ui-kit-from-code

Pass

Audited by Gen Agent Trust Hub on Oct 14, 2026

Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes various system commands to perform discovery and synchronization tasks.
  • references/sync-kit.py invokes git and gh CLI tools via the subprocess module to manage repository state and push updates.
  • SKILL.md workflows involve running Node.js and Python scripts to automate the UI kit generation process.
  • [DYNAMIC_EXECUTION]: The skill extracts design tokens by dynamically generating and executing code within the project's environment.
  • Workflow W02 creates a temporary test script (lib/zz-kit-dump.test.ts), executes it using vitest to dump token values, and then removes the file.
  • [DATA_EXFILTRATION]: The skill is configured to upload generated UI kits to a remote repository.
  • The sync-kit.py script defaults to syncing kits to a private repository under the Rheinmir organization. While it verifies the repository is private, this involves transferring project-specific design data to a remote destination. This behavior can be disabled using the UI_KIT_SYNC environment variable.
  • [INDIRECT_PROMPT_INJECTION]: The skill analyzes external websites, which introduces a surface for indirect prompt injection.
  • Ingestion points: references/discover.mjs uses Playwright to visit external URLs and extract HTML and style information.
  • Capability inventory: The skill can write files, execute shell commands, and interact with external repositories (file: SKILL.md, references/sync-kit.py).
  • Boundary markers: The skill includes logic to respect robots.txt and provides mechanisms to skip specific patterns, but lacks explicit prompt delimiters to prevent the agent from following instructions embedded in scanned web content.
  • Sanitization: The skill does not perform robust sanitization of the HTML content extracted from external sites before it is processed by the agent.
  • [EXTERNAL_DOWNLOADS]: The skill retrieves data from external web origins during its discovery phase.
  • references/discover.mjs fetches robots.txt and navigates to external websites using page.goto() to gather UI component data.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 14, 2026, 02:25 AM