The Agent Skills Directory

[DATA_EXFILTRATION]: The skill performs network operations to external services.
Evidence:
search.js fetches search results from https://search.brave.com based on user queries.
content.js and search.js fetch arbitrary content from external URLs provided by the user or found in search results.
These operations are consistent with the skill's stated purpose of web search and content extraction.
[EXTERNAL_DOWNLOADS]: The skill relies on external npm packages and performs runtime data fetching.
Evidence:
Dependencies include @mozilla/readability, jsdom, turndown, and turndown-plugin-gfm.
The installation instructions in SKILL.md require running npm ci to download these dependencies.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to the processing of untrusted web content.
Ingestion points: Web page content is fetched in search.js (snippets and full page content) and content.js (full page content).
Boundary markers: The skill uses basic structural markers such as --- Result N ---, Title:, and Snippet: in search.js, which provide limited separation between control text and untrusted data.
Capability inventory: The skill has network access (fetch) and outputs content directly to the agent's console, which is then processed as context.
Sanitization: No specific sanitization or filtering is performed to identify or neutralize embedded instructions in the fetched HTML or converted Markdown; only basic structural cleaning is performed in htmlToMarkdown functions.

brave-search