The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external HTML source code and project metadata retrieved via specialized design tools. While this creates a surface for indirect prompt injection if the processed project data contains malicious instructions, the risk is considered safe due to the use of trusted service sources and the specific scope of the analysis task.
Ingestion points: HTML source code and project metadata are fetched using the web_fetch tool from URLs provided by the Stitch MCP server as described in SKILL.md.
Boundary markers: The instructions do not define specific delimiters or instructions to isolate the fetched external code from the agent's core workflow.
Capability inventory: The skill is authorized to use web_fetch for data retrieval and Write to generate the documentation file.
Sanitization: No explicit sanitization of the fetched HTML content is performed before the agent extracts design tokens.

design-md