skill-creator

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

• Third-party content exposure detected (medium risk: 0.30). In skill-creator, the runtime workflow reads the user’s outsider-authored eval prompt text and embeds it for the reviewer via eval-viewer/generate_review.py → build_run() (from eval_metadata.json/transcript.md) → generate_html() (embedded into HTML), so attackers can inject free text through the test prompts that are displayed to the LLM later during review/feedback.

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

• Potentially malicious external URL detected (high risk: 0.70). The eval viewer (eval-viewer/viewer.html) loads and executes a third-party JavaScript library at runtime from https://cdn.sheetjs.com/xlsx-0.20.3/package/dist/xlsx.full.min.js to render .xlsx files in-browser, which is fetched during skill runtime and runs remote code the viewer depends on.