GitHub - dylandrop/business-in-a-box: Reusable Claude Code slash commands for taking a codebase from prototype to production

4 min read Original article ↗

Reusable Claude Code skills for taking a codebase from prototype to production.

Getting Started

  1. Drop a business description into your repo (e.g. in CLAUDE.md or any file Claude can read) describing what your product does and who it's for
  2. Run /project:go-live-checklist — it identifies what's missing and works through decisions with you
  3. Run /project:production-audit — it picks up the checklist and iteratively implements, assesses, and refines until production-ready

That's it. The two commands form a pipeline:

go-live-checklist → production-audit

You can also run either command standalone.

Skills

/project:go-live-checklist

Go-live readiness assessment. Evaluates whether your project is ready to launch as a production business:

  1. Assess — parallel agents scan the codebase for: billing, analytics, testing, error handling, feedback funnels, auth, security, infrastructure, CI/CD, and App Store / Google Play readiness (iOS/Android)
  2. Recommend — compiles findings into external service needs, major business decisions, and minor items, then works through each decision with you interactively
  3. Checklist — generates a prioritized go-live checklist at .claude/go-live/GO-LIVE-CHECKLIST.md

Progress checkpointed to .claude/go-live/state.json.

Runtime output

.claude/go-live/
  state.json                  # durable state — resume from any phase
  assessments/                # per-area assessment files (one per agent)
  recommendations.md          # compiled recommendations
  decisions.md                # user decisions log
  GO-LIVE-CHECKLIST.md        # final prioritized checklist (consumed by production-audit)

/project:production-audit

Automated production-readiness audit loop. Drop into any repo, run the command, and it will:

  1. Discover all user-facing and admin/operational workflows in the codebase — if a go-live checklist exists, its items are incorporated as workflows to implement
  2. Assess every workflow in parallel across 3 lenses — UX, Security, Performance — with platform-specific criteria for web, iOS, and Android
  3. Split findings into conflict-free work packages (no shared files between splits)
  4. Fix all Critical/High/Medium findings via parallel agents in isolated git worktrees
  5. Loop — re-assess, fix new issues, repeat until clean (default 3 iterations, configurable)
  6. Report — generates a final report with findings summary, resolved/open items, and a Manual Steps Required section cataloging everything Claude can't do itself (secrets to configure, services to sign up for, OAuth apps to register, store submissions to complete, DNS records to create, etc.)

Pass an optional iteration count: /project:production-audit 5

All progress is checkpointed to .claude/audit/state.json, so the command survives context flushes, token limits, and session restarts.

Design principles

  • Cross-platform — supports web, iOS (Swift/SwiftUI/UIKit), Android (Kotlin/Compose/XML), React Native, and Flutter
  • Repository-agnostic — discovers project structure, test runners, and build systems dynamically
  • Portable — templates resolve project-local first, then global (~/.claude/skills/), so skills work in any repo without copying files
  • Context-resilient — state machine on disk, phase-specific instruction loading, one-line agent returns
  • Token-optimized — slim 45-line dispatcher, JIT-loaded phase instructions, findings never enter orchestrator context
  • Parallel-safe — file-disjoint splits, worktree isolation, port/DB isolation per agent, post-merge integration verification

File layout

.claude/skills/
  production-audit/
    SKILL.md                                       # slim dispatcher + state machine
    orchestrator-discover-and-assess.md            # phases 1-3 instructions (JIT-loaded)
    orchestrator-execute-and-merge.md              # phases 4-7 instructions (JIT-loaded)
    discover-workflows.md                          # agent: workflow mapping (user + admin)
    assess-ux.md                                   # agent: UX audit (web + mobile + AI-generated code patterns)
    assess-security.md                             # agent: security audit (OWASP Top 10 + Mobile Top 10)
    assess-performance.md                          # agent: performance audit (web + mobile)
    execute-split.md                               # agent: implement fixes in worktree
  go-live-checklist/
    SKILL.md                                       # go-live readiness assessment

Runtime output

.claude/audit/
  state.json                  # durable state — resume from any phase
  workflows.md                # discovered workflows (user-facing + admin/operational)
  findings/iter-{N}/          # assessment findings, versioned per iteration
  splits/iter-{N}/            # work splits + execution results
  REPORT.md                   # final audit report

Installation

Global install (recommended)

Install once and use in any repo — no per-project setup needed:

git clone https://github.com/dylandrop/business-in-a-box.git /tmp/biab
mkdir -p ~/.claude/skills
cp -r /tmp/biab/.claude/skills/* ~/.claude/skills/
rm -rf /tmp/biab

Then run in any repo with Claude Code:

/user:go-live-checklist
/user:production-audit

Per-project install

Copy into a specific repo's .claude/skills/:

# From inside your target repo
git clone https://github.com/dylandrop/business-in-a-box.git /tmp/biab
cp -r /tmp/biab/.claude/skills/ .claude/skills/
rm -rf /tmp/biab

Then run in Claude Code:

/project:go-live-checklist
/project:production-audit

Requirements

  • Claude Code CLI
  • Git (for worktree-based parallel execution)

License

MIT