The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess module in several scripts (scripts/run_eval.py, scripts/run_loop.py, scripts/package_skill.py) to execute the claude CLI and other local utilities. This is necessary for its core functionality of running and testing skills in the local environment.
[EXTERNAL_DOWNLOADS]: The eval-viewer/viewer.html file loads the SheetJS (xlsx) library from cdn.sheetjs.com. This is a well-known service used to render Excel spreadsheets directly within the local review dashboard. Additionally, scripts/improve_description.py uses the official Anthropic Python SDK to communicate with the Anthropic API for generating optimized skill descriptions.
[PROMPT_INJECTION]: The SKILL.md instructions guide the model to write 'pushy' skill descriptions to ensure they trigger reliably. This is presented as a design pattern for skill discovery and optimization rather than an attempt to bypass safety filters or ignore system instructions.
[DATA_EXFILTRATION]: The eval-viewer/generate_review.py script starts a local HTTP server on localhost:3117 to host a developer dashboard. It reads files from the designated skill workspace to embed them into the review interface. This operation is confined to the local machine and the specified project directories.
[COMMAND_EXECUTION]: scripts/run_eval.py dynamically creates temporary command files in .claude/commands/ based on user input and then executes them via the claude CLI. This is a standard part of the skill testing and discovery evaluation process.

skill-creator