video-script

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

• Third-party content exposure detected (medium risk: 0.30). The skill ingests outsider-authored free text when ${var} is an arbitrary user-provided <url> (or topic) and the runtime “fetch[es] the live source now” in Step 1 to verify claims, meaning the LLM reads external page content authored by others.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches live sources at runtime (via "WebFetch / gh api" to read README and recent releases) and will inject external repository content from https://github.com into the model context to build the script, so remote content can directly control prompts.