The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources which could contain malicious instructions designed to subvert the agent's behavior.
Ingestion points: The skill reads external content from live website URLs using WebFetch and Figma design files using MCP tools as specified in SKILL.md.
Boundary markers: The instructions do not provide explicit delimiters or "ignore previous instructions" directives when processing data from these external sources.
Capability inventory: The skill possesses file-writing capabilities, specifically creating docs/DESIGN.md and docs/DESIGN.html on the local file system.
Sanitization: There is no evidence of filtering or sanitization for text extracted from the external references before it is interpolated into the generated documents.
[DYNAMIC_EXECUTION]: The skill automatically generates executable client-side code (HTML and JavaScript) based on derived design tokens.
Evidence: SKILL.md instructs the agent to generate a docs/DESIGN.html file that includes a "vanilla-JS theme toggle" script.
Risk: While the intended script is simple and functional, the process involves generating code that will be executed in the user's browser, potentially incorporating content derived from untrusted external inputs.

studio-design-design-system