The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It retrieves and processes content from external websites using the WebFetch tool. If a target website contains malicious instructions (e.g., hidden in HTML comments or metadata), the agent might follow those instructions during the audit process.
Ingestion points: Website homepage, sitemaps, and internal page content fetched via WebFetch in SKILL.md.
Boundary markers: None identified; the instructions do not specify delimiters or sanitization for the fetched content to prevent it from being interpreted as instructions.
Capability inventory: WebFetch (network read), Write (local write), Bash (shell), Read, Grep, and Glob tools are allowed.
Sanitization: None identified for external web content.
[EXTERNAL_DOWNLOADS]: The skill utilizes the WebFetch tool to retrieve content from arbitrary URLs for the purpose of the audit. While this is the intended functionality, it involves the ingestion of untrusted remote data.
[COMMAND_EXECUTION]: The skill configuration explicitly allows the Bash tool. While no specific malicious shell scripts are hardcoded, the availability of a shell environment alongside the processing of untrusted web data increases the risk profile in the event of a successful prompt injection.

geo-audit