plaud

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

• Third-party content exposure detected (medium risk: 0.30). Plaud’s required runtime workflow ingests LLM-readable outsider-authored free text indirectly via Plaud recordings when the workflow uses list_files to select recordings and then calls get_note/get_transcript to retrieve the (user/attacker-authored) transcript and AI notes.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill explicitly instructs callers to fetch live "get_instructions"/schema from the AgentPMT MCP at runtime and "treat returned live schema and instructions as more specific," meaning content fetched from https://api.agentpmt.com/mcp/ can directly control agent prompts and behavior.