The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses Python's subprocess module in several scripts (scripts/run_eval.py, scripts/aggregate_benchmark.py, eval-viewer/generate_review.py) to execute local CLI commands. These include running claude -p for trigger evaluation and lsof to manage ports for the local review server. This behavior is necessary for the skill's primary purpose as a development and testing tool.
[EXTERNAL_DOWNLOADS]: The evaluation viewer (eval-viewer/viewer.html) loads external resources including the SheetJS library from cdn.sheetjs.com and fonts from Google Fonts. SheetJS is a well-known service used for rendering spreadsheets within the UI, and its usage here is transparent and functional.
[DATA_EXFILTRATION]: The skill communicates with the Anthropic API via the anthropic Python SDK in scripts/improve_description.py and scripts/run_loop.py. This is the intended mechanism for the 'Description Optimization' feature, sending skill metadata and test results to the LLM to generate improved descriptions.
[PROMPT_INJECTION]: The skill processes untrusted user input (skill content) and interpolates it into prompts for the description optimizer in scripts/improve_description.py. While this presents an indirect prompt injection surface, the skill employs boundary markers (XML-like tags) to delimit content, and the risk is contained within the local development context.

skill-creator