google-meet

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

• Third-party content exposure detected (high risk: 0.85). The “Meeting Review Workflow” reads meeting transcripts via list_transcript_entries (and related transcript/recording retrieval), where the transcript text is authored by meeting participants (outsider speakers) and is ingested into the agent’s LLM context as readable text fields at runtime.

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

• Potentially malicious external URL detected (high risk: 0.90). This skill explicitly instructs agents to call the AgentPMT MCP server to fetch live schema/instructions at runtime (e.g., via tools/call to https://api.agentpmt.com/mcp/ for actions like get_schema/get_instructions), which would directly control prompts/instructions returned to the agent.