skill-creator

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

• Third-party content exposure detected (medium risk: 0.30). The eval viewer generator (eval-viewer/generate_review.py) recursively reads transcript.md/eval_metadata.json plus all text-like files under each run’s outputs/ and embeds their free text into HTML at runtime, so any outsider-authored text that can be written into the workspace’s outputs (e.g., via user-provided eval files or earlier runs) is ingested without selecting a specific trusted item.

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 remote JavaScript from https://cdn.sheetjs.com/xlsx-0.20.3/package/dist/xlsx.full.min.js at runtime to render .xlsx files in the browser, so the skill depends on executing remote code during its review workflow.