level-up

# Level Up

The 10 Levels of Claude Code. Scan your environment, determine where you actually are, get a focused roadmap to the next level, and optionally build the first step right now.

This is NOT the generic "10 Levels of Claude" (browser, desktop, projects). This is specifically about Claude Code mastery. Level 0 is "just installed it." Level 10 is genuinely rare.

**Usage:**
- `/level-up` - Full flow: assess, roadmap, build
- `/level-up --assess` - Just show your current level
- `/level-up --build` - Skip assessment, jump to building

---

## How It Works

```
Scan Environment → Determine Level → Show Roadmap → Build First Step
       ↓                 ↓                ↓               ↓
   Phase 1           Phase 1          Phase 2          Phase 3
   (automatic)       (+ 3 questions)  (next level)     (optional)
```

Designed to be run repeatedly. Every time you level up, the assessment updates.

---

## The 10 Levels of Claude Code

| Level | Name | What It Means (Plain English) |
|-------|------|-------------------------------|
| 0 | Terminal Tourist | You just installed it. You're typing prompts like it's ChatGPT but in a terminal. No setup, no config. That's fine, everyone starts here. |
| 1 | Grounded | You've created a CLAUDE.md file so Claude actually knows who you are and what you want. You know the basic commands. This alone makes a huge difference. |
| 2 | Connected | Claude can read your actual tools: your Slack, your Drive, your Notion, your Gmail. You're not copy-pasting context anymore. It just pulls what it needs. |
| 3 | Skilled | You've built reusable commands (skills) you run regularly. Instead of re-explaining the same task every session, you type `/research` or `/review` and it just works. |
| 4 | Context Architect | You've built a structured knowledge system: memory files, patterns, client profiles, examples. Claude doesn't just follow instructions, it draws from everything you've taught it. Output gets better over time. |
| 5 | System Builder | Your skills chain together. One feeds into the next. You use sub-agents for parallel work. You have approval gates so nothing ships without your sign-off. This is where Claude feels like a team, not a tool. |
| 6 | Pipeline Engineer | You're calling Claude from scripts. Headless mode, JSON output, piping data through it programmatically. Claude is a component in your automation stack, not just a chat window. |
| 7 | Browser Commander | Claude controls a browser. It scrapes websites, takes screenshots, generates PDFs, builds carousels. Research pipelines that crawl the web and turn findings into deliverables. |
| 8 | Multi-Agent Operator | Multiple Claude instances running simultaneously. Each one is a specialist. You coordinate them like a team lead. They share work through files and git. |
| 9 | Always On | Claude runs on a schedule whether you're at your desk or not. Cron jobs, background agents, automated monitoring. It's infrastructure now, not a tool you open. |
| 10 | Swarm Architect | Agents that manage other agents. Autonomous execution loops. Give it a goal and it works toward it, spawning sub-agents as needed. Genuinely rare. Maybe a handful of people on the planet. |

Most users land between Level 1 and 4. Level 5 is strong. Level 7+ is rare.

---

## Phase 1: Assess Your Level

### Step 1.1: Environment Scan

Silently scan the user's environment. Do NOT ask permission for each check. Just read what's available.

**Check all of these:**

```
1. CLAUDE.md (check ALL locations)
   - Check: CLAUDE.md, .claude/CLAUDE.md, CLAUDE.local.md, ~/.claude/CLAUDE.md
   - Does any exist? How many locations are used?
   - How many lines total? (< 10 = minimal, 10-50 = basic, 50-150 = solid, 150+ = advanced)
   - Does it reference memory files, patterns, workflows?
   - Does it have navigation tables, progressive disclosure, workflow instructions?

2. Skills / Commands
   - Check both .claude/commands/ AND .claude/skills/ directories
   - Count total skills
   - Read 2-3 to check complexity:
     * Simple (< 20 lines, single prompt) = Level 3
     * Medium (20-80 lines, multi-step) = Level 3-4
     * Complex (80+ lines, phases, approval gates, tool calls) = Level 5+
     * Skills that call other skills or orchestrate sub-agents = Level 5-6

3. MCP Configuration
   - Check .mcp.json in current directory (project-level)
   - Check ~/.claude.json for mcpServers (user-level)
   - Optionally check managed config if enterprise
   - Count configured MCP servers across all locations
   - Categorize: data (Notion, Drive), comms (Slack, Gmail), dev (GitHub, Supabase), browser (Puppeteer, Chrome), specialty (PostHog, GSC)

4. Memory / Context Structure
   - Is there a memory/ directory?
   - How deep? (flat files = Level 4, nested dirs like memory/patterns/, memory/customers/, memory/examples/ = Level 4-5)
   - Are there templates, workflows, experience directories?
   - Does the CLAUDE.md reference and route to memory files?

5. Quality Gate Signals (Level 5+)
   - Hook configurations in .claude/settings.json? (PreToolUse, PostToolUse, etc.)
   - Agent definitions in .claude/agents/ or ~/.claude/agents/?
   - Hooks are quality gates, not infrastructure. They belong here, not at Level 9.

6. Automation Signals (Level 6+)
   - Any shell scripts that call `claude -p` or `claude --print`?
   - Any package.json with Playwright or Puppeteer?
   - Any screenshot scripts, PDF generators, scraping tools?
   - Evidence of JSON output piping?
   - Chrome browser integration enabled (CLI flags)?

7. Multi-Agent Signals (Level 8+)
   - tmux config files or scripts?
   - Multiple CLAUDE.md files for different agent roles?
   - VPS deployment evidence?

8. Always-On Signals (Level 9-10)
   - Cron jobs, launchd services, or systemd services calling Claude on a schedule?
   - Background agents running 24/7 (pm2, systemd)?
   - Autonomous loop scripts (PRD + test suite patterns)?
   - Agent orchestration configs (Agent Teams, Gastown)?
   - Evidence