The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The script scripts/fetch-from-od.mjs performs network operations to fetch design system configurations and template HTML files from an external repository located at github.com/nexu-io/open-design.
[INDIRECT_PROMPT_INJECTION]: The skill architecture utilizes data fetched from an external third-party source to generate the markdown prompts that dictate agent behavior, creating a vulnerability surface.
Ingestion points: Untrusted design data enters the system context through scripts/fetch-from-od.mjs and is used to populate aesthetics/**/prompt.md files.
Boundary markers: Absent. The design tokens extracted from external files are integrated into the final prompts used by the agent without using boundary markers to isolate untrusted content.
Capability inventory: The skill enables the agent to generate code (HTML/CSS) based on the provided tokens and instructions.
Sanitization: Absent. The extraction logic in scripts/build-prompts.mjs uses regular expressions to capture variable values without performing sanitization or validation of the content, which could allow for the injection of malicious instructions from the source repository.

sdesign-design