The Agent Skills Directory

Indirect Prompt Injection Surface: The skill defines mechanisms for generating UI components from user-provided prompts and application state. This architectural pattern involves processing potentially untrusted data to influence structural output.
Ingestion points: Data enters the agent context via the buildUserPrompt and experimental_composeSpec functions as described in SKILL.md.
Boundary markers: The documentation references the use of system prompts and custom rules to guide AI generation, providing a layer of control over the output.
Capability inventory: The skill focus is limited to generating JSON specifications for UI rendering and does not involve high-risk capabilities like direct file system access or arbitrary command execution.
Sanitization: The library includes validateSpec and autoFixSpec functions designed to verify and repair the structural validity of the generated content.
Dynamic Prop Expressions: The skill supports the use of dynamic expressions like $computed and $template which are resolved at render time. These allow for interactive UI behavior by calling registered functions, which is a common and expected pattern in data-driven frameworks.
Secure Secret Management: The documentation correctly demonstrates the use of environment variables (process.env.AI_GATEWAY_API_KEY) for managing sensitive credentials, which aligns with standard security best practices.

core