SkillGuard#

Full security audit of a SKILL.md — three checks, one call.

Start the server#

bash

uvicorn skillguard.app:app --port 8005

Audit a skill#

bash

curl -s -X POST http://localhost:8005/v1/audit-skill \
  -H "Content-Type: application/json" \
  -d "{\"skill_content\": $(cat path/to/SKILL.md | jq -Rs)}" | jq

Returns verdict (SAFE/CAUTION/DANGEROUS), total_findings, and three sub-reports:

• scan — malware detection (safety_score, findings, verdict)
• scope — permission analysis (declared vs detected, undeclared_access)
• injection — prompt injection (risk_score, patterns_detected)

Why use SkillGuard instead of individual tools?#

One call instead of three. Same price. Combined verdict logic: if the malware scan finds anything, it's DANGEROUS. If only scope or injection issues, it's CAUTION. Clean skill = SAFE.