The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to download and install external components and tools. Specifically, it uses npx shadcn@latest add https://ui.inference.sh/r/widgets.json to fetch a remote component registry and npx skills add to install the belt-sh/cli and various inference-sh related components.
[INDIRECT_PROMPT_INJECTION]: The skill's primary purpose is to render "rich interactive UIs from structured agent responses." This creates an attack surface where a compromised or manipulated agent could generate malicious JSON structures.
Ingestion points: The WidgetRenderer component accepts a widget prop containing JSON data intended to be generated by an AI agent (as seen in SKILL.md).
Boundary markers: None are present; the skill lacks instructions for validating or delimiting the agent's JSON output before it is rendered.
Capability inventory: The widget system supports interactive elements including buttons with onClickAction, input fields, textareas, and selects, which can be used to capture user data or trigger application actions.
Sanitization: There is no mention of sanitization or schema validation to ensure the generated UI elements do not perform deceptive actions or exfiltrate data through form submissions.
[COMMAND_EXECUTION]: The skill relies on npx commands to install its dependencies and CLI tools, which involves executing shell commands provided in the documentation.

widgets-ui