short-form-eval

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

• Third-party content exposure detected (high risk: 1.00). The skill's Pre-Dispatch and Dispatch Protocol explicitly require fetching and appending the (a public short-form post URL such as TikTok/Reels/Shorts/X video) into agent prompts and agents (hook-strength, eval-runner, pattern-extractor) read and act on that fetched post content to score and drive decisions, so it ingests untrusted public/social-media content that can influence tool behavior.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires fetching and appending the runtime into agent prompts ("Append brief, post-URL fetched content..."), so arbitrary remote post content would be injected into model context and can directly control agent behavior — flagged URL: .