The Agent Skills Directory

[COMMAND_EXECUTION]: The skill automates complex workflows by executing the claude CLI and various helper Python scripts through subprocess calls. This includes running evaluations (run_eval.py), aggregating results (aggregate_benchmark.py), and launching a local review server (generate_review.py). These actions are integral to the skill's function as a development environment.
[EXTERNAL_DOWNLOADS]: The skill uses the anthropic Python client to interact with AI models for description optimization and utilizes PyYAML for parsing skill configurations. Additionally, the web-based review tool (viewer.html) loads the SheetJS library from a well-known CDN (cdn.sheetjs.com) to render spreadsheet outputs for the user. These dependencies are standard and sourced from reputable providers.
[PROMPT_INJECTION]: The skill processes content from draft skills and evaluation transcripts which are technically untrusted data. Malicious instructions within a draft skill's output could potentially influence the grader or analyzer subagents (Indirect Prompt Injection). The skill manages this by using specialized agent instructions (grader.md, analyzer.md) and human-in-the-loop review via the eval-viewer.

skill-creator