HN Showcase
Back to browse
GitHub Repository

an implementation of the ideas in Erik Meijer's "Guardians of the Agents: Formal Verification of AI Workflows" (Communications of the ACM, January 2026)

67 starsPython

Guardians – Verify tool-using agent workflows before execution

by namin·Apr 9, 2026·8 points·0 comments
Visit ProjectView on HN

AI Analysis

●●●BangerWizardryBig BrainZero to One

Applies formal verification to prevent prompt injection before any tool executes.

Strengths
  • •Three independent verification checks catch issues LLMs miss: taint, automata, Z3 proving.
  • •Separates planning from execution — structured plans verified before any real tool calls.
  • •~1900 lines with 100 tests shows real implementation, not just a paper concept.
Weaknesses
  • •Requires workflows to be expressed as structured plans — limits ad-hoc agent flexibility.
  • •Z3 theorem proving adds computational overhead for complex workflow verification.
Category
Security
Target Audience

AI/ML engineers building agentic systems

Similar Projects

AI/ML●●Solid

OQP – A verification protocol for AI agents

OpenAPI for agent verification with risk scoring before merge — ambitious standardization play.

Bold BetBig Brain
Aamir21
822mo ago
Security●●Solid

SkillFortify, a formal verification for AI agent skills

Formal verification for agent skills when heuristic scanners always fail.

Big BrainNiche Gem
varunpratap369
223mo ago