lark-minutes

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

• Third-party content exposure detected (high risk: 0.75). Required runtime inputs include user-provided free text fields like materialText (and potentially materialFile/materialUrl content) which are inserted into the skill’s context via the contextComposer form schema; if that text originates from outsiders (e.g., pasted meeting/chat content by others), it becomes LLM-readable prompt content.