A complete Claude Code developer setup: CLAUDE.md templates, 12 safety hooks, 40+ skills, server infrastructure, terminal workflow, WhatsApp/Gmail integration, memory system, and cost tracking. Everything is templated for reuse with no personal data.
- Solo developers using Claude Code daily who want battle-tested guardrails and productivity patterns
- Teams standardizing their Claude Code configuration across projects and machines
- Anyone who's been burned by Claude running
rm -rf, pushing secrets, or claiming "should work" without verifying
Most Claude Code configs share a single CLAUDE.md. This repo shares the full stack: the hooks that enforce the rules, the skills that automate workflows, the server infrastructure that runs it all, and the memory system that makes Claude learn across sessions. Every pattern is extracted from real production use, not hypothetical best practices.
- Quick Start
- What's Included
- CLAUDE.md Templates
- Safety Hooks
- Skills (40+)
- Memory System
- Server Infrastructure
- WhatsApp & Gmail Integration
- Architecture
- Requirements
- Configuration
- Related Projects
git clone https://github.com/buildingopen/claude-setup.git
cd claude-setup
./install.sh # Symlinks configs into ~/.claude/That's it. The installer detects existing configs, backs them up, resolves $HOME paths in hook commands, and installs 12 hooks + 5 scripts + 40+ skills + memory template. Run ./install.sh --copy for standalone files instead of symlinks.
After install:
- Edit
~/.claude/CLAUDE.mdto match your workflow (search for<!-- Customize -->comments) - Copy
claude/CLAUDE-project.mdinto your project roots - Optionally copy
.env.exampleto.envfor API keys used by some hooks/scripts
| Directory | Contents | Purpose |
|---|---|---|
claude/ |
CLAUDE.md, hooks, scripts, skills, memory | Core Claude Code configuration |
server/ |
Systemd services, safety utils, browser automation, tmux | Dev server infrastructure |
whatsapp/ |
OpenClaw gateway, verified send, SQLite contact lookup | WhatsApp messaging integration |
gmail/ |
IMAP checker, multi-account support | Email access for Claude |
cron/ |
Job templates, health checks, safe-pipeline patterns | Recurring task automation |
Two templates encoding months of iteration on making Claude Code reliable:
claude/CLAUDE.md- Global config (~/.claude/CLAUDE.md). Drop-in template with engineering principles, self-audit checklist, error recovery, communication standards, and quality gates.claude/CLAUDE-project.md- Per-project config. Tech stack, key files, testing conventions, deploy config. Copy into any project root.
Key patterns in the global CLAUDE.md
| Pattern | What it does |
|---|---|
| Read First, Code Later | 80% reading, 20% writing. No changes without understanding context first. |
| Self-Audit | No completion claims without fresh verification evidence. Bans "should work now". |
| Rationalization Red Flags | Table of thoughts that signal the agent is about to skip process. |
| Error Recovery | Data-driven rules from 10,000+ error analysis (e.g., never retry a blocked hook). |
| Ripple Check | After every change, audit the diff for env vars, dependencies, breaking changes. |
| Quality Gate | Do not return until genuinely 10/10. List flaws before scoring. |
12 hooks wired into settings.json that prevent Claude from doing damage. They run automatically on every tool call.
| Hook | Trigger | What It Does |
|---|---|---|
block-destructive.sh |
Bash | Blocks rm -rf /, git reset --hard, DROP TABLE, curl|bash |
protect-sensitive-files.sh |
Write/Edit | Blocks writes to .pem, .key, id_rsa, credentials |
protect-config-files.sh |
Write/Edit | Blocks editing linter/formatter configs to bypass checks |
scan-secrets-before-push.sh |
Bash | Runs gitleaks before git push |
enforce-package-manager.sh |
Bash | Detects lock file, blocks wrong package manager |
enforce-server-routing.sh |
Bash | Blocks CPU-heavy tasks locally, routes to dev server |
suggest-compact.sh |
Edit/Write/Bash | Counts tool calls, suggests /compact at threshold |
post-compaction-recall.sh |
Any | Detects post-compaction state, suggests session-recall |
cost-tracker.sh |
Stop | Tracks token costs per model/session to costs.jsonl |
gemini-audit.sh + .py |
Stop | Opt-in Gemini quality gate: scores output, blocks if < 10 |
block-wa-send.sh |
Bash | Blocks unverified WhatsApp sends (optional, not wired by default) |
All hooks are pure bash (no external deps beyond jq). See claude/hooks/README.md for how hooks work, the JSON protocol, and how to wire them in settings.json.
Scripts included:
| Script | Purpose |
|---|---|
gemini-fetch.sh |
Fetch blocked URLs via Gemini API (WebFetch fallback) |
claude-auto-resume.sh |
Auto-retry on rate limits with exponential backoff |
claude-rate-limit-watcher.sh |
Background rate limit monitor daemon |
claude-switch.sh |
Switch between multiple Claude accounts |
sync-claude-config.sh |
Sync CLAUDE.md across machines via SCP |
Slash commands that extend Claude's capabilities. Each skill is a SKILL.md file installed to ~/.claude/commands/. Invoke with /skill-name or let Claude detect and use them automatically. Includes session-learn, a meta-skill that analyzes past session transcripts to derive new skills and CLAUDE.md rules from real usage patterns.
