The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install external dependencies including the @inferencesh/sdk package and the belt-sh/cli skill via npx. It also uses npx shadcn to fetch a component configuration from https://ui.inference.sh/r/agent.json. These resources originate from the vendor's official infrastructure.
[COMMAND_EXECUTION]: The documentation provides bash commands for project setup, including npm install, npx shadcn add, and npx skills add. These commands are standard for the integration of the React component.
[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion and rendering of agent-generated content and declarative UI widgets, which presents an attack surface for indirect prompt injection if the agent processes untrusted data.
Ingestion points: Untrusted data enters the context through agentConfig properties (description, system_prompt) and dynamically generated agent responses used to build widgets, as described in SKILL.md.
Boundary markers: No specific delimiters or "ignore embedded instructions" warnings are documented in the usage examples.
Capability inventory: The component supports browser-based tool execution via createScopedTools and network communication via the proxyUrl endpoint, as seen in the client-side tools example in SKILL.md.
Sanitization: There is no explicit mention of output sanitization or schema validation for the agent-generated JSON UI in the provided instructions.

agent-ui