The .claude/ setup that ships real products.
_{A complete, battle-tested Claude Code configuration — rules, agents, and commands ready to drop into any project.}

---

Most Claude Code setups are one CLAUDE.md file and vibes. This is what running Claude Code at production level actually looks like — quality gates with real thresholds, a five-agent pipeline, git discipline that prevents accidents, security rules that catch secrets before they ship, and /build, /review, /debug commands you can run on every project without thinking.

Fork it. Strip what you don't need. Add your stack. Five minutes of setup, months of better output.

---

What's Included

.claude/
├── rules/
│   ├── 00-security.md        # Secrets, auth, dependency scanning
│   ├── 01-testing.md         # TDD, coverage thresholds, determinism
│   ├── 02-code-conventions.md # TypeScript strict, naming, no god functions
│   ├── 03-deployment.md      # Deploy gates, environment hygiene
│   ├── 04-quality-gates.md   # Binary pass/fail thresholds
│   └── 05-git-discipline.md  # Commit format, branch strategy, session closure
├── agents/
│   ├── architect.md          # Plans structure before any code is written
│   ├── senior-engineer.md    # Implements features with context discipline
│   ├── code-reviewer.md      # Reviews against rules, not vibes
│   ├── test-runner.md        # Validates build, coverage, and regressions
│   └── ui-inspector.md       # Catches layout bugs before they ship
└── commands/
    ├── build.md              # /build — feature pipeline from branch to push
    ├── review.md             # /review — structured quality check
    └── debug.md              # /debug — four-phase debugging protocol

---

Why This Exists

Claude Code without structure produces inconsistent results. The same task run twice gives different quality. Mistakes compound across a long session. You spend tokens fixing things that rules would have prevented.

The files in this repo encode the guardrails that make Claude Code reliable:

• Rules tell Claude what it must always do (and never do) on your project
• Agents give each role a scoped identity so context doesn't bleed between concerns
• Commands give you repeatable workflows you can trigger with /build, /review, /debug

These aren't theoretical. They were refined across multiple production builds — catching real bugs, preventing real deploys from breaking, and cutting the back-and-forth that wastes tokens.

---

Quick Start

# Copy into your project
cp -r .claude/ /path/to/your/project/

# Add your project context
echo "# Project context here" > /path/to/your/project/.claude/rules/00-project.md

Then in Claude Code:

• /build <feature description> — implement a feature end to end
• /review — quality gate check before pushing
• /debug — structured debugging when something breaks

---

Rules Overview

00-security.md — No secrets in source. Zero CRITICAL/HIGH vulns before merge. Auth defaults to strong. Input validation on all external data.

01-testing.md — TDD first. ≥80% line coverage. Deterministic tests. E2E for critical paths.

02-code-conventions.md — TypeScript strict. No any. No god functions over 50 lines. No magic numbers. No commented-out code.

03-deployment.md — Never push to main directly. Human reviews before prod merge. Preview URL verified before merging.

04-quality-gates.md — Binary pass/fail. ≥90% PRD completeness before building. ≥80% coverage before shipping. Never start the build phase before the planning gate passes.

05-git-discipline.md — Commit after every logical change. Conventional commit format. Push before ending any session.

---

Commands Overview

/build <feature> — Branches from main → plans → implements (fresh subagent per task) → validates → pushes.

/review — Structured pass/fail across build, lint, tests, security, UI, API routes. File:line references for every issue.

/debug — Four-phase protocol: Investigate → Analyse → Hypothesise → Fix and Verify. One change per test run.

---

Agent Roles

Agent	Model	Role
architect	Opus	Plans structure, file tree, dependency map
senior-engineer	Sonnet	Implements features, writes commits
code-reviewer	Haiku	Reviews code against rules
test-runner	Haiku	Validates build, coverage, routes
ui-inspector	Sonnet	Checks layout, positioning, cursor rules

---

Adapting for Your Stack

• 02-code-conventions.md — Replace TypeScript rules with your language's conventions
• 03-deployment.md — Update build commands and hosting provider references
• 04-quality-gates.md — Adjust coverage thresholds to your project's maturity
• Add .claude/rules/00-project.md with your specific stack and architecture decisions

---

Licence

MIT