visual-reference-calibration
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and interpret external visual references such as screenshots or links to external sites (CodePen, Dribbble). This creates a surface where malicious instructions or deceptive UI metaphors could be embedded in the provided data to influence the agent's code generation.
- Ingestion points: User-provided visual artifacts, screenshots, and URLs in
SKILL.md. - Boundary markers: The workflow mandates a 'Reference contract' and 'Calibration questions' to explicitly confirm intent with the user, serving as a procedural safeguard against unintended instruction following.
- Capability inventory: The skill involves file-writing, creating preview branches, and building prototypes as described in the implementation guardrails.
- Sanitization: No technical sanitization of artifacts is described; the skill relies on manual verification and user confirmation steps.
- [EXTERNAL_DOWNLOADS]: The skill references an external visual example from CodePen (
codepen.io), which is a well-known community service for front-end developers.
Audit Metadata