The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: A URL referenced in the premium patterns documentation (mivora-wbs.framer.website) has been flagged by automated security scanners as a confirmed phishing site. This domain is used as a source for design techniques within the skill.
[REMOTE_CODE_EXECUTION]: The design reference for the cursor-recode mockup contains the high-risk command pattern curl https://cursor.com/install -fsS | bash. While this is presented as part of a UI demonstration, the inclusion of piped remote script execution patterns is a significant security risk if processed or accidentally executed by an AI agent.
[INDIRECT_PROMPT_INJECTION]: The skill uses browser tools to ingest content from arbitrary external production websites to inform design decisions.
Ingestion points: Files references/design-inspiration.md and references/design-inspiration-urls.md instruct the agent to use mcp__fuse-browser__* to navigate to and capture content from external URLs.
Boundary markers: Absent. The skill does not provide explicit delimiters or warnings to the model to ignore instructions found within the browsed websites.
Capability inventory: The agent has the capability to write and edit files on the local filesystem and invoke external generation APIs via the Gemini Design MCP.
Sanitization: Absent. There are no mechanisms described to filter or sanitize text extracted from external sites before it is used in subsequent prompting phases.

design-web