The Agent Skills Directory

[SAFE]: The skill includes explicit instructions for frameworks like Next.js, Vite, and SvelteKit to implement guards (e.g., process.env.NODE_ENV === "production" or import.meta.env.DEV) that ensure the throwaway lab pages are unreachable in production environments.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided component names, file paths, and design briefs, which creates a potential surface for indirect prompt injection if the source files contain malicious instructions.
Ingestion points: Component names, file paths, and free-text design instructions used to generate code variants in the /lab/ directory.
Boundary markers: The skill uses a structured sequence of operations and hard rules to guide the agent's behavior, although no explicit data delimiters are defined for the processed component source code.
Capability inventory: The skill uses file system operations to read component source, write multiple variant files (variant-1.tsx, etc.), and modify routing configuration within the developer's local project.
Sanitization: The accompanying React utility code (useUrlParam) performs validation on URL-driven states by checking input against a list of allowed values (values.includes(v)).

design-variant