Definition
An attack where a crafted software interface description (such as a Swagger/OpenAPI file) is fed into a tool that auto-generates AI plugin or agent code. Because the generator trusts the spec's contents, it can inject operating-system commands or malicious code paths into the code it produces, and can even pull attacker-controlled configuration into the plugin manifest. Developers routinely pull such specs from public registries, vendor exports, or pull requests.
Why it matters
This is a supply-chain-style foothold: one poisoned API document can hand an attacker code execution inside an AI coding agent, so any external API spec should be treated as untrusted input.