The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts run_eval.py, run_loop.py, and package_skill.py use the subprocess module to execute external binaries including the claude CLI, python, and lsof. These are used for running test cases and managing the local benchmark viewer server.
[REMOTE_CODE_EXECUTION]: scripts/run_eval.py executes arbitrary queries through the claude CLI. These queries are typically generated by the AI agent during the skill development or optimization phase. This creates a mechanism where instructions generated by an AI model are executed directly against the user's local environment via the CLI tool.
[DATA_EXFILTRATION]: The eval-viewer/generate_review.py script launches a local HTTP server using HTTPServer. It reads files from the designated workspace directory (including transcripts, grading results, and skill outputs) and embeds their full content as base64-encoded strings into an HTML page. This mechanism exposes local filesystem data to the browser environment.
[EXTERNAL_DOWNLOADS]: The eval-viewer/viewer.html template fetches the SheetJS library from cdn.sheetjs.com to enable spreadsheet rendering within the review interface. This is a well-known service for this library.

skill-creator