Anyone can publish an AI agent skill. Not everyone can publish a verified one. This post explains exactly how the Agent Skill Exchange verification process works — what we check, what our tiers mean, and why it matters for the teams deploying these skills.

Why Verification Matters for Agent Skills

AI agent skills execute autonomously. A poorly scoped skill might send emails to the wrong recipients, provision cloud resources without limits, or expose credentials through insecure logging. The risks are real — and they scale with how often the skill runs. Verification exists to surface these risks before deployment, not after.

What We Actually Check

Every skill submitted to Agent Skill Exchange goes through a structured review covering: permission scope (does the skill request only what it needs?), failure behavior (does it fail safely and loudly?), documentation quality (can another agent — or engineer — understand what this skill does without running it?), output format consistency (does it produce predictable, parseable output?), and security posture (does it handle secrets safely, avoid logging sensitive data, and scope API access correctly?).

The Three Verification Tiers

Listed: The skill has been submitted and meets our basic quality bar — it has a SKILL.md, clear description, and no obvious red flags. It hasn’t been deeply reviewed yet.

Verified Metadata: A reviewer has confirmed the skill’s inputs, outputs, permissions, and documentation are accurate and complete. The skill does what it says it does.

Security Reviewed: The skill has passed a full security review — permission scope, data handling, secret management, and failure modes have all been explicitly assessed. This is our highest trust tier.

What Verification Doesn’t Cover

We’re transparent about limitations. Verification doesn’t guarantee a skill will work in every environment, doesn’t replace testing in your own infrastructure, and doesn’t assess performance at scale. It’s a trust signal, not a warranty.

How to Get Your Skill Verified

Submit your skill at agentskillexchange.com/create-skill/. Include a complete SKILL.md, real example inputs/outputs, and clear permission documentation. Skills with better documentation move through review faster.

Browse skills by verification tier at .