The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation includes instructions to download and install components from external sources as part of its setup process. This includes fetching a component registry from https://ui.inference.sh/r/agent.json via the shadcn CLI and installing additional functionality from the belt-sh and inference-sh repositories using the skills CLI.
[INDIRECT_PROMPT_INJECTION]: The skill implements a Generative UI framework where agent responses dynamically create widgets and trigger client-side tools. This creates an attack surface for indirect prompt injection. 1. Ingestion points: The Agent component processes message streams from configured models (e.g., Claude via OpenRouter) to determine UI state. 2. Boundary markers: Not explicitly documented in the provided implementation examples. 3. Capability inventory: The skill enables client-side tool execution, such as scan_ui and fill_field, which can interact with browser DOM elements based on agent instructions. 4. Sanitization: Not explicitly shown in the snippets; the library relies on declarative JSON structures for widget rendering.

agent-ui