Claude Code Router

Bundled commands use $PLUGIN_ROOT ($env:PLUGIN_ROOT in PowerShell; same path suffix) for the plugin root. Set it once: use the host's plugin-root variable when defined (Claude Code: PLUGIN_ROOT="$CLAUDE_PLUGIN_ROOT"), otherwise the absolute path of this plugin's root directory.

Use this skill when the user asks to operate, diagnose, automate, wrap, or explain Claude Code CLI, including claude, claude --print, claude plugin, claude mcp, claude agents, claude doctor, claude auto-mode, worktrees, hooks, settings, permissions, custom agents, or ultrareview.

For the full command/safety matrix, read $PLUGIN_ROOT/references/cli-operation-contracts.md.

First Move

Prefer live local facts over memory. Inspect the installed CLI before building commands or debugging version-sensitive behavior:

python3 "$PLUGIN_ROOT/scripts/claude_code_inspector.py" --json

If the user supplies a Claude executable path, use it for the current run:

python3 "$PLUGIN_ROOT/scripts/claude_code_inspector.py" --claude "$CLAUDE_CLI_PATH" --json

Do not commit personal absolute paths into source files, manifests, docs, or examples. Use CLAUDE_CLI, PATH, ANTHROPIC_API_KEY, or user-provided runtime arguments instead.

Routing

• Non-interactive --print, JSON/stream-json, JSON schema, input streaming, prompt files/stdin, budget caps, fallback model, and no-persistence runs: use claude-print-automation.
• Plugin marketplaces, plugin install/update/remove/details/validate/tag/prune, session-only --plugin-dir or --plugin-url, and MCP server lifecycle: use claude-plugin-mcp-manager.
• Broken config, safe mode, bare mode, debug logs, doctor, update/install, auth/token setup, auto-mode classifier, IDE/Chrome startup issues: use claude-doctor-debugger.
• Background agents, claude agents --json, worktrees, tmux, resume/continue/from-pr, fork-session, session names, remote control, and ultrareview: use claude-agent-worktrees.
• Settings JSON, hooks, CLAUDE.md, tool allow/deny rules, custom agents, plugin customizations, and setting-source boundaries: use claude-hooks-settings.

If several apply, inspect the CLI first, diagnose the safety/config surface second, then run or recommend the narrow workflow.

Safety Rules

• Treat --dangerously-skip-permissions, --allow-dangerously-skip-permissions, and --permission-mode bypassPermissions as high-risk. Use them only with an explicit external sandbox boundary.
• Prefer default permission flow or --permission-mode plan for exploratory work.
• Use --safe-mode for broken customizations and --bare for minimal explicit-context troubleshooting.
• Do not run doctor, project purge, setup-token, auth, install/update, cloud ultrareview, plugin marketplace updates/removals, MCP removals/resets, or long-lived background sessions without clear user intent.
• Keep secrets out of command examples, settings snippets, MCP headers, debug files, hook logs, and final answers.

Completion Standard

A Claude Code CLI task is done when the answer includes the exact command or source change, the permission/tool/settings mode chosen, the cwd assumptions, and proof appropriate to the request: inspector output, subcommand help, plugin validate, plugin/MCP list/details output, safe-mode/bare comparison, settings JSON validation, or a clearly reported blocker.