Meta-Framework Proposal: Generic AI Trading Agent Interface (Input → Output Unification)

[samchenku]

Should we consider building a meta-framework inside TradingAgents that defines a generic interface for any AI trading agent — where:

Input = ticker (and optional context/data)

Output = classification or rating (e.g., Buy / Hold / Sell)

Basically, instead of hard-coding specific agent types, have a standard wrapper or interface that allows ML, LLM, or RL agents to plug in and produce comparable outputs.

Questions for discussion:

1. Do we want a unified input/output schema for all agent types?
2. Should the MetaAgent concept manage registration, evaluation, and orchestration?
3. How would we benchmark different agents under one standard interface?
4. Would this fit the long-term design of TradingAgents (as a platform, not just a pipeline)?

Just opening this to see if there’s interest or existing discussion around it.

Thanks!
Sam