Claude Code: Global setup command - installs plugins, commands, agents

setup-claude-code.md

Setup Claude Code

You are setting up Claude Code globally on this machine. This captures best practices from agentic coding workflows.

Step 0: Check Existing Setup

First, check what's already configured:

ls ~/.claude/commands/ 2>/dev/null
ls ~/.claude/agents/ 2>/dev/null
cat ~/.claude/settings.json 2>/dev/null

If things are already set up, ask the user:

• "I see you already have [X] configured. Would you like me to add missing pieces, replace everything, or skip setup?"

Step 1: Choose Setup Path

Ask the user:

Option A: Instant Setup (Recommended)

• Installs everything with best-practice defaults
• Skips multi-model features that require external CLI auth (Gemini, Codex)
• Gets you productive immediately

Option B: Custom Setup

• Interview for each feature
• Configure multi-model ensembling if you have Gemini/Codex CLIs
• More control over what gets installed

Step 2: Interview (Custom Path Only)

Skip this step for Instant Setup.

Core Commands (Recommended)

• Atomic commits (/commit-smart) - For multi-agent workflows where you only commit your changes. Want this?
• Context transfer (/transfer-context) - Prepare context for new chat when degraded. Want this?
• Debugging (/debug) - Systematic debugging when stuck. Want this?

Multi-Model Ensembling (Optional - requires external CLIs)

• Do you want multi-model ensembling capabilities? (get opinions from Claude, Gemini, Codex)
  • If yes: Do you have Gemini CLI installed? (which gemini)
  • If yes: Do you have Codex CLI installed? (which codex)
  • Commands created based on what's available

Plugins (Optional)

• code-simplifier - Simplifies complex code at end of sessions. Want this?
• frontend-design - Helps with design decisions and component structure. Want this?

Step 3: Configure Global Permissions

Update ~/.claude/settings.json with safe defaults. These apply globally to all projects.

{
  "permissions": {
    "ask": [
      "Bash(rm -rf:*)",
      "Bash(rm -fr:*)",
      "Bash(sudo:*)",
      "Bash(chmod 777:*)",
      "Bash(git push --force:*)",
      "Bash(git push -f:*)",
      "Bash(git reset --hard:*)",
      "Bash(git clean -fd:*)",
      "Bash(npm publish:*)",
      "Bash(eval:*)",
      "Read(.env)",
      "Read(.env.*)",
      "Read(**/.env)",
      "Read(**/.env.*)",
      "Read(~/.ssh/**)",
      "Read(~/.aws/**)",
      "Read(**/*credential*)",
      "Read(**/*secret*)",
      "Read(**/*.pem)",
      "Read(**/*.key)"
    ]
  }
}

This makes Claude ask before:

• Destructive operations (rm -rf, git reset --hard, git push --force)
• Publishing packages
• Reading sensitive files (.env, credentials, keys)

Merge with existing settings—don't overwrite other config.

Step 4: Install Plugins

Instant path: Install both Custom path: Only install what user confirmed

# Code simplifier - cleans up complex code
claude plugins install code-simplifier

# Frontend design - helps with design decisions
claude plugins install frontend-design

Step 5: Create Global Commands

Create files in ~/.claude/commands/. Fetch content from gists:

Instant Path - Install All Core Commands:

Command	Gist
commit-smart.md	https://gist.github.com/SilenNaihin/d4b3870178667475b08e1f48d6cdbc30
transfer-context.md	https://gist.github.com/SilenNaihin/e4be0e8750343d9cbafdaab88366115c
debug.md	https://gist.github.com/SilenNaihin/6833c01f597c82912af5aca4e3467a35
refactor.md	https://gist.github.com/SilenNaihin/cd321a0ada16963867ad8984f44922cf
update-claudemd.md	https://gist.github.com/SilenNaihin/916e4f2cbcd96f487b2845edaed469a3
setup-repo.md	https://gist.github.com/SilenNaihin/e402188c89aab94de61df3da1c10d6ca

Custom Path - Multi-Model Commands (if requested and available):

Command	Gist	Requires
ensemble-opinion.md	https://gist.github.com/SilenNaihin/3e9b43522b61e155bd256fe7193493cd	Gemini or Codex
codex-delegate.md	https://gist.github.com/SilenNaihin/ff19b2d65d17137b0ee1f609f25205c5	Codex CLI
gemini-delegate.md	https://gist.github.com/SilenNaihin/0c27ee7816da489c4ce110bf115f5ab6	Gemini CLI

Fetch each gist and write to ~/.claude/commands/[name].md.

Step 6: Create Agents

If multi-model ensembling requested (Custom path only), create ~/.claude/agents/claude_critic.md:

---
name: claude_critic
description: "Provides critical analysis - identifies risks, blind spots, and alternative approaches without bias"
---

You are Claude Critic, one voice in an ensemble of AI models analyzing a problem.

Your role is to provide independent critical analysis. Be objective and evidence-based.

## Guidelines

- Present facts and observations, not opinions dressed as facts
- Identify what you know vs what you're assuming
- Be contrarian when useful - challenge assumptions
- Focus on failure modes and edge cases others might miss
- No cheerleading or validation - be honest about weaknesses

## Output Format

### Observations
- [Factual observations about the problem/approach]

### Risks & Failure Modes
- [What could go wrong, with reasoning]

### Blind Spots
- [What the user might be missing]

### Alternative Approaches
- [Other ways to solve this, with tradeoffs]

### Assessment
[Honest 2-3 sentence assessment - what's the real situation here?]

Step 7: Verification

Run checks:

ls ~/.claude/commands/
ls ~/.claude/agents/
cat ~/.claude/settings.json

Confirm everything was created.

Step 8: Report to User

Instant path report:

✓ Global permissions configured (asks before destructive ops)
✓ Plugins installed: code-simplifier, frontend-design
✓ Commands installed: /commit-smart, /transfer-context, /debug, /refactor, /update-claudemd, /setup-repo

Next: Run /setup-repo in any project to configure repo-specific settings.

Custom path report:

Installed:
- Permissions: [configured/skipped]
- Plugins: [list what was installed]
- Commands: [list what was created]
- Agents: [list if created]

How to use:
- /commit-smart - atomic commits for multi-agent work
- /transfer-context - when context degrades
- /debug - systematic debugging when stuck
- /refactor - code quality cleanup with jscpd, knip, code-simplifier
- /update-claudemd - update CLAUDE.md with recent learnings
- /ensemble-opinion - multi-model perspectives (if installed)
- /setup-repo - set up a new project with best practices

Next: Run /setup-repo in any project to configure repo-specific settings.