May 4, 2026 — CLAUDE.md template updated. The CLAUDE.md template has been modernized with 3 new variants (CLAUDE-template-1.md, CLAUDE-template-2.md, CLAUDE-template-3.md) following official Anthropic best practices — including dual-memory architecture, progressive disclosure, and standalone behavioral rules. Existing users: use CLAUDE-migrate-to-new-template.md as an AI prompt to migrate your current CLAUDE.md to any of the new templates while preserving all your project-specific content. See the CLAUDE.md Templates section for details.

• My AI journey Substack to follow me in more indepth AI adventures - https://ai.georgeliu.com
• Threads - https://www.threads.com/@george_sl_liu
• BlueSky - https://bsky.app/profile/georgesl.bsky.social

My Claude Code Project's Starter Settings

Alternate Read Me Guides

The beauty of using AI is that I can get AI to generate alternate Read Me guides using different formats and styles to suit different readers and learners. So I asked Claude Code to generate alternate Read Me guides in different styles - pick one that suits you best or read all of them to get a better understanding of how to use this project.

• README.md - Original written version I did myself below
• README-v2.md - Progressive Disclosure Guide (Beginner → Intermediate → Advanced sections)
• README-v3.md - Task-Based Guide ("I want to..." navigation for goal-oriented users)
• README-v4.md - Technical Reference Manual (Chapter-based, dense reference format)

Overview

My Claude Code project's starter settings and Claude Code hooks and slash commands are provided in this repository for users to try out. The CLAUDE.md is setup as set of memory bank files to better retain context over many chat sessions. The repo now includes 3 CLAUDE.md templates following official Anthropic best practices with progressive disclosure, dual-memory architecture, and a migration guide for existing users. Be sure to read the official Claude Code docs first at https://docs.anthropic.com/en/docs/claude-code/overview and sign up for a paid Claude AI account to use Claude Code. You can pay for Claude Pro $20/month, Claude Max $100/month or Claude Max $200/month. The paid Claude tier plans will include varying quotas for usage and rate limits outlined here. You can also use Claude Code with Z.AI to get higher token usage quotas and get access to Z.AI GLM-4.7 LLM models within Claude Code. Use Z.AI invite code for additional 10% discount which can stack with current 50-60% yearly discounts.

1. Copy the files in this Github repo to your project directory (where you intended codebase will be).
2. Modify the template files and CLAUDE.mdto your liking. .claude/settings.jsonneeds to install Terminal-Notifier for macOS https://github.com/centminmod/terminal-notifier-setup. If you're not using macOS, you can remove.claude/settings.json`.
3. After launching Claude Code for the first time within your project directory, run /init so that Claude Code analyses your code base and then populates your memory bank system files as per CLAUDE.md` instructions.
4. Optional step highly recommended: Install Visual Studio Code (beginners YouTube video guide and here) and Claude Code VSC Extension.
5. Optional step highly recommended: Sign up for Github.com account and install Git for Visual Studio Code. Checkout YouTube guides here and here.
6. CLAUDE.md updated to instruct models to use faster tools so for macOS: brew install ripgrep fd jq
7. Optional step to setup Claude Code, Codex GPT-5, Gemini CLI, OpenCode, Vercel CLI, Cloudflare Wrangler, Amazon AWS CLI, all in a single isolated Visual Studio Code dev container running Debian 12. Allowing you to run YOLO modes for Claude Code with dangerously_skip_permissions enabled and Codex CLI with sandbox_mode = danger-full-access etc.
8. Claude Code via Claude Desktop apps use Git Worktrees. You may need to create a .worktreeinclude file as outlined here.
9. If you use Cloudflare and ClerkOS platforms in your apps, you can keep either CLAUDE-cloudflare.md or CLAUDE-cloudflare-mini.md supplementary reference docs and update CLAUDE.md referencing either file to help AI understand Cloudflare and ClerkOS platforum documentation and products. Edit documentation templates as needed i.e. if you do not use ClerkOS platform, you can remove those sections.
10. If you use Convex database in your apps, you can use CLAUDE-convex.md supplementary reference docs for building Next.js and React apps with Convex backend deployed on Cloudflare Pages.
11. Useful read Advent of Claude: 31 Days of Claude Code by Ado Kukic from Anthropic.
12. Useful Claude Code with Claude Opus 4.5 YouTube video by Alex Finn and Claude Code YouTube video by Matt Maher.
13. Configure Claude Code with Z.AI to get higher token usage quotas and get access to Z.AI GLM-4.7 LLM models within Claude Code.
14. For folks intending to use Claude LLM models with OpenClaw, read this first for security, deployment and cost optimizations https://github.com/centminmod/explain-openclaw.

