claude-team-init

Stop burning $40/day on repeated builds. Get a senior engineering team for $3/month.

One command scans your codebase, generates specialized AI agents, and sets up a shared cache so your agents never repeat each other's work. Then /team-audit finds every problem in your code and fixes it.

You Vibe Coded Your App. Now What?

You shipped fast with AI. You have users. But you have that constant low-grade anxiety:

• "What if there's a security hole I don't know about?"
• "Each fix introduces two new problems"
• "30k lines of code I didn't write and can't debug"
• "I'm burning tokens because every agent re-runs the same build"

claude-team-init fixes all of this.

One Command

npx claude-team-init

  Scanning your codebase...
  Found: Next.js project

  Detected:
    Framework:    Next.js
    Database:     Supabase
    Auth:         Supabase Auth
    Payments:     Stripe
    API routes:   yes
    Mobile:       yes

  Agents to generate:
    build-gate          haiku   Build + lint verification
    qa-signoff          haiku   Final QA with physical proof
    regression-checker  haiku   Side effect detection
    orchestrator        opus    3-solution finder (97% confidence)
    team-audit          opus    Full code audit + auto-fix
    team-review         opus    Fast agent team fitness check
    api-prober          haiku   API endpoint testing
    security-reviewer   sonnet  DB security audit (Supabase)
    mobile-expert       sonnet  Mobile compatibility check

  Auto-run agents on edits and commits? Yes
  Run team audit now? Yes

  Launching team audit...

What /team-audit Finds (and Fixes)

Not just warnings. Fixes. To 95% confidence.

| What it catches | What linters catch | |---|---| | 3 API routes with zero auth | Unused variable | | SQL injection in search endpoint | Missing semicolon | | 2,400 lines of dead code | Wrong indentation | | Duplicated fetch logic in 8 files | Import order | | God component with 847 lines | Trailing whitespace | | No error boundaries anywhere | Inconsistent quotes |

After scanning, it works through every finding — critical first — fixes the code, validates the build, and iterates until the codebase is solid.

  TEAM-AUDIT-REPORT.md

  Health Score: 62/100

  Dead Code: 2,400 lines removed
  Duplication: 8 fetch calls → shared api.ts
  Security: 3 critical holes patched
  Missing: Error boundaries, loading states, input validation added

  Fixes Applied: 23
  Build: PASS
  Tests: PASS (47 passed, 0 failed)
  Overall Confidence: 96%

Token Savings (the real money)

Your agents repeat each other's work constantly. Build-gate runs the build. Then QA runs the same build. Then regression runs it again. That's 3x the tokens for the same output.

The shared cache eliminates this:

| | Without Cache | With Cache | |---|---|---| | Tokens per QA cycle | ~34,000 | ~250 | | Daily cost (heavy use) | $40-60 | $3-10 | | Monthly savings | - | $900-1,200 |

Real numbers: 4.3M tokens saved in 2 days on a production project. The cache costs $3/month and pays for itself in the first hour.

Inside Claude Code

After setup, just type slash commands:

/team-audit       Full codebase scan + fix everything
/team-review      Quick agent team fitness check
/team-launch      Parallel orchestrator for multiple problems
/team-init        Re-run setup

Or run agents directly:

claude --agent build-gate
claude --agent qa-signoff
claude --agent security-reviewer

Auto-Triggers

Say yes to hooks during setup and agents run automatically:

• After code edits — build-gate catches type errors immediately
• After commits — build-gate, regression-checker, qa-signoff run in sequence
• On auth file changes — auth-verifier activates
• On database changes — security-reviewer activates
• On payment code changes — billing-bot activates

You don't have to remember to run anything.

The Orchestrator

Give it a hard problem. It:

1. Generates 3 completely different solutions (not variants — different strategies)
2. Validates each through build-gate + regression-checker + qa-signoff
3. Scores confidence: build (30pts) + regression (30pts) + QA (25pts) + judgment (15pts)
4. Iterates until 97% confidence (92% acceptable after 5 rounds)

./scripts/team-launch.sh "fix login timeout" "add dark mode" "optimize image loading"

Each problem gets its own tmux window and git worktree. Go to sleep. Wake up to tested, reviewed solutions on branches ready to merge.

Supported Stacks

Detects and adapts to your specific tools:

| | Detected | |---|---| | Languages | JavaScript/TypeScript, Python, Ruby, Go, Rust, Java/Kotlin, Dart, Swift, Elixir, PHP | | Frameworks | Next.js, Nuxt, Remix, Astro, SvelteKit, React, Vue, Express, Django, FastAPI, Flask, Rails, Gin, Axum | | Databases | Supabase, Prisma, Drizzle, MongoDB, PostgreSQL, Firebase, SQLite | | Auth | NextAuth, Clerk, Supabase Auth, Firebase Auth, Passport, Lucia | | Payments | Stripe, Lemon Squeezy, Paddle | | Storage | S3, Supabase Storage, Google Cloud Storage, Cloudinary, Vercel Blob | | Testing | Jest, Vitest, Playwright, Cypress, pytest, RSpec, go test, cargo test | | Mobile | iOS/SwiftUI, Android/Kotlin, Flutter, React Native |

Monorepos too — scans web/, apps/, packages/ subdirectories automatically.

Quick Start

Option A: npm

npm install -g claude-team-init
cd your-project
claude-team-init

Option B: npx (no install)

cd your-project
npx claude-team-init

Option C: Inside Claude Code

/team-init

What Gets Created

your-project/
  .claude/
    agents/           # 6-12 agents tailored to your stack
    settings.local.json  # Permissions + auto-trigger hooks
    rules.json        # Agent activation rules
  scripts/
    team-launch.sh    # Parallel orchestrator launcher
    cache-hook.sh     # Fly.io cache integration
  CLAUDE.md           # Workflow rules (appended, not overwritten)
  TEAM-AUDIT-REPORT.md  # Generated by /team-audit

Your source code is never modified during setup. Only /team-audit touches code (to fix what it finds).

FAQ

Do I need Fly.io? No. Agents work without it — they just can't share results. The cache is optional but saves $20-40/day.

How much does it cost? The tool is free. Cache server is ~$3/month on Fly.io. Agents use your Claude API credits (haiku: ~$0.01/run, opus orchestrator: ~$0.50-2.00/cycle).

Can I customize the agents? Yes. They're markdown files. Edit anything.

What if I already have a CLAUDE.md? It appends. Your existing rules stay.

Is it safe? The tool only creates config files during setup. /team-audit makes code changes but validates everything (build + tests must pass, 95% confidence minimum).

Built By Someone Who Burns Tokens Too

This started because I was burning 600K tokens a night running parallel AI agents on my own projects. Built the cache, saved $40/day. Built the audit, caught security holes I didn't know existed. Figured every vibe coder could use the same thing.

License

MIT