security-scan

name	security-scan
description	Run SkillGuard before installing third-party Agent Skills, with optional semantic review only when the user explicitly allows network/model use.

SkillGuard Security Scan

Use this skill before installing or enabling a third-party Agent Skill.

Default Static Scan

Run a deterministic local scan first:

skillguard scan <skill-path> --preset strict --policy strict

Do not install the skill if the scan fails unless the user explicitly accepts the risk after reviewing the findings.

Semantic Review

Run semantic review only when the user explicitly approves sending the candidate skill content to a model provider:

skillguard scan <skill-path> \
  --preset paranoid \
  --policy strict \
  --allow-network \
  --semantic \
  --semantic-provider openai

Semantic review requires provider credentials, normally OPENAI_API_KEY. For OpenAI-compatible local endpoints, use:

skillguard scan <skill-path> \
  --preset paranoid \
  --policy strict \
  --allow-network \
  --semantic \
  --semantic-provider openai-compatible \
  --semantic-base-url http://localhost:11434/v1

Never include unrelated user documents, chat history, workspace secrets, or credentials in the scan target.