Show HN: Guardians – Verify tool-using agent workflows before execution

Category: ai-ml

Tags: ai-safety, static-analysis, agent-framework, formal-verification

Score: 7.0/10 (Innovation: 7, Technical: 8, Documentation: 7, Utility: 6)

Guardians is a Python library for statically verifying AI agent workflows before execution, applying formal methods like taint analysis, security automata, and Z3 theorem proving to prevent prompt injection and other security violations. It's interesting because it translates established software security paradigms (like SQL injection prevention) to the emerging problem of securing AI agentic systems, enforcing separation of code and data at the planning stage.

Target audience: ai-engineers, ml-engineers, backend-devs

Repository: https://github.com/metareflection/guardians · Python · MIT · 6 stars

View on Hacker News