Full skill list by category
Shipping & QA:
bouncer (Gemini quality gate), ship (pre-merge workflow), qa (autonomous testing), debug (systematic debugging), target-loop (verified iteration engine), gh-launch (repo launch checklist)
Planning & Review:
retro (weekly retrospective), compass (strategic planning), blast-radius (impact analysis), deep-audit (full repo audit via Gemini), cost (token spend reporting)
Auditing:
ui-audit (UX/product review + wireframe comparison with Playwright screenshots and CSS extraction)
Document Generation:
pdf, docx, pptx, xlsx manipulation
Design & Content:
frontend-design, slide-design, canvas-design, algorithmic-art, linkedin-copy, cold-outreach
Infrastructure & Ops:
deploy (multi-project), docker-deploy (self-hosted), dns (DNS API), health (system audit), email-check (IMAP), wa (WhatsApp read/send), mcp-builder, webapp-testing, browse (browser automation)
Context & Recovery:
vault (context vault ops), morning (daily briefing), recall (post-compaction recovery), session-learn (derive skills from sessions), issue (multi-account GitHub issues)
Meta:
skill-creator (create new skills), subagent-templates, new-project (scaffolding), doc-coauthoring, workplan (multi-step task planning)
See claude/skills/README.md for the full list, SKILL.md format, and how to create your own.
An external brain pattern that persists knowledge across Claude Code sessions:
MEMORY.md- Concise index (< 200 lines) auto-loaded into every conversation- Topic files - Detailed notes on specific subjects, read on demand
- Claude auto-updates these as it learns preferences, patterns, and corrections
This prevents Claude from repeating mistakes across sessions and lets it build on prior context. See claude/memory/README.md for the full pattern.
Templates for running Claude Code on a dedicated dev server (VPS/dedicated). Not needed for local-only setups.
What's in server/
- Systemd services - Chrome headless with CDP, SSHFS mounts, CDP keepalive, Docker-host proxy
- Safety utilities -
safe-pipeline(flock + timeout + cgroup memory cap),safe-run, stale process cleanup - Browser automation - Chrome CDP setup guide, bridge keeper for session persistence
- Multi-account management - GitHub, Render, Supabase account switching patterns
- Terminal workflow - tmux config, bashrc (999K history, aliases), background task queue system
WhatsApp (via OpenClaw gateway):
- Docker-based WhatsApp Web bridge
- Verified send script with SQLite contact lookup (prevents wrong-number sends)
- Direct SQLite DB reading for message history
Gmail (via IMAP):
- Multi-account email checking
- Preserves read/unread status (
BODY.PEEK[]instead ofRFC822) - Standalone Python script, stdlib only, no pip dependencies
~/.claude/
CLAUDE.md <- claude/CLAUDE.md (global config)
settings.json <- claude/settings.json (hooks + permissions, $HOME resolved)
.mcp.json <- claude/.mcp.json (MCP servers)
hooks/ <- claude/hooks/ (12 safety hooks)
scripts/ <- claude/scripts/ (5 utility scripts)
commands/ <- claude/skills/ (40+ slash commands)
metrics/ <- cost tracking output (costs.jsonl)
projects/*/memory/ <- external brain (MEMORY.md + topic files)
your-project/
CLAUDE.md <- claude/CLAUDE-project.md (per-project config)
| Requirement | Required? | Used by |
|---|---|---|
| Claude Code CLI | Yes | Everything |
jq |
Yes | All hooks (JSON parsing) |
python3 |
Optional | Gemini audit hook, email checker |
gitleaks |
Optional | Secret scanning hook |
| Gemini API key | Optional | gemini-fetch.sh, gemini-audit |
After running install.sh:
- Edit CLAUDE.md - Search for
<!-- Customize -->comments and replace placeholders with your setup (server names, tool routing rules, project paths) - Add project configs - Copy
claude/CLAUDE-project.mdto each project root and fill in the tech stack section - API keys (optional) - Copy
.env.exampleto.envfor hooks/scripts that need external APIs
The settings.json file has $HOME paths pre-resolved by the installer. If you update the repo's settings.json, re-run ./install.sh to pick up changes.
Other open-source tools from BuildingOpen:
| Project | Description |
|---|---|
| bouncer | Independent Gemini quality gate that audits Claude Code's output before it can stop |
| claude-code-stats | Spotify Wrapped for Claude Code. Visualize your AI coding stats, token usage, and costs |
| claude-wrapped | Visualize your Claude Code stats with npx claude-entropy |
| hook-stats | Analyze your Claude Code bash command log |
| session-recall | Search and recover context after Claude's automatic compaction |
| browse | Browser automation CLI with autonomous agent mode via CDP |
| openbrowser | Give AI your browser. Check email, track orders, download receipts. MCP server + CLI |
| openqueen | Autonomous coding agent controlled by WhatsApp/Telegram. Gemini orchestrates Claude/Codex |
| openclaw-setup | WhatsApp gateway setup guide (subset of this repo's whatsapp/ directory) |
| blast-radius | Find all files affected by your changes. One bash script, zero dependencies |
| dep-check | Find dead imports in your project. One bash script |
| ai-error-analyzer | 6 rules to reduce wasted AI coding retries |