plaud

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

• Third-party content exposure detected (medium risk: 0.30). In the MCP HTTP variant (https://mcp.plaud.ai/mcp) used by browser-based AI clients (Claude Web/ChatGPT Web), the runtime tool calls can ingest meeting/transcript/note text that originates from recordings created by the user’s Plaud accounts, i.e., outsider-authored content can flow into the LLM conversation via Plaud’s returned source_list/note_list fields.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill config/installation uses the external MCP endpoint https://mcp.plaud.ai/mcp at runtime (for Claude Web / ChatGPT Web and as the remote MCP server), which lets a remote service supply/drive tool responses and model context (i.e., it can inject instructions into the agent), and the skill relies on that endpoint as a required runtime dependency.