OpenClaw Skill Trust Score: How to Read It (and What It Doesn’t Guarantee)

Updated: 6 min read

UseClawPro shows a trust score on skill pages to help you compare skills quickly.

Think of it as a risk signal, not a guarantee.

What the trust score is for

The trust score is meant to answer one question fast:

“If I install this skill today, how likely is it to blow up my environment?”

It is most useful when you are:

  • comparing multiple skills in the same category
  • deciding whether to allow network or shell
  • choosing a default “safe” toolbox for your setup

What a trust score cannot do

A trust score cannot:

  • prove intent (“this is not malicious”)
  • prevent runtime prompt injection from external content
  • eliminate supply-chain risk from dependencies you install later

If a skill has shell or network, you still need process and controls.

How to use trust score with permissions (the right mental model)

Use these three layers together:

  1. Permissions (least privilege): /guides/permissions-explained
  2. Verification (pattern review): /verifier
  3. Isolation (sandboxing): /guides/sandbox-setup

High trust score + minimal permissions is the “easy yes.”

Low trust score + broad permissions is the “easy no.”

Practical thresholds (simple rules)

You can use a rough policy like:

  • No network + no shell: safe for daily use in most environments.
  • Network only: allow only if you can name the exact domains it must call.
  • Shell: sandbox required; treat as RCE.

Then use trust score to sort within each bucket.

Best next click