HN Showcase
Back to browse
GitHub Repository

Pre-execution policy engine for AI agents. Every tool call checked before execution.

7 starsTypeScript

Your AI agent logged the mistake. Mine wasn't allowed to make it

by Soenke_Cramme·Feb 25, 2026·2 points·2 comments
Visit ProjectView on HN

AI Analysis

●●●BangerSolve My ProblemBig Brain

Deny-before-allow policy engine stops prompt injection attacks on agent tool calls.

Strengths
  • •Fail-secure architecture: logic-based rules can't be bypassed by prompt injection, unlike LLM-gatekeeping
  • •Tamper-proof W3C Trace Context audit trail makes every decision verifiable and repeatable
  • •Framework-agnostic: works with LangChain, Vercel AI SDK, CrewAI, n8n without code changes
Weaknesses
  • •Only 3 GitHub stars and minimal adoption signals; unclear production usage or scaling proof
  • •Policy syntax and operator set (11 condition types) needs real-world usage examples beyond demo
Target Audience

Teams deploying AI agents with access to sensitive APIs and databases

Similar To

Boundary.io (HashiCorp) · OPA/Rego (policy-as-code)

Similar Projects

Developer Tools●Mid

MemLineage: governed writes for AI agents

PR-like review for agent writes, but OpenClaw adoption is niche and early.

Niche GemShip It
celastin
113mo ago