CLAUDE.md Memory Bank system

CLAUDE.md uses a memory bank system of files for Claude Code to better retain context over many chat sessions. Example of Claude Code thinking output when I ask it to update memory bank after a successful task completion and git commit:

> update memory bank

∴ Thinking…

  The user wants me to update the memory bank files. I should update the relevant CLAUDE-*.md files to reflect the v0.1.6 changes and current state. Let me check
   which memory bank files exist and update them.

∴ Thinking…

  I need to continue updating the memory bank files. I'll update:
  1. CLAUDE-patterns.md - Add export path handling pattern (Pattern 22)
  2. CLAUDE-decisions.md - Add ADR-017 for Phoneme API migration and ADR-018 for export path handling
  3. CLAUDE-troubleshooting.md - Add new issues and solutions for v0.1.6

∴ Thinking…

  The memory bank files have been modified but not committed. The v0.1.6 release was already committed and tagged, but the memory bank documentation updates are
  still uncommitted. Let me provide a summary of the completed work and offer to commit these documentation changes.

⏺ Summary: v0.1.6 Release Complete ✅

  Memory Bank Updates Complete:
  - CLAUDE-activeContext.md - v0.1.6 session context
  - CLAUDE-patterns.md - Added patterns 22-24 (Export Path, Format Detection, API Migration)
  - CLAUDE-decisions.md - Added ADR-017 & ADR-018
  - CLAUDE-troubleshooting.md - Added 4 new issue/fix entries

  v0.1.6 Fixes Applied:
  1. Export to absolute paths (e.g., --export /tmp/test.wav)
  2. Format detection from filename extension (prevents .mp3.wav)
  3. Phoneme API migrated to /dev/phonemize
  4. Test scripts updated for --export "" syntax

  Test Results: All 29 tests passing

  The memory bank documentation updates are ready to commit. Would you like me to commit them

CLAUDE.md Templates

This repo provides 3 modernized CLAUDE.md templates following official Anthropic best practices:

Template	Lines	Philosophy	Best For
CLAUDE-template-1.md	~101	Compact self-contained + memory resilience	Quick starts, small projects
CLAUDE-template-2.md	~153	Memory bank headline + dual memory	Existing memory bank users
CLAUDE-template-3.md	~105	Progressive disclosure native	Teams, max context efficiency

The current CLAUDE.md uses the Template 3 format. All templates include:

• Dual-memory architecture (git-shared CLAUDE-*.md + machine-local auto memory for resilience)
• Progressive disclosure (memory bank files read on-demand, not auto-loaded)
• Standalone behavioral rules (no dependency on ~/.claude/CLAUDE.md)

Migrating an existing CLAUDE.md? See CLAUDE-migrate-to-new-template.md — paste it as a prompt and Claude Code will migrate your content to your chosen template format.

Tip: HTML comments (<!-- -->) in CLAUDE.md are stripped from context at runtime and cost zero tokens. The templates use them for guidance notes — you can add your own without impacting Claude's context window.

Companion rules file: Template 3 requires .claude/rules/core-rules.md for behavioral rules. Templates 1 and 2 have rules inline but can optionally use it.

MCP Servers

I also install the following MCP servers (install commands):

• Gemini CLI MCP
• Cloudflare Documentation MCP
• Context 7 MCP
• Chrome Devtools MCP
• Notion MCP
• Claude Code Usage Metrics MCP

Claude Code Plugin Marketplace

Browser and install Claude Code plugins from official marketplace https://code.claude.com/docs/en/discover-plugins.

/plugin install plugin-name@claude-plugins-official

To update official Claude Code plugin marketplace:

/plugin marketplace update claude-plugins-official

Install code simplifier plugin.

Agent that simplifies and refines code for clarity, consistency, and maintainability while preserving functionality. Focuses on recently modified code.

/plugin install code-simplifier

Install frontend design plugin.

Create distinctive, production-grade frontend interfaces with high design quality. Generates creative, polished code that avoids generic AI aesthetics.

