Git Exploit Runs Code Before Trust Prompt in 7 AI Agents, Leaving 4 Unpatched

A security flaw allows malicious Git repositories to execute attacker code before an AI assistant requests user permission. Four of the seven affected tools re…