How I Built a Static Analyzer for Prompt Injection

▤ Summary

This article introduces agentic-guard, a static analyzer for LLM agent code that spots prompt-injection risks before runtime. It focuses on two detection rules: confused-deputy flows between untrusted source tools and privileged sink tools, and dynamic system prompts built from runtime data. The tool supports LangGraph and the OpenAI Agents SDK, uses a framework-agnostic IR, and ships with a YAML-based taxonomy for tool classification. The author also reports validation across popular open-source agent codebases and outlines roadmap items such as deeper library-call analysis and support for more frameworks.

▦ Linked Companies