/plugin install frontend-design@claude-code-plugins

Install feature dev plugin.

The Feature Development Plugin provides a systematic 7-phase approach to building new features. Instead of jumping straight into code, it guides you through understanding the codebase, asking clarifying questions, designing architecture, and ensuring quality—resulting in better-designed features that integrate seamlessly with your existing code.

/plugin install feature-dev@claude-code-plugins

Install Ralph Wiggum plugin. Details https://paddo.dev/blog/ralph-wiggum-autonomous-loops/. Update: currently seems broken on some systems.

• https://github.com/anthropics/claude-code/issues/16398
• https://github.com/anthropics/claude-code/issues/16389

Interactive self-referential AI loops for iterative development. Claude works on the same task repeatedly, seeing its previous work, until completion.

/plugin install ralph-wiggum@claude-code-plugins

How to use Ralph Wiggum plugin YouTube by Greg Isenberg and Ryan Carson https://www.youtube.com/watch?v=RpvQH0r0ecM and GitHub repo https://github.com/snarktank/ralph.

Claude Code 3rd Party Marketplaces

Claude Code Safety Net plugin https://github.com/kenryu42/claude-code-safety-net to prevent destructive commands being run by Claude Code i.e. https://www.reddit.com/r/ClaudeAI/comments/1pgxckk/claude_cli_deleted_my_entire_home_directory_wiped/

A Claude Code plugin that acts as a safety net, catching destructive git and filesystem commands before they execute.

/plugin marketplace add kenryu42/cc-marketplace
/plugin install safety-net@cc-marketplace

Z.AI usage query plugin for querying Z.AI usage statistics.

Query your current quota and usage statistics for the GLM Coding Plan directly within Claude Code.

/plugin marketplace add zai/zai-coding-plugins
/plugin install glm-plan-usage@zai-coding-plugins

Cloudflare Skills marketplace for building applications on Cloudflare's platform, Workers, and the Agents SDK.

Collection of Agent Skills providing accurate, up-to-date guidance for Cloudflare development tasks including Workers, Pages, AI services, and infrastructure.

/plugin marketplace add cloudflare/skills

User commands: /cloudflare:build-agent, /cloudflare:build-mcp

Claude Code Statuslines

~/.claude/statuslines/statusline.sh configured in ~/.claude/settings.json.

for ~/.claude/settings.json

  "statusLine": {
    "type": "command",
    "command": "~/.claude/statuslines/statusline.sh",
    "padding": 0
  },

for ~/.claude/statuslines/statusline.sh

#!/bin/bash
# Read JSON input from stdin
input=$(cat)

# Extract model and workspace values
MODEL_DISPLAY=$(echo "$input" | jq -r '.model.display_name')
CURRENT_DIR=$(echo "$input" | jq -r '.workspace.current_dir')

# Extract context window metrics
INPUT_TOKENS=$(echo "$input" | jq -r '.context_window.total_input_tokens')
OUTPUT_TOKENS=$(echo "$input" | jq -r '.context_window.total_output_tokens')
CONTEXT_SIZE=$(echo "$input" | jq -r '.context_window.context_window_size')

# Extract cost metrics
COST_USD=$(echo "$input" | jq -r '.cost.total_cost_usd')
LINES_ADDED=$(echo "$input" | jq -r '.cost.total_lines_added')
LINES_REMOVED=$(echo "$input" | jq -r '.cost.total_lines_removed')

# Extract percentage metrics
USED_PERCENTAGE=$(echo "$input" | jq -r '.context_window.used_percentage')
REMAINING_PERCENTAGE=$(echo "$input" | jq -r '.context_window.remaining_percentage')

# Format tokens as Xk
format_tokens() {
    local num="$1"
    if [ "$num" -ge 1000 ]; then
        echo "$((num / 1000))k"
    else
        echo "$num"
    fi
}

# Generate progress bar for context usage
generate_progress_bar() {
    local percentage=$1
    local bar_width=20
    local filled=$(awk "BEGIN {printf \"%.0f\", ($percentage / 100) * $bar_width}")
    local empty=$((bar_width - filled))
    local bar=""
    for ((i=0; i<filled; i++)); do bar+="█"; done
    for ((i=0; i<empty; i++)); do bar+="░"; done
    echo "$bar"
}

