The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted web content from search results and external websites, creating a surface for indirect prompt injection attacks.
Ingestion points: Search results (titles and snippets) in search.js and full page content fetched and parsed in both search.js and content.js.
Boundary markers: Output is delimited using text headers like --- Result 1 ---, but the skill does not provide explicit instructions to the agent to disregard commands embedded within the retrieved data.
Capability inventory: The skill's scripts are limited to performing network requests via fetch. There are no capabilities for executing shell commands, writing to the file system, or accessing sensitive environment credentials.
Sanitization: The skill uses @mozilla/readability and jsdom to parse and clean HTML content, removing dangerous elements such as <script> and <style> tags before converting the results to markdown format.

brave-search