Trust center
Every skill in the registry carries a trust tier for how far it got: Scanned, Verified or Certified. This page states the rules behind each tier and lists what we blocked and rejected.
The checks describe reviewed content. They do not guarantee that a skill is safe or useful. Read the methodology
How a skill earns its tier
Each tier adds a check on top of the one before, and a minimum trust score. The skill page shows the result as a trust tier from 0 to 4.
ScannedTrust tier 2
Passed the 57-pattern security scan, and a trust score of 30 or more.
VerifiedTrust tier 3
A model review of intent and scope came back as a pass or with concerns noted, and a trust score of 60 or more.
CertifiedTrust tier 4
A confirmed source, a model review score of 80 or more, a completed human review and a trust score of 80 or more.
- Trust tier 1
- No scan result yet, or a trust score under 30.
- Blocked, trust tier 0
- On the blocklist. The trust score is 0.
vskill installrefuses it unless you pass--force. - Tainted
- Another skill by the same author is blocked. The skill can rise no higher than Scanned.
- Vendor organizations
- Skills from the organizations we list as vendors (Anthropic, OpenAI, Google, Google Gemini, Microsoft, Vercel, Cloudflare and the vskill maintainer's own) skip the scan and are certified on publication.
Loading trust distribution...
The checks behind each tier
Each check answers one question, in this order. Every skill page shows the result of each one, including the 57-pattern security scan.
- Source check
- We confirm that the person who submitted the skill owns or collaborates on its GitHub repository, check the skill and its author against the blocklist, and record a hash of the exact content we scanned, so a changed file is caught before it is published.
- Security scanEarns Scanned
- A static scan checks 57 patterns in 9 categories, including command injection, credential theft, data exfiltration and prompt injection. Each finding lowers a score that starts at 100, and 70 or above is a pass. Each critical finding takes off 25, each high 15, each medium 8 and each low 3. 70 or more passes, 40 to 69 is concerns, below 40 fails.
- Model reviewEarns Verified
- A model reads the skill for intent and scope and confirms or dismisses each critical finding from the scan. A pass, or a pass with concerns noted, lifts the skill to Verified. The model rates malicious intent, scope alignment, hidden behaviors, privilege misuse, capability expansion and prompt injection.
- Human reviewEarns Certified
- A person reviews the skill. With a confirmed source, a model review score of 80 or more and a trust score of 80 or more, it reaches Certified. Skills from vendor organizations skip the scan and both reviews and are certified on publication.
How the trust score adds up
Four parts, 100 points in all. Each tier needs a minimum score as well as its check, so a clean scan alone cannot make a skill Verified or Certified.
| Part | How it is scored | Points |
|---|---|---|
| Security scan | 30 for a pass, 15 for concerns, 0 for a fail | 30 |
| Model review | The review's 0 to 100 score, scaled to 30 | 30 |
| Provenance | 20 when the source repository checks out, minus 10 on a mismatch | 20 |
| Track record | 5 each: 90 days old, over 100 installs, no open reports, never blocklisted | 20 |
Published, blocked and rejected
Blocked skills are on the blocklist, and the CLI refuses to install them without --force. Rejected submissions failed the scan or the review and were never published. Submissions still being checked are in the queue.
Found a problem in a skill
Use Report a skill at the top of this page when a skill does something its description does not say. While a report is open, the skill loses 5 points of its trust score.