Claude Workflow Rules — CLAUDE.md Template for Claude Code

A drop-in CLAUDE.md template with opinionated rules and workflows for Claude Code, Anthropic's AI coding assistant. Copy one file into your project root to get consistent, high-quality AI output: planning discipline, persistent project memory, language-aware testing, security defaults, and atomic git practices.

Quick Start

1. Copy CLAUDE.md into your project root.

   macOS / Linux:

   curl -o CLAUDE.md https://raw.githubusercontent.com/rolandfarkasCOM/claude-workflow-rules/main/CLAUDE.md

   Windows (PowerShell):

   Invoke-WebRequest -Uri https://raw.githubusercontent.com/rolandfarkasCOM/claude-workflow-rules/main/CLAUDE.md -OutFile CLAUDE.md

   Or clone this repo and copy the file manually.

2. Open Claude Code in your project and tell it to set everything up:

Read CLAUDE.md and set up the project — create the memory directory and task files as described.

Claude will create .claude/memory/ (with decisions.md, patterns.md, context.md, session.md) and tasks/ (with todo.md, lessons.md). On every future session, Claude Code loads CLAUDE.md automatically at startup.

What's Included

• First-Time Setup — Claude asks about your git, PR, and testing preferences on first run
• How to Work — planning, subagents, verification, elegance, autonomous bug fixing, self-improvement loop
• Project Memory — structured files for architectural decisions, code patterns, domain context, and session logs
• Testing — language-appropriate test frameworks (Vitest, Jest, Playwright, pytest, PHPUnit, go test, cargo test), CI setup, what to test
• Error Handling — search-first debugging, language-appropriate patterns, fail loudly in dev / gracefully in prod
• Task Tracking — structured approach using tasks/todo.md and tasks/lessons.md
• Tool Integration — GSD for structured execution, Claude-Mem for automatic session memory
• Code Standards — read before write, minimal changes, root cause fixes, edge case handling
• Security — input validation, parameterized queries, rate limiting, audit logging
• Dependencies — minimize packages, audit before adding, record decisions
• Documentation — update when context changes, not for its own sake
• Git Discipline — atomic commits, no force push, user controls commit workflow
• Checklists — new project setup and session start protocols

Memory System

Two layers of memory work together to keep Claude Code aware of your project across sessions:

Layer	How it works	What it captures
Claude-Mem (automatic)	Plugin captures every session silently	Code changes, errors, decisions — everything
.claude/memory/ (deliberate)	You update these files intentionally	Architectural decisions, patterns, domain knowledge, session notes

Claude-mem is great for "what did I do last week?" — the memory files are for "why did we choose Hono over Express?"

Optional Tools

Referenced in the rules but not required:

# GSD — structured project execution with phases and verification gates
npx get-shit-done-cc@latest

# Claude-Mem — persistent session memory with semantic search (requires Bun)
npx claude-mem install

Customization

The rules are a starting point. Add project-specific sections to CLAUDE.md for:

• Build commands (npm run dev, npm run build, npm run typecheck)
• Tech stack details (framework, database, hosting)
• Project-specific conventions (naming, file structure, component patterns)
• Key directories and their purpose
• Performance requirements or constraints

FAQ

What is CLAUDE.md?

CLAUDE.md is the convention Claude Code uses for project-specific instructions. When you open a project, Claude Code automatically loads CLAUDE.md from the repo root and treats it as system prompt context for the entire session — so any rules, conventions, or commands you put there shape every response.

How is this different from a regular README?

A README documents the project for humans. CLAUDE.md documents it for the AI coding assistant — preferred workflow, testing approach, security rules, things to avoid. The two complement each other.

Can I use these rules with Cursor, Cody, or other AI coding tools?

The underlying rules (planning, testing, security, git discipline) apply to any AI assistant. The Claude Code–specific bits — the CLAUDE.md auto-loading convention, the /gsd-* slash commands, and claude-mem integration — only work with Claude Code itself.

Where do I put project-specific overrides?

Edit CLAUDE.md directly after copying it. The file lives in your repo and is yours to customize — add a section at the bottom for your stack, build commands, and conventions.

License

MIT