Claude Code Template

This repository provides a starter template for configuring Claude Code with well-structured building blocks. The starter-template/ directory contains ready-to-use configuration files that you can copy into your projects.

This project serves as both a Proof of Concept and a Living Reference. It uses the same starter-template/ configuration as its own operational setup. An automated ♻️ pipeline continuously evaluates improvements and best practices, updating the template to maintain alignment with current standards. The agents continuously reference official sources to ensure the template stays accurate and up to date.

Quickstart

Copy the starter template into your project.

cp -rn starter-template/. your-project/

The -n flag skips existing files, so you can re-run this safely after template updates to pick up new files.

Building Blocks

Block	What It Does	Reliability	Eval	Docs
CLAUDE.md	Your project instructions. Claude reads them at the start of every session.	Always read. Instructions are advisory, not enforced.	4/4	Docs
Hooks	Scripts that run before or after Claude uses a tool. Can block dangerous actions.	Always executed. Blocking decisions are enforced.	24/24	Docs
Settings	Controls what Claude can and cannot do. Permissions and environment.	Always applied. Deny rules are enforced.	2/2	Docs
Rules	Instructions for specific file types. Load automatically when files match.	Always read. Same as CLAUDE.md: advisory, not enforced.	-	Docs
Skills	Reusable workflows. Trigger with /command or let Claude decide.	/command always works. Claude may skip auto-trigger.	2/2	Docs
Memory	Knowledge Claude saves and recalls across sessions.	File loading is reliable. Auto-save depends on Claude's judgment.	-	Docs
Subagents	Helper agents that handle tasks independently and report back.	Reliable for most tasks.	-	Docs
MCP	Connects Claude to external tools and services.	Local tools connect reliably. Remote tools may disconnect.	-	Docs
Agent Teams	Multiple Claude agents working together on shared tasks.	Experimental. Not enabled by default.	-	Docs
Plugins	Add-ons like code intelligence for your programming language.	Code intelligence is stable. Others vary.	-	Docs

Eval column shows results from our eval suite. Hook tests are deterministic. LLM evals are single-run and results may vary. Reliability is based on official docs and community reports (March 2026).

Starter Template

Every file includes inline guidance. Open and extend for your project.

starter-template/
├── CLAUDE.md                  Project instructions
├── CLAUDE.local.md            Personal overrides
├── .mcp.json                  MCP server connections
└── .claude/
    ├── settings.json          Permissions and environment
    ├── settings.local.json    Personal permission overrides
    ├── skills/                Skill template
    ├── rules/                 Rule template
    ├── agents/                Subagent template
    └── hooks/                 Hook script template

File	What It Provides
CLAUDE.md	Project instructions with sections for commands, style, architecture, and conventions
.claude/skills/	Skill template with all configuration options documented
.claude/rules/	Rule template showing global and file-scoped patterns
.claude/agents/	Subagent template with all configuration options documented
.mcp.json	MCP server configuration for connecting external tools
.claude/hooks/	Hook script template with event list and common patterns
.claude/settings.json	🌟 Shared permissions, hooks, and environment variables. TypeScript LSP plugin is included ("enabledPlugins": {"typescript-lsp@claude-plugins-official": true}) and can be configured (docs). Auto memory is enabled by default ("autoMemoryEnabled": true) and can be configured (docs).
CLAUDE.local.md	Personal project overrides
.claude/settings.local.json	Personal permission overrides

📌 Configuration Levels

Claude Code supports 4 levels. Higher levels override lower ones.

Level	Location	Shared?	Priority
Managed	System directories	Yes (system-wide)	Highest
Local	.claude/*.local.*	No	High
Project	.claude/ in repo	Yes (git)	Medium
User	~/.claude/	You only	Base

💾 Memory & Configuration Guide

Comprehensive guide covering:

• Directory Structure — Project vs user level (.claude/ vs ~/.claude/)
• CLAUDE.md vs Rules vs Memory — Three distinct systems and when to use each
• Auto Memory (MEMORY.md) — Claude's automatic learning system
• Modular Rules — Organizing instructions with .claude/rules/
• Hierarchical Loading — Multiple CLAUDE.md files in monorepos

Read the full Memory & Configuration Guide →

🧩 Context Engineering

Tool	Description
Context7	Up-to-date, version-specific library documentation for AI coding tools

For a deeper dive, see Context Engineering for Coding Agents by Martin Fowler.

🌱 Living Reference

This repository uses its own template. Root .claude/ is this project's real config; starter-template/ is the generic template you copy.

Pipeline

graph TB
  subgraph Trigger
    direction LR
    I[GitHub Issue: track URL]
    C[Copilot: @claude mention]
    R[Claude Remote Control]
  end

  subgraph Agentic Workflow
    direction LR
    SA[source-analyzer: fetch and audit]
    EV[evaluator: assess and decide]
    TU[template-updater: apply changes]
  end

  subgraph Output
    direction LR
    AP[approved.md + PR created]
    RJ[rejected.md: reason logged]
    _[ ]
  end

  I --> SA
  C --> SA
  R --> SA
  SA --> EV
  EV -->|implement| TU
  EV -->|on hold| RJ
  EV -->|not relevant| RJ
  TU --> AP

  classDef trigger fill:#dbeafe,stroke:#3b82f6,color:#1e3a5f
  classDef workflow fill:#fef3c7,stroke:#f59e0b,color:#78350f
  classDef approved fill:#d1fae5,stroke:#10b981,color:#064e3b
  classDef rejected fill:#fee2e2,stroke:#ef4444,color:#7f1d1d
  classDef hidden fill:none,stroke:none,color:transparent

  class I,C,R trigger
  class SA,EV,TU workflow
  class AP approved
  class RJ rejected
  class _ hidden

Loading

Pipeline decisions are logged in agentic-workflow-output/:

File	Logs	Written by
approved.md	Implemented updates with PR links	Main session, after template-updater completes
rejected.md	Skipped or on-hold updates with reasons	Main session, after evaluator decides

🔖 Official Sources

Built from the official Claude Code documentation.

• Claude Code Docs
• Best Practices
• Effective CLAUDE.md
• The Complete Guide to Building Skills
• Context Engineering for Coding Agents

Contributing

Found something worth tracking? Share it by creating an issue with,

• Title: track: <URL>
• Body: Optional notes or context

I will review and run the pipeline if relevant.