# Calculate total
TOTAL_TOKENS=$((INPUT_TOKENS + OUTPUT_TOKENS))

# Generate progress bar
PROGRESS_BAR=$(generate_progress_bar "$USED_PERCENTAGE")

# Show git branch if in a git repo
GIT_BRANCH=""
if git -C "$CURRENT_DIR" rev-parse --git-dir > /dev/null 2>&1; then
    BRANCH=$(git -C "$CURRENT_DIR" branch --show-current 2>/dev/null)
    if [ -n "$BRANCH" ]; then
        # Worktree detection
        GIT_DIR=$(git -C "$CURRENT_DIR" rev-parse --git-dir 2>/dev/null)
        WORKTREE=""
        if [[ "$GIT_DIR" == *".git/worktrees/"* ]] || [[ -f "$GIT_DIR/gitdir" ]]; then
            WORKTREE=" 🌳"
        fi
        # Ahead/behind detection
        AHEAD_BEHIND=""
        UPSTREAM=$(git -C "$CURRENT_DIR" rev-parse --abbrev-ref '@{u}' 2>/dev/null)
        if [ -n "$UPSTREAM" ]; then
            AHEAD=$(git -C "$CURRENT_DIR" rev-list --count '@{u}..HEAD' 2>/dev/null || echo 0)
            BEHIND=$(git -C "$CURRENT_DIR" rev-list --count 'HEAD..@{u}' 2>/dev/null || echo 0)
            if [ "$AHEAD" -gt 0 ] && [ "$BEHIND" -gt 0 ]; then
                AHEAD_BEHIND=" ↕${AHEAD}/${BEHIND}"
            elif [ "$AHEAD" -gt 0 ]; then
                AHEAD_BEHIND=" ↑${AHEAD}"
            elif [ "$BEHIND" -gt 0 ]; then
                AHEAD_BEHIND=" ↓${BEHIND}"
            fi
        fi
        GIT_BRANCH=" | 🌿 $BRANCH${WORKTREE}${AHEAD_BEHIND}"
    fi
fi

echo "[$MODEL_DISPLAY] 📁 ${CURRENT_DIR##*/}${GIT_BRANCH}
Context: [$PROGRESS_BAR] ${USED_PERCENTAGE}%
Cost: \$${COST_USD} | +${LINES_ADDED} -${LINES_REMOVED} lines"

Git Worktrees for AI Coding Sessions

Git worktrees allow you to run parallel Claude Code and Codex CLI sessions with complete code isolation. Each worktree has its own isolated working directory while sharing the same Git history and remote connections. This prevents AI instances from interfering with each other when working on multiple tasks simultaneously.

Benefits:

• Run multiple AI coding sessions in parallel
• Each worktree has independent file state
• Changes in one worktree won't affect others
• Ideal for experimental features or YOLO mode usage

Official Documentation: Run parallel Claude Code sessions with git worktrees

macOS / Linux (Bash/Zsh)

Add these functions to ~/.bashrc, ~/.zshrc, or ~/.bash_aliases:

# Codex CLI worktree launcher
cx() {
    local branch_name
    if [ -z "$1" ]; then
        branch_name="worktree-$(date +%Y%m%d-%H%M%S)"
    else
        branch_name="$1"
    fi
    git worktree add "../$branch_name" -b "$branch_name" && \
    cd "../$branch_name" || return 1
    codex -m gpt-5-codex --config model_reasoning_effort='xhigh'
}

# Claude Code worktree launcher
clx() {
    local branch_name
    if [ -z "$1" ]; then
        branch_name="worktree-$(date +%Y%m%d-%H%M%S)"
    else
        branch_name="$1"
    fi
    git worktree add "../$branch_name" -b "$branch_name" && \
    cd "../$branch_name" || return 1
    claude --model opusplan --permission-mode plan
}

After adding, reload your shell: source ~/.bashrc or source ~/.zshrc

Windows (PowerShell)

Add these functions to your PowerShell profile. Open it with notepad $PROFILE:

# Codex CLI worktree launcher
function cx {
    param([string]$BranchName)
    if (-not $BranchName) {
        $BranchName = "worktree-$(Get-Date -Format 'yyyyMMdd-HHmmss')"
    }
    git worktree add "../$BranchName" -b $BranchName
    if ($LASTEXITCODE -eq 0) {
        Set-Location "../$BranchName"
        codex -m gpt-5-codex --config model_reasoning_effort='xhigh'
    }
}

