The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local commands such as ls and Read to analyze project structures, UI directories, and configuration files like package.json or tailwind.config. These operations are limited to structural analysis for design token extraction.
[EXTERNAL_DOWNLOADS]: The skill uses WebFetch to retrieve data from user-provided URLs. This is intended for analyzing live websites to extract color palettes, typography, and component patterns. References to services like Apple, Stripe, and Linear are used as templates or inspiration sources.
[DATA_EXFILTRATION]: The skill processes local brand assets, code repositories, and documents. While it does not perform unauthorized exfiltration, its primary workflow involves aggregating this data into a DESIGN.md file for the user to upload to an external platform (claude.ai/design). The skill explicitly instructs users to anonymize sensitive information (customer data or revenue) before processing.
[PROMPT_INJECTION]: The skill is subject to Indirect Prompt Injection (Category 8) because it ingests untrusted content from external websites and third-party code repositories.
Ingestion points: External URLs via WebFetch, GitHub repository content, and local files (PPTX, PDF, Source Code).
Boundary markers: The instructions do not specify explicit delimiters or "ignore" instructions for the processed external data.
Capability inventory: The skill has the ability to list directories (ls), read files (Read), and write/save files to the local filesystem (./design-system-prep/).
Sanitization: There is no explicit mention of sanitizing or escaping the content retrieved from websites or repositories before it is processed by the AI.

claude-design-system-prep