The Agent Skills Directory

[COMMAND_EXECUTION]: Path traversal vulnerability in filesystem persistence logic. The function persist_design_system in scripts/design_system.py constructs file paths for writing markdown documents using the project_name and page parameters derived from user-controlled input. The logic performs minimal sanitization (lowercasing and replacing spaces with dashes), which does not prevent directory traversal sequences such as ... This could allow an attacker to write files to arbitrary locations on the host system beyond the intended design-system/ directory.- [PROMPT_INJECTION]: Indirect prompt injection surface identified through untrusted data ingestion. The skill loads and parses structured data from data/ui-reasoning.csv, including a Decision_Rules field containing JSON-formatted instructions that are used to guide the agent's reasoning. This content lacks boundary markers and sanitization, allowing for the possibility of influencing agent behavior via malicious data payloads. Evidence Chain: 1. Ingestion point: data/ui-reasoning.csv; 2. Boundary markers: Absent; 3. Capability inventory: File system writing (scripts/design_system.py); 4. Sanitization: Absent.

ui-ux-pro