The Agent Skills Directory

[COMMAND_EXECUTION]: Several Python scripts (improve_description.py, run_eval.py, run_loop.py) utilize the subprocess module to interact with the system's claude CLI. These calls use argument lists rather than shell strings, preventing shell injection vulnerabilities. The use of the CLI is necessary for the skill's functionality to evaluate and improve other skills.
[EXTERNAL_DOWNLOADS]: The evaluation viewer (eval-viewer/viewer.html) loads the SheetJS library from cdn.sheetjs.com. This is a well-known and trusted service used to render spreadsheet data in the browser for the user's review. Per the analysis rules, this reference is considered safe.
[INDIRECT_PROMPT_INJECTION]: As a tool designed to generate and optimize prompts based on user input, the skill inherently possesses an attack surface for indirect prompt injection. However, it employs boundary markers (e.g., XML-style tags) when processing user-provided content. Given its role as a developer-facing generation tool, this is a standard characteristic of its operation and is categorized as low risk.
[SAFE_PRACTICE]: The local web server started by the evaluation viewer binds specifically to 127.0.0.1, ensuring the review interface is only accessible from the local machine.

skill-creator