The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill features 'Audit' commands that read local design files (src/styles/tokens.css, src/styles/typography.css) and interpolate their raw content into LLM prompts.
Ingestion points: Files at src/styles/tokens.css and src/styles/typography.css are read via cat and passed to the gemini CLI.
Boundary markers: Absent. The file content is interpolated directly into the prompt string without delimiters or instructions for the AI to ignore embedded commands.
Capability inventory: The skill utilizes the gemini CLI for text generation and processing, and standard shell utilities (cat, jq) for file manipulation.
Sanitization: None detected. Content from external style files is used as-is.
Command Execution (SAFE): The skill contains bash snippets for exporting design tokens. These utilize standard, non-malicious tools like cat and jq to transform JSON data into CSS, JS, and SCSS formats.

design