The Agent Skills Directory

External Resource Integration: The skill fetches documentation and library scripts from the project's official domain (ui-sdk.dev). This practice is routine for web libraries providing CDN-based distribution and hosted documentation.
UI-to-Agent Interface Surface: The skill facilitates a bridge between UI components and AI agents via the createAgent API. This creates a surface where the agent interacts with application data.
Ingestion points: The agent observes the interface state and control labels through the agent.snapshot() function, primarily within SKILL.md and references/agent-ready.md.
Boundary markers: The skill incorporates specific attributes like data-agent-hidden and data-agent-secret to explicitly prevent the agent from seeing or reading sensitive regions, as detailed in references/agent-ready.md.
Capability inventory: The agent.act() method enables the agent to perform actions such as setValue or press on designated UI controls.
Sanitization and Control: To mitigate risks, the architecture includes a mandatory allow callback. This is designed to implement human-in-the-loop verification for any action marked with data-agent-confirm.
Dynamic Tool Registration: The skill utilizes registerTools to expose capabilities to browser-based agents via the WebMCP protocol. This is an intended integration pattern for the library's agent-ready functionality.

ui-sdk