The Agent Skills Directory

[Remote Code Execution] (CRITICAL): The skill utilizes a dangerous installation method detected by automated scans: curl -fsSL https://squirrelscan.com/install | bash. Piping a remote script from an untrusted domain into a shell allows for arbitrary command execution on the host machine.
[Indirect Prompt Injection] (HIGH): The skill creates a significant attack surface by parsing content from external websites and feeding it to an AI agent.
Ingestion points: Web content such as meta descriptions, titles, and site information from audited URLs (references/OUTPUT-FORMAT.md).
Boundary markers: The XML-based llm format provides structural separation but lacks explicit instructions for the agent to ignore instructions embedded within the text fields.
Capability inventory: The tool is designed to be used in a pipeline with agents (e.g., squirrel audit ... | claude), meaning injected instructions could trigger downstream agent actions.
Sanitization: The tool performs XML escaping, which prevents XML syntax errors but is ineffective against natural language prompt injection.
[External Downloads] (MEDIUM): The skill relies on installation from an unverified third-party domain (squirrelscan.com) that is not listed as a Trusted External Source.

audit-website