interesting, I can open a PR with a first draft so you can see the style and decide how deep you want it integrated.

[wxai-space]

Could you provide me with more information? I'm very interested in your ideas. Or you can simply submit the PR so that I can review it. Thank you very much.

[onestardao]

Ok I will submit PR tmr , thanks ^^