The Agent Skills Directory

[COMMAND_EXECUTION]: The skill performs extensive command execution to manage the environment:
Runs claude mcp commands to register, list, and verify MCP servers.
Executes npm view and npm config to manage package versions and cache.
Invokes platform-specific package managers (apt-get, yum, brew, winget, scoop) to install system dependencies like ripgrep.
Executes language-specific tool installers (pip, dotnet tool, composer).
[EXTERNAL_DOWNLOADS]: The skill downloads and installs various packages from public registries and GitHub repositories:
Fetches vendor-specific MCP servers via npx (e.g., @levnikolaevich/hex-line-mcp).
Downloads system utility helpers like @vscode/ripgrep-postinstall and @sourcegraph/scip-python from NPM.
Installs language analysis tools from official repositories, including basedpyright (PyPI), csharp-ls (.NET), and scip-dotnet (.NET).
References vendor-maintained patches for tools on GitHub, such as scip-python and scip-php.
[COMMAND_INJECTION]: The skill demonstrates safe command construction, specifically using MSYS_NO_PATHCONV=1 on Windows systems to prevent shell-specific path conversion issues during MCP registration.
[INDIRECT_PROMPT_INJECTION]: The skill processes project-level files (package.json, .claude/commands/*.md, and .claude/skills/*/SKILL.md) to migrate allowed-tools configurations. While this is an ingestion surface for untrusted data, the skill uses structured parsing to limit risks and focuses only on specific YAML frontmatter fields.
[PRIVILEGE_ESCALATION]: The skill may request sudo access for package installation on Linux and provides a mechanism to modify the allowDangerouslySkipPermissions setting in IDEs (Cursor/VSCode). These actions are explicitly documented as detection-only by default and require both a specific invocation flag (apply_ide_override=true) and direct user confirmation.

ln-012-mcp-configurator