Skills
Navigate
skills/full-stack-skills/pencil-skills/pencil-design-from-stitch-html/Gen Agent Trust Hub

pencil-design-from-stitch-html

Pass

Audited by Gen Agent Trust Hub on Oct 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted data from external Stitch URLs and user-provided HTML to generate design commands.
  • Ingestion points: Input originates from Stitch project/screen URLs, the htmlCode.downloadUrl property, and manual HTML pastes provided by the user in SKILL.md.
  • Boundary markers: The instructions lack explicit delimitation or 'ignore embedded instructions' warnings for the agent when processing the external HTML content.
  • Capability inventory: The skill utilizes Write for file system operations, Bash for resource fetching, and mcp_pencil tools (batch_design) to manipulate design documents.
  • Sanitization: There is no evidence of sanitization or filtering logic intended to prevent embedded prompt instructions within the HTML from influencing the agent's behavior.
  • [EXTERNAL_DOWNLOADS]: The skill fetches design assets from Google's Stitch platform.
  • Evidence: The skill uses web_fetch or Bash to download HTML and screenshots from stitch.withgoogle.com (a well-known service) as part of the get_screen tool workflow described in SKILL.md and usage.md.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 7, 2026, 08:19 AM