[Feature Request] Enable secure execution of tools with output masking for LLM

I have searched the Issue Tracker and Discussions that this hasn't already been reported. (+1 or comment there if it has.)
Consider asking first in a Discussion.

Internal tools would be executed and raw output (which can be sensitive) could be exposed to the LLM. This prevents security risks, privacy issues.

We can introduce a "Secure Execution Wrapper" or extend the existing external_tool interface paradigm. This wrapper would:

Be responsible for actually invoking the specified tool.
Capture the full output from the tool.
Internally manage or store this sensitive output (e.g., in a secure, session-specific context).
Return only a non-sensitive placeholder, handle, status code, or a sanitized summary to the LLM, instead of the raw tool output.

This allows the LLM to orchestrate the execution of tools without gaining access to their potentially sensitive results.

No response

No response

No response

Provide feedback