AI-Coded App Vulnerability Checklist Introduced with 33 LLM-Specific Items

The post introduces an open-source checklist of 258 vulnerabilities in applications built with AI coding assistants, organized into 17 categories. Among these, Category 6 includes 33 vulnerabilities specific to LLM integration, such as MCP tool poisoning, agent memory poisoning, cross-agent prompt injection, and insecure agent handoff, each with detection methods (static analysis, runtime inspection, or config audits). The checklist provides severity ratings and is accompanied by a prompt.md file to run all checks against a codebase using an LLM CLI, returning file paths, line numbers, and remediation steps. The project is licensed under Apache 2.0.