ui-design
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONPERSISTENCEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: Downloads design-related skills from the Google Labs official GitHub repository and installs CLI tools from the public NPM registry.
- [COMMAND_EXECUTION]: Performs automated installation and configuration tasks using system binaries including npm, git, and agent harness CLIs.
- [PERSISTENCE]: Modifies the user's shell profile to include the local installation path in the system PATH variable, facilitating durable access to the design tools.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external UI design data and project state which may contain untrusted content.
- Ingestion points: Project data from Stitch remote services and document state from the pen.dev desktop application (SKILL.md, smoke.mjs).
- Boundary markers: Provides guidelines for preserving repository design ownership and using isolated artifact paths.
- Capability inventory: Includes file system modification for design implementation, network access for design synchronization, and harness configuration updates (setup.py, run.mjs).
- Sanitization: Enforces the use of authenticated MCP/SDK transports and warns against passing secrets as command-line arguments.
- [DYNAMIC_EXECUTION]: Creates a local shell launcher and employs Node.js module injection to manage proxy configurations at runtime.
- [CREDENTIALS_SAFE]: Manages service credentials in a private configuration file with restricted owner-only permissions (0600) and performs runtime ownership checks before loading secrets.
Audit Metadata