aether

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

• Third-party content exposure detected (high risk: 1.00). The skill explicitly ingests public, user-generated live chat from YouTube/Twitch/Bilibili (see references/chat-platforms.md and the SKILL.md workflow), normalizes and feeds those messages into the LLM→TTS→avatar pipeline (UnifiedChatMessage, command handling, and routing), so untrusted third-party content can directly influence agent actions like LLM responses, TTS playback, and avatar behavior.