I Built a Tool That Roasts Your CLAUDE.md. Mine Got Called an 'S-Tier Micromanager'.

I've been building serious tools for Claude Code operators. Diagnostic scanners. Usage analytics. Audit logs. Cost calculators.

Then I thought: what if one of them was just mean?

So I built CC Roast — pick your CLAUDE.md and get roasted.

What It Does

You pick (or paste) your Claude Code instruction file. It analyzes the text and delivers:

1. A verdict — one of 11 personality types for your config
2. Four scores — Verbosity, Strictness, Complexity, Thoroughness (each 0-100)
3. Personalized roasts — based on what it actually finds in your file
4. Genuine compliments — because even roast comics say something nice

No AI involved. Pure client-side text analysis. Your CLAUDE.md never leaves your browser.

The 11 Verdicts

Verdict	What triggers it
🕵️ S-Tier Micromanager	25+ strict rules AND 200+ lines
🪂 Helicopter Operator	15+ strict directives
📚 The Novelist	500+ lines
🔒 Security Paranoid	Heavy security + git safety
⚙️ Overengineered Perfection	High complexity, 100+ lines
✌️ Vibes Only	Under 3 strict rules, under 50 lines
🧘 The Minimalist	15-60 lines, clean
📝 Sticky Note Energy	Under 15 lines
⚖️ Well-Balanced Config	Good thoroughness, moderate strictness
🎰 Trust Fund AI	Under 30 lines, barely any rules
🔥 Standard Issue Operator	None of the above

My CLAUDE.md Got Demolished

I tested it on my actual CLAUDE.md. It's... 342 lines across three files.

Verdict: 🕵️ S-Tier Micromanager
"Your AI has less autonomy than a vending machine"

The scores:

• Verbosity: 75/100 (342 lines — that's a lot of opinions)
• Strictness: 80/100 (33 strict directives. The tool counts both English and Japanese keywords.)
• Complexity: 68/100 (hooks, MCP, tables, code blocks)
• Thoroughness: 85/100 (git safety, testing, security — all covered)

The roasts:

• "342 lines. That's a lot of opinions about how an AI should behave. You could fit a working app in fewer lines."
• "52 section headers. Your CLAUDE.md has more chapters than a self-help book."
• "2 TODO/FIXME/HACK found. Even your instruction file has tech debt."

But it also said: "Your git safety rules are genuinely good. Your repos will thank you."

Fair.

Why Comedy Works for Learning

Every roast maps to a real lesson:

• "No git safety rules? Living dangerously." → Add git safety rules
• "Zero mentions of testing." → Add test requirements
• "15 hooks. Safety net made of safety nets." → Maybe simplify your hooks

The format makes you actually read the feedback. A score of "42/100 Thoroughness" is forgettable. "Your .env file is doing a victory lap on GitHub" sticks.

How It Works (Technical)

Pure regex-based text analysis:

• Count NEVER, DO NOT, ALWAYS, MUST, IMPORTANT, CRITICAL
• Count Japanese equivalents: 禁止, するな, 絶対, 必ず, 厳命, 不可
• Measure line/word/char counts
• Detect headers, code blocks, tables, emoji
• Check for key patterns (git safety, test mentions, security rules)
• Calculate density metrics (strict rules per 1000 words)
• Match against verdict conditions (priority-ordered)

Zero dependencies. Single HTML file. ~1000 lines.

Try It

1. Open CC Roast
2. Click "Pick CLAUDE.md File" and select yours — or paste the contents directly
3. Get roasted
4. Share the verdict (if you dare)

What verdict did your config get?

---

Free Tools for Claude Code Operators

Tool	What it does
cc-health-check	20-check setup diagnostic (CLI + web)
cc-session-stats	Usage analytics from session data
cc-audit-log	Human-readable audit trail
cc-cost-check	Cost per commit calculator
cc-wrapped	Your AI year in review
cc-roast	Your CLAUDE.md, brutally honest
claude-code-hooks	Fix what the roast found — 10 hooks + 5 templates

Just for fun: CC Bingo — 50 Claude Code moments. How many have you hit?

More tools: Dev Toolkit — 100+ free browser-based tools for developers. JSON, regex, colors, CSS, SQL, and more. All single HTML files, no signup.