Any plans to add sandbox execution environment?

Hi team,

I noticed that the concept of "sandbox" is already referenced in the prompt text within the codebase:

- `adk/middlewares/filesystem/prompt.go` describes the execute tool as "Executes a given command in the sandbox environment"
- `adk/prebuilt/deep/prompt.go` mentions "When sandboxing improves reliability (e.g. code execution, structured searches, data formatting)"

However, I didn't find any actual sandbox implementation in the codebase. Currently, when agents execute commands via the execute tool, there is no isolation
mechanism, which could pose security and reliability risks.

I'm curious — are there any plans to add sandbox functionality for tool/code execution? If so, is there a preferred direction (e.g., container-based,
process-level isolation, WASM, etc.)?

Thanks!

Start by reading adk/middlewares/filesystem/prompt.go and adk/prebuilt/deep/prompt.go, then trace the execute tool referenced there. Compare the current command-execution path with the isolation risks described in the issue. Done should be a documented sandbox direction and an agreed implementation scope, rather than a small isolated edit.

Written by the indexing model from the issue text.