The Agent Skills Directory

[SAFE]: The skill consists entirely of instructional markdown and metadata. No executable scripts or remote dependencies are present, and the overall design promotes the creation of isolated, throwaway code.\n- [DATA_EXFILTRATION]: The skill instructs developers to avoid using production databases for prototypes, favoring in-memory state or isolated scratch files, which prevents accidental exposure of sensitive production data.\n- [COMMAND_EXECUTION]: Mentions of task runners (pnpm, python, bun) are examples of how users can run their generated prototypes; these are not commands executed by the agent or skill itself.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes user prompts and existing code (ingestion points) to generate prototypes. It lacks explicit input sanitization or boundary markers but mitigates risk by instructing the agent to mark all output as throwaway code and isolate it from main production logic. Capabilities are limited to file creation and providing user instructions.

prototype