The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to download and install UI components and SDKs from the Inference.sh domain. Specifically, it uses npx shadcn to add a component from https://ui.inference.sh/r/agent.json and installs the @inferencesh/sdk package.
[COMMAND_EXECUTION]: Provides standard command-line instructions for environment setup and dependency management, including npm install, npx shadcn, and npx skills add for platform-specific CLI tools and related skill blocks.
[INDIRECT_PROMPT_INJECTION]: The Agent component renders LLM outputs and executes client-side tools based on agent responses, which presents an inherent surface for indirect prompt injection if the agent processes untrusted external data.
Ingestion points: The component processes agent responses (agentConfig) and renders generative UI widgets based on those responses.
Boundary markers: No specific delimiters or "ignore instructions" markers are defined in the provided configuration examples.
Capability inventory: The component supports client-side tool execution (e.g., scan_ui, fill_field), form filling via refs, and declarative JSON widget rendering.
Sanitization: Sanitization is expected to be handled by the internal component logic of the Agent UI block, though no explicit sanitization patterns are exposed in the skill metadata.

agent-ui