The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/run_eval.py uses subprocess.Popen to invoke the claude CLI tool. This is the intended mechanism for running evaluations within the Claude Code environment. Additionally, eval-viewer/generate_review.py uses subprocess.run to call lsof for local port management, which is a standard utility operation.
[EXTERNAL_DOWNLOADS]: The skill utilizes the anthropic Python library to communicate with the Anthropic API for description optimization tasks. This involves standard API calls to a trusted service provider and is essential for the skill's optimization logic.
[DATA_EXFILTRATION]: No evidence of malicious data exfiltration was found. The skill processes local project data, evaluation results, and user feedback, sending only relevant queries to the official Anthropic API for processing.
[PROMPT_INJECTION]: The instructions in SKILL.md are correctly focused on the workflow for skill development. They do not contain any patterns typical of prompt injection, such as attempts to bypass safety filters or disregard system instructions.
[DYNAMIC_EXECUTION]: The evaluation viewer (eval-viewer/generate_review.py) dynamically constructs HTML reports by embedding local assets (such as logs and images) using Base64 encoding. This is a legitimate implementation for generating self-contained, offline-compatible developer reports.
[CREDENTIALS_UNSAFE]: The code follows best practices for secret management. It does not contain hardcoded API keys or tokens; it relies on standard environment variables (e.g., ANTHROPIC_API_KEY) for authentication with external services.

skill-creator