Instructions for AgentTool summarization

[sgpstefanov]

I'm reading the docs about wrapping an agent as a tool using AgentTool - https://google.github.io/adk-docs/tools/function-tools/#customization
I see that there's a skip_summarization option.
This is where I learn that some responses have automatic summarization.
Could you show me where the prompt for this summarization step is? I can't seem to find it in the code base.

[dexhunter]

There is no separate, fixed "summarization prompt" attached to AgentTool. Here, summarization means allowing the calling agent's normal LLM loop to continue after receiving the tool result. Its existing instructions, conversation history, and the tool response determine what it says next.

AgentTool(..., skip_summarization=True) sets tool_context.actions.skip_summarization. That flag makes the resulting event a final response, which stops the calling agent's follow-up model step.

So, to control how the caller presents the tool's answer, leave skip_summarization=False and put that guidance in the calling agent's instruction, for example: "After calling the research tool, answer the user's question using its findings and preserve any source URLs." To control the tool agent's own output, edit the wrapped agent's instructions instead.