The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Python subprocess module to execute the claude CLI and system utilities like lsof and kill. These are used for running skill evaluations, optimizing descriptions, and managing the local web server used for result reviews.
[EXTERNAL_DOWNLOADS]: The eval-viewer/viewer.html template includes a reference to a third-party script (xlsx.full.min.js) hosted on cdn.sheetjs.com. This is a well-known service used by the skill to render Excel file outputs in the evaluation viewer.
[DATA_EXFILTRATION]: The eval-viewer/generate_review.py script starts a local HTTP server on the loopback address (127.0.0.1) to serve evaluation outputs. This is the intended mechanism for human review of skill performance. Access is restricted to the local machine and the content served is limited to the skill's workspace directory.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) as it processes untrusted data including user-provided queries, transcripts of AI behavior, and the source code of other skills. To mitigate this, the skill utilizes boundary markers (e.g., XML-like tags such as <skill_content>) when interpolating this data into prompts for the grader and optimizer agents.

skill-creator