design-to-code
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data by analyzing images and browsing live websites. It provides robust instructions to mitigate risks.
- Ingestion points: User-supplied screenshots and URLs processed via visual analysis and browser interaction (SKILL.md, reference-forensics.md).
- Boundary markers: Present. Instructions explicitly state to use external sites only as visual/behavioral evidence and to ignore proprietary source code, scripts, or private data.
- Capability inventory: The skill writes frontend code to the repository and utilizes network tools to inspect public references.
- Sanitization: Present. The skill mandates using inert or mocked data for sensitive flows like payments or account recovery.
- [EXTERNAL_DOWNLOADS]: The skill references standard web development resources such as the Geist font family (Vercel) and the Fontsource package registry. These are well-known and trusted entities.
- [COMMAND_EXECUTION]: The instructions guide the agent to perform standard development tasks, including running project-specific linting, testing, and build scripts.
- [SAFE]: The skill prioritizes security by forbidding the creation of deceptive surfaces and emphasizing the use of project-native primitives and established libraries.
Audit Metadata