The Agent Skills Directory

[COMMAND_EXECUTION]: Path traversal vulnerability in the design system persistence feature. The function persist_design_system in scripts/design_system.py uses a project_slug derived from user-provided input to construct directory and file paths. This string is not sanitized for directory traversal characters (such as ../), which could allow the skill to create or overwrite markdown files in arbitrary locations relative to the working directory if the agent is granted write permissions.
[PROMPT_INJECTION]: Indirect prompt injection risk. The skill ingests untrusted user input from the search query and interpolates it directly into generated design system files (MASTER.md and page-specific files) via the persistence feature. These files lack boundary markers or explicit instructions to ignore embedded content, creating a surface where malicious instructions could be persisted and subsequently obeyed by other agents reading the documentation.
Ingestion points: User-provided query in scripts/search.py passed to design_system.py.
Boundary markers: Absent in generated markdown files.
Capability inventory: Filesystem write access via persist_design_system in scripts/design_system.py.
Sanitization: Absent for project names and page titles.
[SAFE]: References to external resources like Google Fonts and the Lucide icon library are used for design recommendations and do not involve automated script execution or sensitive data exfiltration.

ui-ux-pro-max