The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands to install system-level software such as LibreOffice, Pandoc, Poppler, and QPDF using platform-specific package managers (choco, brew, apt, dnf).
[COMMAND_EXECUTION]: On Linux systems, the skill explicitly uses sudo for package installation, which is a privilege escalation vector. However, the skill includes a safety constraint (MUST NOT Rule #1) requiring user confirmation before execution.
[EXTERNAL_DOWNLOADS]: The setup process triggers multiple downloads of external dependencies from official registries, including npm (docx, pptxgenjs, etc.) and PyPI (pandas, lxml, etc.).
[COMMAND_EXECUTION]: The skill modifies configuration files (~/.claude/CLAUDE.md and ./CLAUDE.md) to inject routing tables, which persists the plugin's behavior across future agent sessions.
[COMMAND_EXECUTION]: The skill performs environment verification by executing version-check commands for all installed binary tools and performing package imports in Python and Node.js to confirm successful setup.
[PROMPT_INJECTION]: The skill uses a local configuration file gateway/install.yaml to determine which packages to install. This presents a surface for indirect influence if a user or external process can modify this file's content before the setup script runs.

setup