The Agent Skills Directory

[COMMAND_EXECUTION]: Several Python scripts utilize the subprocess module to orchestrate the skill development lifecycle. This includes calling the claude CLI for evaluations, git for repository management, npm for dependency research, and system utilities like lsof and kill to manage the local evaluation viewer server.
[EXTERNAL_DOWNLOADS]: The skill's documentation and prerequisites guide the user to install well-known security tools (like gitleaks from GitHub) and software packages from official registries (PyPI, NPM). These references target trusted organizations and well-known services, posing no unusual risk to the environment.
[DATA_EXPOSURE]: The skill contains instructions for the agent to search for project-specific configuration files, such as ~/.claude.json and ~/.claude/CLAUDE.md, to research existing workflows. This behavior is contextually appropriate for a tool designed to integrate with the Claude Code platform.
[PROMPT_INJECTION]: The skill uses structured boundary markers (XML-style tags) when interpolating user-provided feedback or generated content into prompts. For example, in the description optimizer (improve_description.py), it wraps content in <skill_content> and <current_description> tags to mitigate accidental instruction following from the processed data.

skill-creator