The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for creating interactive widgets that ingest data via props (e.g., query or initialCount) and perform actions using the useCallTool hook. This defines an attack surface where an agent could be influenced by malicious data passed into the widget's context.
Ingestion points: Widget component props and schema definitions in SKILL.md and references/design-principles.md.
Boundary markers: No specific delimiters or "ignore instructions" warnings are included in the provided design examples.
Capability inventory: The examples demonstrate network operations via fetch() and the ability to execute other agent tools via useCallTool.
Sanitization: The documentation recommends using React Error Boundaries and try/catch blocks for data parsing to prevent widget crashes.
[DYNAMIC_EXECUTION]: The skill demonstrates how to use template literals with the text/html+skybridge MIME type to inject HTML and external JavaScript into the widget environment.
Evidence: The pizza-map.ts example in references/design-principles.md shows a resource handler returning a script tag pointing to {{PIZZAZ_JS_URL}} and a link tag for {{PIZZAZ_CSS_URL}}.
Context: This is a documented feature of the xmcp platform for creating custom UI components and is used strictly in an instructional capacity within this skill.

widget-design