medium10 pts · first answer countsprompt-engineeringagentscontext

Tool output filling the context

Each call to a "search_orders" tool returns 30 KB of raw JSON. After a few calls, the agent's context is full and quality drops. What is a good fix?

A.Increase temperature so the model skims the JSONB.Ask the model to ignore tool results it finds longC.Send each tool result twice so it is rememberedD.Return only the fields the agent needs, in a compact form

Discussion 0

Join free to post in the discussion.

Tool output filling the context

Discussion 0

Related questions