audio-to-srt

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

• Third-party content exposure detected (high risk: 0.70). Risk comes from the “Claude 逐段精修” step that reads 輸入檔.vocab.srt (derived from Groq STT JSON / resegmented SRT containing outsider-authored transcription text from the user-provided media), which then becomes the free-text input context to the agent’s LLM for cleanup.