We Built a Claude Agent That Doesn't Know Its Own API Keys

listenlabs.ai

8 pointsoelmgren2 comments

I wonder how many applications have exactly the vulnerability… But I’m sure Anthropic is going to deprecate the SDK in 2 months and change it.

Yeah, this is exactly what I was thinking.

Maybe they’ll add the ability to sandbox the env variables from the agent’s bash execution via the Agent SDK settings at some point.