frontend-design
Pass
Audited by Gen Agent Trust Hub on Feb 16, 2026
Risk Level: LOWEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
- EXTERNAL_DOWNLOADS (LOW): The file
source.mdreferences a GitHub repository under theanthropicsorganization. This is a recognized trusted source per [TRUST-SCOPE-RULE], which downgrades the risk to LOW. - PROMPT_INJECTION (SAFE): The skill uses instructional markers such as 'CRITICAL', 'IMPORTANT', and 'NEVER'. These are identified as natural instructional language for guiding creative output and do not attempt to bypass agent safety protocols or override core instructions.
- INDIRECT_PROMPT_INJECTION (LOW): The skill identifies an ingestion point for untrusted data (user-provided frontend requirements). However, its capability tier is 'INFO/Display only' because the skill instructions focus solely on generating code for display/return to the user without requesting file system writes, command execution, or network operations. Evidence: (1) Ingestion in
SKILL.md; (2) No boundary markers; (3) No dangerous capabilities; (4) No sanitization.
Audit Metadata