The Agent Skills Directory

[COMMAND_EXECUTION]: Path Traversal Vulnerability in File Persistence.
Evidence: The persist_design_system function in scripts/design_system.py constructs directory and file paths using the project_name and page arguments provided via the command-line interface in scripts/search.py.
Details: The implementation only replaces spaces with hyphens but fails to sanitize or strip directory traversal sequences such as ... This allows a user or an automated process to write markdown files outside the intended design-system/ directory, potentially creating or overwriting files in arbitrary locations within the agent's accessible workspace.
[PROMPT_INJECTION]: Indirect Prompt Injection Attack Surface.
Evidence: The skill is designed to generate a "Design System Master File" (MASTER.md) and page-specific override files, which it then instructs the agent to read and prioritize as a source of truth during development tasks.
Details:
Ingestion points: The system ingests data from local CSV files (e.g., ui-reasoning.csv) and interpolates user-provided query strings directly into generated markdown instructions.
Boundary markers: The generated files do not include delimiters or instructions for the agent to ignore potentially embedded malicious commands.
Capability inventory: The skill possesses file-writing capabilities and explicitly directs the agent to follow rules stored in these files in future turns.
Sanitization: There is no sanitization of the content retrieved from CSVs or user queries before they are written to the persistent instruction files, creating a vulnerability where a compromised data source or a malicious user query could permanently alter the agent's behavior for a specific project.

ui-ux-pro-max