Stave

A reusable project template for Harness Engineering with Claude Code — Anthropic's CLI agent for software development.

Harness Engineering is a methodology where humans steer and agents execute. The engineer's role shifts from writing code to designing environments, specifying intent, and building feedback loops that allow Claude to do reliable work.

What's Inside

Clone this repo to bootstrap any new project with a complete agent-friendly documentation and tooling structure.

CLAUDE.md                    # Map/TOC — Claude's entry point (~100 lines)
ARCHITECTURE.md              # System layers, dependency rules, module layout

docs/
├── design-docs/             # Design decisions and rationale
│   ├── index.md             # Catalog of all design docs
│   ├── core-beliefs.md      # Golden principles / agent-first rules
│   └── TEMPLATE.md          # Template for new design docs
├── exec-plans/              # Versioned execution plans
│   ├── active/              # Plans currently in progress
│   ├── completed/           # Done plans (for reference)
│   ├── tech-debt-tracker.md # Tech debt log with priority tracking
│   └── TEMPLATE.md          # Template for new plans
├── product-specs/           # Product specifications
│   ├── index.md             # Catalog of all specs
│   └── TEMPLATE.md          # Template for new specs
├── references/
│   └── design-system.md     # Design system reference
├── DESIGN.md                # Design patterns and conventions
├── FRONTEND.md              # Frontend architecture and conventions
├── QUALITY_SCORE.md         # Quality grades per domain/layer
├── RELIABILITY.md           # Reliability standards and SLOs
└── SECURITY.md              # Security policies and boundaries

.claude/
├── settings.json            # Claude Code settings with hooks
└── commands/
    ├── review.md            # /review — self-review before PR
    ├── cleanup.md           # /cleanup — garbage collection scan
    └── plan.md              # /plan — create execution plan

scripts/
└── lint-architecture.sh     # Architectural boundary linting

.github/workflows/
└── claude-review.yml        # Claude-powered PR review CI

Quick Start

1. Clone the template

   git clone https://github.com/GaetanJuvin/stave.git my-project
   cd my-project

2. Find-and-replace [PROJECT_NAME] across all files with your project name

3. Customize the architecture — edit ARCHITECTURE.md with your actual layers, domains, and tech stack

4. Fill in the design system — update docs/references/design-system.md with your colors, typography, and components

5. Set your quality targets — update docs/RELIABILITY.md with real SLOs and docs/SECURITY.md with your auth approach

6. Start building — Claude reads CLAUDE.md first and follows the map to deeper docs as needed

Custom Commands

Command	Purpose
/review	Self-review all changes before opening a PR
/cleanup	Scan for deviations from golden principles and tech debt
/plan	Create a versioned execution plan for complex tasks

Key Principles

1. Repository is the source of truth — if Claude can't read it, it doesn't exist
2. CLAUDE.md is a map, not a manual — ~100 lines pointing to deeper docs
3. Enforce architecture mechanically — linters and CI, not willpower
4. Plans are first-class artifacts — versioned and committed to the repo
5. Corrections are cheap, waiting is expensive — trust the guardrails, ship fast

Read the full methodology in HARNESS_ENGINEERING.md.

Background

Adapted from OpenAI's Harness Engineering methodology, translated for teams using Claude Code. See COMPARISON_OMC.md for how this compares to other Claude Code scaffolding approaches.

License

MIT