TrustFall: malicious repos trigger code execution in Claude Code, Cursor, Gemini, and Copilot CLIs
Researchers disclosed 'TrustFall', a class of conventions in agentic coding CLIs (Claude Code, Cursor CLI, Gemini CLI, GitHub Copilot CLI) where opening a malicious repository triggers code execution with minimal or no user interaction. The vendors' warning dialogs underdescribe what's being authorized. Reproducible technique published.
Treat agentic CLIs the same as you'd treat 'npm install from a stranger': run them in a disposable container or VM, never against live cloud credentials. Update internal dev guidance before someone clones a customer repo on a laptop with prod AWS keys cached.