The Agent Skills Directory

[SAFE]: The skill implements a robust workflow for skill development, including evaluation cycles and description optimization. All scripts (aggregate_benchmark.py, generate_report.py, package_skill.py, etc.) perform legitimate development tasks such as data processing, file packaging, and local reporting.
[COMMAND_EXECUTION]: The skill uses subprocess.Popen and os.kill in eval-viewer/generate_review.py and scripts/run_eval.py to manage a local development server and execute claude -p for testing. These are standard operations for a local development tool and do not involve untrusted input execution.
[EXTERNAL_DOWNLOADS]: The eval-viewer/viewer.html file includes references to Google Fonts and a well-known CDN (SheetJS on cdn.sheetjs.com). These are used for UI rendering and spreadsheet processing in the local review tool, which is standard and safe.
[DATA_EXFILTRATION]: No evidence of unauthorized data transfer was found. Network operations are limited to a local HTTP server (127.0.0.1) used for the evaluation viewer and legitimate API calls to Anthropic for the description optimization feature.

skill-creator