deepgram

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

• Third-party content exposure detected (high risk: 0.80). The skill's SKILL.md explicitly includes actions that fetch and ingest third-party content—e.g., the "Transcribe Audio from URL" action and the "membrane request CONNECTION_ID /path/to/endpoint" proxy—meaning the agent can retrieve arbitrary public URLs and then analyze/transcribe that untrusted content as part of its workflow, which could enable indirect prompt injection.