Author here. We've been building agent trust infrastructure for ~1 year (12 microservices for identity, behavior monitoring, data governance, etc).
When the CSA published their Agentic Trust Framework spec, we realized our stack maps to all 5 elements — but the spec has no reference implementation. So we built one.
What we added that the spec leaves open:
- Maturity model runtime (agents earn autonomy through 5 promotion gates)
- Policy-as-code segmentation engine (real-time access decisions)
- Circuit breaker for incident response (auto-containment)
Everything is MIT. You can clone the repo and run `npm test` — 25 contract validation tests.
Repo: https://github.com/yogami/atf-reference-implementation
Interactive demo: https://berlinailabs.de/atf-demo.html
Happy to answer architecture questions.
Show HN: Open-source zero-trust framework for AI agents- 12 services, all tested | Heykuki News