The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the web_fetch tool to download project assets such as HTML source code and screenshots from dynamically generated URLs provided by the Stitch MCP server. These downloads are necessary for the skill to perform its intended design analysis.
[DATA_EXPOSURE]: The agent is instructed to read project metadata, including project IDs and design themes, using the Stitch MCP server tools. This information is used to populate the generated DESIGN.md documentation and does not involve exfiltration to unauthorized third parties.
[PROMPT_INJECTION]: Analysis of the instructions found no attempts to bypass safety filters, extract system prompts, or override agent constraints. The logic is strictly focused on design system synthesis.
[INDIRECT_PROMPT_INJECTION]: The skill processes external HTML code which represents a surface for indirect prompt injection.
Ingestion points: HTML content is fetched via web_fetch and read_url_content in SKILL.md.
Boundary markers: The skill lacks explicit delimiters or instructions to ignore embedded prompts within the retrieved code.
Capability inventory: The skill has permissions to write files to the project directory and perform further network fetches.
Sanitization: There is no mention of sanitizing or escaping the retrieved HTML content before it is processed by the agent.

design-md