cognitive-toolkit

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

• Third-party content exposure detected (high risk: 0.85). At runtime, the Hermes/Telegram adapter waits for and then ingests the operating user’s Telegram replies (free-form text) into the agent’s LLM context; this is outsider-authored content because it comes from the Telegram user, not from the skill author or the operating user’s own prompts.