The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs network requests to external URLs to retrieve search results and web page content.
Evidence: search.js fetches data from https://search.brave.com/search. Both content.js and search.js (via the --content flag) can fetch content from any URL provided as a command-line argument.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted natural language content from external websites.
Ingestion points: Untrusted content is fetched in content.js and search.js using the fetch API.
Boundary markers: The skill provides delimiters in its output (e.g., --- Result 1 ---) in search.js to organize results, though these are not robust security boundaries.
Capability inventory: The skill has network access via fetch but does not possess filesystem modification or arbitrary command execution capabilities.
Sanitization: The skill uses the @mozilla/readability library and turndown in content.js and search.js to convert HTML to Markdown, which strips scripts but does not filter natural language instructions.

brave-search