VideoToObsidian

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

• Third-party content exposure detected (high risk: 0.85). Alta probabilidad: el script lee el fichero de transcripción generado por TranscribeYoutube (transcript_content = transcript_path.read_text(...)) y lo incluye en el JSON (transcript_content) que luego se inserta en el contexto del LLM; esa transcripción proviene de subtítulos de YouTube (texto de terceros, no elegido por el usuario).

HIGH W008: Secret detected in skill content (API keys, tokens, passwords).

• Secret detected (high risk: 1.00). I found a high-entropy API key literal assembled in the script and used to build INNERTUBE_URL. The code sets:

_IK = os.environ.get("INNERTUBE_API_KEY") or "".join([ "AIzaS", "yAO_FJ2SlqU8Q4STEHLGCi", "lw_Y9_11qcW8" ])

which concatenates to: "AIzaSyAO_FJ2SlqU8Q4STEHLGCilw_Y9_11qcW8"

Reasons for flagging:

• The string is high-entropy and matches the common Google API key prefix ("AIza").
• It is a concrete, non-placeholder value (not "YOUR_API_KEY" or truncated).
• It is used directly to call the YouTube InnerTube endpoint (INNERTUBE_URL), so it provides access when present.
• Although the comment says "no es una clave personal" and there is an env var override, that does not change that the literal key is present and usable in the code.

No other high-entropy secrets (private keys, tokens, or passwords) are present. Other strings (paths, user-agent, simple placeholders) are non-sensitive by the criteria.