The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill generates prototype logic and UI based on user prompts and URL search parameters.
Ingestion points: User prompts (SKILL.md) and the variant URL parameter (UI.md) influence the generated output and runtime behavior.
Boundary markers: No explicit boundary markers or delimiters are instructed for processing these inputs.
Capability inventory: The skill generates and creates new files, including React components and standalone HTML files (SKILL.md, UI.md, LOGIC.md).
Sanitization: The UI switcher implementation uses a controlled set of variant keys, which mitigates arbitrary parameter injection.
[DYNAMIC_EXECUTION]: The skill is designed to generate executable code at runtime, including standalone HTML files with embedded JavaScript and React components, based on user requirements.
[COMMAND_EXECUTION]: The instructions direct the agent to execute project-level task runners (e.g., pnpm, python, bun) to launch the generated prototypes (SKILL.md).
[METADATA_POISONING]: The REVIEW.md file contains self-asserted safety claims and references to external security scanner scores (e.g., "Gen safe", "Socket safe"). These claims are treated as data to be evaluated and are not taken as authoritative for the final verdict.

prototype