watch

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill tells the agent to AskUserQuestion for a Groq/OpenAI API key and then write that key into ~/.config/watch/.env (setting GROQ_API_KEY=... or OPENAI_API_KEY=...), which requires the LLM to receive and place secret values verbatim into a file/command—an exfiltration risk.

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

• Third-party content exposure detected (medium risk: 0.30). At runtime, the workflow accepts user-provided video URLs/paths, runs yt-dlp to fetch native caption VTT (outsider-authored text from the video host/platform) or falls back to Whisper transcription, then passes the resulting caption/transcript text into the LLM context via the report and subsequent “Read” framing.