stella 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.