The only AI agent skills
you can trust.
Every skill passes a 2-stage automated security pipeline — pattern-based static analysis and LLM semantic review — before it appears in the registry.
$ npx skillvault install frontend-design
Fetching frontend-design from registry...
publisher community/frontend-design
version 1.3.0
── stage 1 · pattern scan ──────────────
✓ prompt injection CLEAN
✓ network calls NONE
✓ credential access NONE
✓ risk score 2 / 100 VERY LOW
── stage 2 · LLM semantic analysis ─────
✓ intent alignment ALIGNED
✓ scope expansion NONE
✓ recommendation APPROVE
Installing to ~/.claude/skills/frontend-design/
✓ Done. Reload Claude Code to activate.
Live scanner — running 24/7
View all flagged skills →0
scanned
0%
elevated risk
0
flagged
0
verified
Connecting to live scanner...
The 2-stage automated security pipeline
Every skill passes both stages. No exceptions.
Regex and heuristic detection across 25+ threat categories: prompt injection, credential path access, shell command execution, network call analysis, social engineering scoring, IoC cross-reference.
Claude reviews the skill's intent and instructions for natural-language threats: sleeper instructions, scope expansion, social engineering, intent misalignment, and subtle manipulation patterns that evade static analysis.
2
Review stages
Pattern scan + LLM semantic analysis
✓
LLM + Pattern Scanned
Automated 2-stage security pipeline
✓
2-Stage Automated Pipeline
Fast, consistent, no human bottleneck
Know what you're installing before you install it.
Free security scan for any public skill. No account required. Full verification with signing available for review submissions.