The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/design_system.py uses a user-supplied or agent-determined project_name parameter to construct filesystem paths for writing 'Master' design files. The logic design_system_dir = base_dir / "design-system" / project_slug is vulnerable to path traversal and absolute path injection. An attacker or a manipulated agent could provide a project name like ../../../.bashrc or an absolute path to write design rules into sensitive system or configuration files, potentially achieving persistence or influencing shell sessions.
[PROMPT_INJECTION]: The skill processes unstructured and structured data from multiple internal CSV files (e.g., styles.csv, ux-guidelines.csv, ui-reasoning.csv). This data is interpolated directly into prompts and generated markdown files (MASTER.md) without sanitization or clear boundary markers. This creates an 'Indirect Prompt Injection' surface where malicious instructions embedded in the data files could override agent behavior during a design task.
[PROMPT_INJECTION]: The persistence pattern generates a MASTER.md file containing explicit instructions for the AI agent to follow its rules 'strictly'. This creates a self-referential instruction loop where data generated by the tool is later treated as authoritative system-level guidance, amplifying the risk of any data poisoning in the source CSVs.
[SAFE]: The skill includes standard instructions for users to install Python 3 using official package managers (Homebrew, APT, Winget) which is appropriate for its execution environment.

ui-ux-pro-max