pencil-design-from-stitch-html

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

• Third-party content exposure detected (high risk: 0.85). The skill’s runtime path can ingest outsider-authored free text when it fetches Stitch HTML from htmlCode.downloadUrl (public/remote page content) and then parses innerText/content from that HTML to create Pencil text nodes (which become LLM-readable prose in the agent’s context during parsing/mapping).

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches the Stitch screen HTML at runtime from the Stitch download URL (htmlCode.downloadUrl, e.g. a URL under https://stitch.withgoogle.com/ or the screen's download URL like https://stitch.withgoogle.com/projects/123/screen/456 → htmlCode.downloadUrl) and then parses that remote HTML to drive batch_design operations, so the fetched content directly controls the agent's instructions.

MEDIUM W021: Hidden or invisible Unicode characters detected (potential obfuscation or prompt injection).

• Hidden Unicode characters detected (1 type(s) found)