The Agent Skills Directory

[SAFE]: The skill operates as described, using legitimate design analysis workflows. All external references, such as the Stitch Effective Prompting Guide, point to official Google domains.
[PROMPT_INJECTION]: The skill processes untrusted external data (HTML and CSS) to extract design tokens, which constitutes an indirect prompt injection surface. This risk is inherent to the skill's purpose of code analysis and is mitigated by the specific task focus.
Ingestion points: HTML source code and screen metadata are fetched via the web_fetch tool using URLs provided by the Stitch MCP server in SKILL.md.
Boundary markers: Absent; the instructions do not specify the use of delimiters to isolate the fetched HTML content from the agent's instructions.
Capability inventory: The skill is allowed to use Write for local file creation and web_fetch for asset retrieval.
Sanitization: No explicit sanitization or filtering of the external HTML/CSS code is performed prior to the analysis and synthesis phase.

design-md