Vibe Coder Score

Scan your AI coding setup and get a score. Detects models, agents, MCP servers, memory systems, deploy pipelines, security practices, and more — then rates your setup across 8 dimensions.

🌐 vibecheck.crunchy.tools

---

Quick Start

npx vibecheck-score

Runs in ~3 seconds. No installs beyond npx. No network calls during scanning.

Sample Output

  VIBE CODER SCORE

  Level 72 · Architect
  "Context windows are your canvas. Connections are your brushstrokes."

    Intelligence  ████████░░   78
    Tooling       ██████████  100
    Continuity    ████████░░   82
    Autonomy      ██████████  100
    Ship          ██████░░░░   60
    Security      ███████░░░   70
    Ops           █████░░░░░   45
    Social        ███░░░░░░░   28

  KEY MECHANISMS
    Claude Code skills: handoff, review, session-closer, memory
    OpenClaw orchestrator
    Custom MCP servers (3 detected)
    CLAUDE.md deep continuity context

  GROWTH AREAS
    🚀 Ship (60/100) — Add test coverage and formalize your deploy pipeline
    📊 Ops (45/100) — Structured task tracking would boost your score here

Options

--json          Output raw JSON (pipe to other tools)
--shallow       Skip global checks (~/.zshrc, crontab) for faster scan
--submit        Upload results to vibecheck.crunchy.tools for a shareable card
--handle <name> Custom handle for your result page
--compare       Create or join a side-by-side comparison

---

What It Scans

The probe runs 11 scanners in parallel, checking ~200 signals across 8 categories:

Category	What it looks for
Intelligence	API keys, model providers, routing aliases, local models
Tooling	CLI tools, MCP servers, IDE integrations
Continuity	CLAUDE.md, memory systems, rules, context files
Autonomy	Subagents, hooks, orchestration, tmux/worktrees
Ship	CI/CD, test frameworks, deploy configs, npm scripts
Security	.gitignore coverage, env var handling, agent permissions
Ops	Task tracking, build scripts, monorepo tooling
Social	Public repos, npm packages, webhooks

Tiers

Score	Tier	Description
0–10	Observer	Getting started
11–20	Apprentice	Basic tooling in place
21–30	Practitioner	Using AI consistently
31–45	Builder	Shipping with AI
46–55	Operator	Deep tool ecosystem
56–65	Commander	AI-native workflows
66–75	Architect	System-level thinking
76–85	Orchestrator	Multi-agent mastery
86–100	Industrialist	Full-stack AI production

Prefer to audit first?

git clone https://github.com/brklyngg/vibe-coding-score.git
cd vibe-coding-score/vibecheck
npm install && npm run build:probe
node packages/probe/dist/index.js

---

Privacy & Security

During scanning:

• No network calls during scanning. The only network call is --submit, which is opt-in.
• No file contents loaded into memory. Shell config files (.zshrc, .bashrc, etc.) and .env files are scanned via grep — the probe checks for key names without ever reading values into the Node process.
• Crontab is grep-piped, not dumped — only matching line counts are captured.
• Git author emails are not collected. Only names and commit metadata are read.
• First-run consent prompt tells you exactly what will be scanned before anything runs.

When submitting (--submit):

• A sanitizeForSubmit() layer strips sensitive details before any network call:
  • File paths like ~/.zshrc are normalized to generic labels (shell-config)
  • Detail fields are whitelisted — only numeric counts and enum-like strings pass through
  • Agent names, alias names, plist filenames, and other identifying details are removed
  • scanResults.detections are emptied (only scanner name + duration are sent)
  • platform is set to "redacted"
• No IP logging or analytics telemetry on the server. The submit endpoint source is open at packages/web/src/app/api/submit/route.ts.

For full details — scanner inventory, sample output, sanitization pipeline, merge flow — see vibecheck.crunchy.tools/security.

---

Scoring

Each detection earns points based on its tier (basic through elite). Points are weighted across the 8 categories to produce a score from 0-100 and a tier title. You also get a 4-letter type code (like MARC or VGCD) describing your style across intelligence, autonomy, shipping, and depth dimensions.

Web App

Results submitted with --submit are viewable at vibecheck.crunchy.tools/result/<handle>, including an OG image card for sharing on Twitter, LinkedIn, or anywhere.

The --compare flag enables side-by-side comparisons: create a session code, share it with a teammate, and see exactly where your setups diverge.

---

Monorepo Structure

vibecheck/
  packages/probe/    # npm: vibecheck-score (TypeScript CLI)
  packages/web/      # Next.js 15 web app (vibecheck.crunchy.tools)

Development

npm install
npm run build:probe   # Compile CLI
npm run build:web     # Build web app
npm run dev:web       # Next.js dev server

Development

Issues and feedback welcome. To run it locally or add a new scanner or detection signal:

1. Add entries to packages/probe/src/taxonomy/registry.json
2. Implement the detection logic in the appropriate scanner under packages/probe/src/scanners/
3. Run npm run build:probe and test with node packages/probe/dist/index.js

License

© Gary Gurevich. All rights reserved. This source is publicly viewable but not licensed for reuse, redistribution, or derivative works.