Preventing Tool Summarization in Child Agents & Passing Full Tool Response to Parent Agent

[dgnk007]

Hi everyone,

I have a scenario where I’m using a root agent and a SqlAgent (child).

• The SqlAgent contains secret information in its prompt that should never be exposed to the user.
• To protect this, I’m keeping the sensitive details inside the tool prompt (not in the conversation context).
• However, I also need the parent agent to receive the full tool response — without any summarization — so the parent has the complete context for further reasoning.

Here’s the challenge:

• If I use skip_summarization, it only prevents the parent from summarizing, but the tool itself still performs summarization before returning output to the parent.
• What I really need is for the tool to return the raw tool_response directly to the parent agent, bypassing any summarization stage entirely, while still not exposing the sensitive prompt to the end user.

Question:
Is there a supported way in ADK to configure a tool/child agent so that:

1. No summarization is performed at the tool level.
2. The exact tool output is passed up to the parent agent.
3. Sensitive prompts remain inaccessible to the user.

If not, could this be a potential feature for the ADK? This would be especially useful for secure agents like database query agents where accuracy and context preservation are critical.

Thanks in advance for any guidance or workarounds!