kai-hook-bench

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

• Third-party content exposure detected (medium risk: 0.30). The workflow executes a python audit collection script (scripts.audit.collect --url <business-url> --mode sales_external) and ingests external Reddit digests, user-provided reviews, and web searches, allowing outsider-authored free text into the hook generation pipeline.

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

• Potentially malicious external URL detected (high risk: 0.70). The skill executes a Python script with a user-supplied business URL parameter (--url <business-url>), which constitutes an unverified external runtime dependency.