The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interacts with the local system and the Claude CLI using Python's subprocess module to perform its core engineering tasks.
scripts/run_eval.py and scripts/run_loop.py execute the claude command-line interface to evaluate how effectively skill descriptions trigger the agent.
eval-viewer/generate_review.py executes the lsof utility to manage local network ports when starting the evaluation viewer server.
The workflow instructions in SKILL.md prompt the execution of repository-specific scripts such as scripts/update_skill_index.sh and tests/common/compile_skill_scripts.py.
[EXTERNAL_DOWNLOADS]: The skill's HTML assets reference external resources from well-known technology providers to enhance the evaluation viewer's functionality.
assets/eval_review.html and eval-viewer/viewer.html load font styles from Google Fonts (fonts.googleapis.com).
eval-viewer/viewer.html fetches the SheetJS library from cdn.sheetjs.com to support the rendering of spreadsheet data within the viewer.
[REMOTE_CODE_EXECUTION]: The skill implements an automated optimization loop that relies on communication with an external AI service.
scripts/improve_description.py uses the Anthropic Python SDK to send current skill metadata to a remote model, which then generates improved versions of the description.
[DATA_EXFILTRATION]: As part of its intended optimization workflow, the skill transmits internal file content to a remote API.
scripts/improve_description.py sends the full text of SKILL.md files to the Anthropic API to facilitate description refinement. This is a functional requirement for the skill's primary purpose.

alicloud-skill-creator