create-skill

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

• Third-party content exposure detected (medium risk: 0.30). In eval-viewer/generate_review.py, the server embeds transcript.md/eval_metadata.json-derived “prompt” and all outputs/ text files (including *.md, *.json, *.csv, etc.) directly into the generated HTML that users view and can paste feedback from, so outsider-authored text arriving in those workspace files is ingested as readable free text at runtime.

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

• Potentially malicious external URL detected (high risk: 0.70). The eval viewer includes a runtime-loaded third-party JS library (SheetJS) fetched from https://cdn.sheetjs.com/xlsx-0.20.3/package/dist/xlsx.full.min.js which will execute remote code in the user's browser when the viewer is opened.