The Agent Skills Directory

[COMMAND_EXECUTION]: The persistence logic in scripts/design_system.py is vulnerable to path traversal. The persist_design_system function constructs directory and file paths using the project_name and page arguments directly. Because it only replaces spaces with dashes and does not sanitize for directory traversal sequences like ../, it allows the creation of directories and writing of files to arbitrary locations outside the intended design-system folder.
[PROMPT_INJECTION]: The skill implements a "Master + Overrides" design pattern where it generates markdown files containing design rules and then instructs the AI agent to "strictly follow" the rules found in these files. This creates a risk where manipulated files (facilitated by the path traversal flaw) could be used to override the agent's behavior or project instructions.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it ingests untrusted user input (the query and project/page names) and interpolates this data into markdown documentation without proper boundary markers or sanitization. This documentation is later used as a primary source of truth for the agent's code generation tasks.
[SAFE]: The skill uses local CSV databases for its BM25 search engine and does not perform any unauthorized network operations. All external URLs provided in the data files point to trusted domains such as fonts.google.com and official framework documentation.

ui-ux-pro-max