Malicious Git Configs Enable AI Coding Agents to Execute Attacker Code

Eight flaws across Claude, Codex, Cursor, and other AI agents allow malicious Git configs to trigger arbitrary code execution outside sandbox without approval prompt.