message-test-designer

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

• Third-party content exposure detected (medium risk: 0.30). In SKILL.md, the runtime reads user-provided/candidate message variants and panel notes (“User-provided”) as untrusted input while designing a message-test spec (step 1/5/8), so an outsider can submit free text that the agent will ingest before selecting specific canon items.