Sandboxing and Other Subagent Security Measures?

Something I've neglected to pay attention to and became an issue recently when a sub agent launched by Claude used my email to register with an external service so that it could access information related to a task it was running. Not a big deal because I was researching security and the content was from OWASP, but I recognise how this could become a problem.

Anyone got some suggestions here? I mainly use the Claude code extension in VS code but I also use open models via Ollama in the VS Code Chat

6 comments

Sandboxing and Other Subagent Security Measures?