watch

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly instructs the agent to AskUserQuestion for Whisper API keys and then write those keys into ~/.config/watch/.env (i.e., collect and embed user secrets), which requires the agent to handle secret values and—if implemented naively—to emit them verbatim in commands or outputs.

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

• Third-party content exposure detected (high risk: 0.75). SKILL.md describes downloading user-supplied URLs and using yt-dlp/transcription to produce readable transcript text and frame-derived metadata that are then combined into the assistant’s response context and written into report.md for ingestion; because that transcript originates from third-party caption sources / Whisper outputs, it counts as outsider-authored free text flowing into the LLM.