extract

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). This skill explicitly sends user-supplied public URLs to Tavily's extraction API and returns the extracted page content (see SKILL.md examples and scripts/extract.py and scripts/extract.sh), so untrusted, user-generated web pages can be ingested and could materially influence downstream behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.80). The extract.sh script invokes "npx -y mcp-remote https://mcp.tavily.com/mcp" at runtime when no token is present, which fetches and executes remote code (the mcp-remote npm package) and relies on the external MCP endpoint https://mcp.tavily.com/mcp for the OAuth flow.