The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions if documentation or requirements are sourced from untrusted parties.
Ingestion points: The skill reads from docs/standards/design-tokens.md, the components/ directory, and interprets unstructured user requirements to plan UI directions.
Boundary markers: There are no explicit delimiters or specific instructions to the agent to ignore embedded commands within the design documents or component metadata.
Capability inventory: The skill has broad capabilities including modifying application source code (pages/overview.js), updating design documentation (design.md), deleting files from the filesystem during cleanup, and executing local tests via node --test.
Sanitization: The skill does not describe any sanitization or validation logic for the input text before using it to generate executable HTML and JavaScript.
[DYNAMIC_EXECUTION]: The skill generates and executes/renders new JavaScript and HTML code based on user input and local data.
Evidence: It produces 'real HTML variants' with specific layout, type, and color spends. These are intended to be rendered and interacted with in a browser environment, representing a script generation and execution pattern.
[COMMAND_EXECUTION]: The skill utilizes system commands to perform development and cleanup tasks.
Evidence: It executes local test suites using the command node --test and performs file system deletions to remove variants and scaffolding after the design phase is completed.

draft-ui