skills

Static Analysis

Every skill repo is cloned in a sandbox. We parse manifests, scan scripts for network calls, filesystem access, and secret extraction patterns. No execution occurs.

Policy Enforcement

Skills must declare capabilities (fs, net, exec, secrets). The gate rejects undeclared or excessive permissions. Score penalizes broad scopes.

Evidence, Not Verdicts

Score any skill, repo, or agent answer 0–8 with file:line evidence, before you trust it. Every point points back at a file and a line.

Read-Only Registry

This site never executes target code. Read-only by design — the auditors never execute the target.