« All posts

Background Subagents Can Leak Secrets

Claude Code's background subagents emit system prompt fragments, including authorization data. A security issue to be aware of!

Developers reported an issue with Claude Code's background subagents stalling and leaking authorization data in system prompt fragments. This is labeled as a security concern and should be taken seriously.

The solution is structural: credentials should be scoped per subagent, subagent outputs treated as untrusted, and the system prompt separated from the working context. These measures will help prevent privacy breaches.