The Agent Skills Directory

[SAFE]: No malicious patterns or security vulnerabilities were detected. The skill operates locally on the user's project files to provide design feedback.
[INDIRECT_PROMPT_INJECTION]: The auditing scripts (scripts/ux_audit.py and scripts/accessibility_checker.py) read and process untrusted content from the local project files. While this creates a potential surface for indirect prompt injection if a project file contains malicious instructions, it is a standard characteristic of code auditing tools.
Ingestion points: The scripts read .html, .jsx, .tsx, .vue, .svelte, and .css files from the provided project path.
Boundary markers: The audit output is presented to the agent as text or JSON without specific isolation markers.
Capability inventory: The agent is intended to use the audit reports to suggest or implement code improvements.
Sanitization: Content is processed via regular expressions for pattern matching, which does not execute the code but extracts strings for reporting.

frontend-design