Certification for agent artifacts
Certify the skills your AI agents run.
Provenance, policy and safety checks for every skill, MCP server and ruleset your agents load — signed, version-pinned, and gated by a private registry you control.
Built on AGENTS.md · MCP · A2A · SKILL.md — runs on your infrastructure.
The problem
Your agents already run code nobody reviewed.
Artifacts are executable
A skill or MCP server runs with the agent's full access — files, credentials, memory, network. The instructions are the code.
Registries don't vet
Publishing to a public catalog is a record, not a review. Listing says nothing about whether an artifact is safe to run.
Signing proves who, not safe
A signature confirms the publisher. It doesn't catch prompt injection buried in a skill — and that's already widespread.
How it works
Certify, gate, audit — on the formats you already use.
Certify
Each artifact is checked for provenance, policy conformance and known-injection patterns, then run against a reproducible test suite — yielding a signed certificate or an A–E trust band.
Gate
Your agents resolve against a private registry of approved, version-pinned artifacts. Nothing uncertified loads, and you revoke instantly.
Audit
An immutable trail of which artifact ran, which version, under which policy, and why it was allowed — ready for review.
Get involved
We're building this with a handful of teams first.
If your agents already load third-party skills or MCP servers, we want to learn from you — and give you certification before anyone else.
Request early access
Get an invite when the private beta opens. Work email, nothing else required.
You're on the list.
We'll email your invite when the beta opens.
Book a 20-minute briefing
Tell us what your agents run today. We'll show what we're seeing across other teams — no pitch.
Request received.
We'll reply with a scheduling link shortly.