superdesign
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill installs the
@superdesign/clipackage globally via npm and fetches remote instruction sets (INIT.md, SUPERDESIGN.md) from the vendor's GitHub repository at runtime. - [COMMAND_EXECUTION]: The agent executes shell commands using the
superdesignCLI to create projects, iterate designs, and manage authentication via thelogincommand. - [DATA_EXFILTRATION]: The skill reads the full implementation of UI components, layouts, and configurations, sending this code context to the external SuperDesign service for design processing.
- [PROMPT_INJECTION]: The skill dynamically fetches markdown-based instructions from external URLs and instructs the agent to follow them without manual review, which can modify agent behavior at runtime.
- [PROMPT_INJECTION]: The skill processes untrusted repository files to build its operational context, creating a surface for indirect prompt injection. Ingestion points: Full source code of components, layouts, and project configuration files. Boundary markers: None present in the instructions. Capability inventory: Shell command execution via the CLI and writing files to the .superdesign/init/ directory. Sanitization: No sanitization of the analyzed source code is mentioned or performed.
Audit Metadata