make-me-claude-god

May 9, 2026 · View on GitHub

A Claude Code skill that turns you into a Claude Code power user without reading a single guide.

Install

In Claude Code, run:

/plugin marketplace add josharsh/make-me-claude-god

Or manually:

mkdir -p ~/.claude/skills/make-me-claude-god
curl -sL https://raw.githubusercontent.com/josharsh/make-me-claude-god/main/skills/make-me-claude-god/SKILL.md \
  -o ~/.claude/skills/make-me-claude-god/SKILL.md

Manual: Copy skills/make-me-claude-god/SKILL.md to ~/.claude/skills/make-me-claude-god/.

Why I Built This

There are 150+ Claude Code skills, a dozen awesome lists, marketplace sites with ratings, and blog posts titled "Top 10 Skills You Need." The ecosystem is huge and growing every week.

I don't want to browse any of that. I don't want to read comparisons. I don't want to figure out which combination of skills, hooks, CLAUDE.md settings, and MCP servers works best for my specific stack and workflow.

I just want to type one command and have Claude figure it out for me.

That's what /make-me-claude-god does. It scans your project, asks a few questions about how you work and what frustrates you, then sets up everything — skills, CLAUDE.md, hooks, settings — optimized for your specific situation. Minimum install, maximum impact. Because the research is clear: 3-4 well-chosen skills beat 10 random ones. Past 8, Claude starts getting confused and your context window fills up with instructions nobody's using.

How It Works

The Wizard

You: /make-me-claude-god

Claude: Let me scan your environment first...

  Found: TypeScript project (Next.js), Vitest, GitHub Actions CI,
  no CLAUDE.md, 2 skills installed (active-listening, prove-it)

  First question: Is this a solo project or team?

You: Team of 4.

Claude: What wastes the most time in your Claude Code sessions? Pick 1-2:
  - Claude doing more than I asked
  - Getting stuck in fix-break loops
  - Tests that don't catch real bugs
  - Inconsistent patterns across the team
  ...

You: Claude doing more than I asked, and inconsistent patterns.

Claude: Here's what I'd set up:

  Essential:
  • stay-focused — flags out-of-scope edits (~400 tokens)
  • CLAUDE.md with team coding conventions (~300 tokens)

  Optional:
  • skillmother validate in CI — catches skill drift on PRs

  Skipping:
  • unstuck — you didn't mention fix-break loops
  • talk-slow — more useful for solo deep-dive sessions

  Context Budget:
    Current: 2 skills (~700 tokens)
    After setup: 3 skills + CLAUDE.md (~1400 tokens)
    Remaining: ~198k tokens (99.3%)
    Verdict: Healthy

  Want me to proceed?

3-4 questions. Then you're a Claude god.

The Audit

Already set up? Run an audit to clean house.

You: /make-me-claude-god audit

Claude: Reviewing your setup...

  ✓ active-listening — well-formed, relevant, ~350 tokens
  ⚠ prove-it — references vitest.config.ts which doesn't exist anymore
  ✗ old-skill — no frontmatter, 2000+ lines, ~1200 tokens

  Recommendations:
  • Update prove-it's file references
  • Remove old-skill — burning 1200 tokens/message for nothing

The Status Check

You: /make-me-claude-god status

Claude Code Setup:
  Skills (3): active-listening, prove-it, stay-focused
  CLAUDE.md: exists (287 tokens)
  Hooks: pre-commit (eslint)

  Context baseline: ~1,437 tokens
  Health: Healthy

What It Sets Up

Not just skills. The whole thing.

Layer	Examples
Skills	Best-fit from the ecosystem (web-searches for latest, not a hardcoded list)
CLAUDE.md	Project conventions, key files, team rules
Hooks	Pre-commit formatting, post-tool checks
Settings	Permissions, model preferences, tool restrictions
MCP Servers	External integrations if needed (Slack, Linear, etc.)

Design Principles

Never more than 5 skills. 3-4 well-chosen beats 10 random.
Your pain points drive everything. Not a checklist — recommendations connect to what you said wastes your time.
Context-budget aware. Shows the token cost of every recommendation.
Honest about what it skips. Tells you what it's NOT recommending and why.
Searches the web for latest. The ecosystem moves fast. Not a frozen list.
One question at a time. Done in 3-4 questions, never more than 6.

Commands

Command	What it does
`/make-me-claude-god`	Run the full setup wizard
`/make-me-claude-god audit`	Review existing setup, find waste
`/make-me-claude-god status`	Show what's installed and context budget

Testing

Tested with skillmother:

skillmother test skills/make-me-claude-god/

Uninstalling

rm -rf ~/.claude/skills/make-me-claude-god

License

MIT