web-design-engineer

Pass

Audited by Gen Agent Trust Hub on Oct 11, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: Instructions in SKILL.md and references/browser-acceptance.md suggest the use of external command-line tools like yt-dlp and ffmpeg for asset extraction, as well as starting local development servers (e.g., npm run dev) for browser-based quality assurance. These operations are intended to support the primary goal of creating high-fidelity web artifacts.
  • [EXTERNAL_DOWNLOADS]: The skill references several external resources from well-known services, including React and Babel via cdn.jsdelivr.net, D3.js via d3js.org, and Chart.js via cdn.jsdelivr.net. It also suggests fetching brand assets from official sites and video platforms to ensure visual accuracy.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as project requirements, brand guidelines, and user-provided codebases, and incorporates it into generated web artifacts. While the workflow includes multiple human-in-the-loop checkpoints, there are no specific instructions for sanitizing or escaping external content to prevent instructions embedded in that data from influencing the agent's behavior.
  • Ingestion points: Requirement gathering in SKILL.md (Step 1) and context gathering in SKILL.md (Step 2) and redesign-protocol.md.
  • Boundary markers: The skill uses structured workflow steps and checkpoints but lacks explicit prompt delimiters for external data.
  • Capability inventory: The skill has capabilities for file writing, shell command execution (CLI tools), and network requests (asset fetching).
  • Sanitization: No specific sanitization or filtering logic is prescribed for external inputs.
  • [DYNAMIC_EXECUTION]: The skill utilizes runtime JSX transpilation via Babel for React prototypes and incorporates dynamic interaction logic using requestAnimationFrame for animation engines. These techniques are standard for the intended front-end engineering use case.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 11, 2026, 08:21 AM