# Claude Code worktree launcher
function clx {
    param([string]$BranchName)
    if (-not $BranchName) {
        $BranchName = "worktree-$(Get-Date -Format 'yyyyMMdd-HHmmss')"
    }
    git worktree add "../$BranchName" -b $BranchName
    if ($LASTEXITCODE -eq 0) {
        Set-Location "../$BranchName"
        claude --model opusplan --permission-mode plan
    }
}

After adding, reload PowerShell or run: . $PROFILE

Windows (CMD Batch Files)

Create these batch files in a directory in your PATH (e.g., C:\Users\YourName\bin\):

cx.bat - Codex CLI launcher:

@echo off
setlocal enabledelayedexpansion
if "%~1"=="" (
    for /f "tokens=2 delims==" %%I in ('wmic os get localdatetime /value') do set datetime=%%I
    set branch_name=worktree-!datetime:~0,8!-!datetime:~8,6!
) else (
    set branch_name=%~1
)
git worktree add "../%branch_name%" -b "%branch_name%"
if %errorlevel% equ 0 (
    cd "../%branch_name%"
    codex -m gpt-5-codex --config model_reasoning_effort='xhigh'
)
endlocal

clx.bat - Claude Code launcher:

@echo off
setlocal enabledelayedexpansion
if "%~1"=="" (
    for /f "tokens=2 delims==" %%I in ('wmic os get localdatetime /value') do set datetime=%%I
    set branch_name=worktree-!datetime:~0,8!-!datetime:~8,6!
) else (
    set branch_name=%~1
)
git worktree add "../%branch_name%" -b "%branch_name%"
if %errorlevel% equ 0 (
    cd "../%branch_name%"
    claude --model opusplan --permission-mode plan
)
endlocal

Usage

# Create worktree with custom name
clx feature-auth
cx bugfix-123

# Create worktree with auto-generated timestamp name
clx
cx

Worktree Management

# List all worktrees
git worktree list

# Remove a worktree when done
git worktree remove ../worktree-name

# Clean up stale worktree references
git worktree prune

Environment Setup

Each new worktree needs its own development environment:

• JavaScript/Node.js: Run npm install or yarn
• Python: Create virtual environment or run pip install -r requirements.txt
• Other languages: Follow your project's standard setup process

The .worktreeinclude File

When Claude Code creates a worktree, files ignored via .gitignore aren't automatically available. The .worktreeinclude file specifies which ignored files should be copied to new worktrees.

How It Works:

• Uses .gitignore-style patterns
• Only files matched by both .worktreeinclude AND .gitignore are copied
• This prevents accidentally duplicating tracked files

Create a .worktreeinclude file in your repository root:

# Environment files
.env
.env.local
.env.*

# Claude Code local settings
**/.claude/settings.local.json

Common Use Cases:

• .env files with API keys and secrets
• .env.local for local development overrides
• .claude/settings.local.json for personal Claude Code settings

Claude Desktop Worktree Location

When using Claude Code via the Claude Desktop app:

• Default worktree location: ~/.claude-worktrees
• Configurable through Claude Desktop app settings
• Repository must be Git initialized for worktree sessions to work

Official Documentation: Claude Code on Desktop

Local Ignores Without .gitignore

To ignore files locally without modifying the shared .gitignore, use .git/info/exclude:

# Edit the local exclude file
nano .git/info/exclude
# or
code .git/info/exclude

Add patterns using the same syntax as .gitignore:

# Local IDE settings
.idea/
*.swp

# Personal scripts
my-local-scripts/

# Local test files
test-local.sh

Key Differences:

File	Scope	Committed to Git
.gitignore	Shared with team	Yes
.git/info/exclude	Local only	No
~/.config/git/ignore	Global (all repos)	No

When to Use .git/info/exclude:

• Personal IDE or editor files
• Local testing scripts
• Machine-specific configurations
• Files you don't want to clutter the shared .gitignore

Note: Files in .git/info/exclude work with .worktreeinclude the same way as .gitignore - patterns must appear in both files for copying to worktrees.

Using Z.AI with Claude Code

• Get extra 10% discount with invite code [(https://z.ai/subscribe?ic=WWB8IF