The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill instructs users to install components using npx shadcn@latest add https://ui.inference.sh/r/widgets.json. The domain ui.inference.sh is not in the trusted sources list, posing a risk of untrusted code execution during the component setup phase.
[REMOTE_CODE_EXECUTION] (HIGH): The documentation suggests adding related skills via npx skills add inferencesh/skills@.... Since inferencesh is not a verified organization in the trusted list, this pattern involves executing potentially malicious remote packages.
[INDIRECT_PROMPT_INJECTION] (HIGH): The core functionality is rendering UIs from agent-generated JSON (Generative UI).
Ingestion points: The WidgetRenderer component accepts a widget object, which is expected to be generated by an AI agent.
Boundary markers: None identified in the schema or usage examples to prevent the agent from being coerced into generating malicious UI structures.
Capability inventory: Supports forms, buttons with actions, inputs, and external images, allowing for complex interactive workflows.
Sanitization: There is no evidence of input sanitization or output encoding to prevent the injection of deceptive UI elements (e.g., fake login forms or malicious redirects).
Risk: An attacker could influence the agent's context to generate a UI that phishes for user credentials or misleads the user into clicking malicious buttons.

widgets-ui