The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data by crawling external websites using the squirrel audit command. This content is then presented to the agent in an audit report. There is a potential for malicious instructions to be embedded in the website's content (e.g., meta tags, headings, or hidden text).
Ingestion points: The squirrel audit and squirrel report commands in SKILL.md ingest data from external URLs and local audit databases.
Boundary markers: The skill uses a structured hybrid XML/text format (--format llm) described in references/OUTPUT-FORMAT.md which provides clear delimiters for different sections of the audit data.
Capability inventory: The skill allows access to Bash, Read, Edit, Grep, and Glob tools for executing CLI commands and modifying project files.
Sanitization: The references/OUTPUT-FORMAT.md file specifies that all text content in the LLM format is XML-escaped for safety.
[EXTERNAL_DOWNLOADS]: The skill directs the user to download and install the squirrel CLI from the vendor's website (squirrelscan.com/download). It also includes self-management commands like squirrel self update and squirrel skills install which fetch and install software components. These are standard operations for the provided utility.
[COMMAND_EXECUTION]: The skill uses the Bash tool to interact with the squirrel binary. It instructs the agent to perform auditing tasks and apply automated fixes to the local codebase using Edit and Read tools. These actions are consistent with the skill's stated purpose of auditing and improving websites.

audit-website