Manifold Security: Seven AI Coding Agents Execute Git-Config Commands Before Trust Prompts

Manifold Security has disclosed GitSpawn, a class of flaws in which seven command-line AI coding agents execute a command named in a repository's own Git configuration before a workspace-trust prompt — and, on some products, before the user authenticates or types a message. Researchers reported eight findings; four were still unpatched when the work was published on September 1, 2026.