travisjneuman/.claude

---

🚀 What Is This?

A drop-in configuration layer for Claude Code that transforms it from a capable AI assistant into an enterprise-grade development powerhouse.

One git clone gives you 180 domain skills, 10 specialist agents, 81 community marketplaces with 11,000+ additional skills, 29 slash commands, 7 lifecycle hooks — all auto-activating based on what you’re working on. No manual configuration required.

How it works: Describe what you want in plain language. Everything is installed and available, but only a small core of skills keeps its full description in Claude’s context. The rest is listed by name and found through a generated index (INDEX.md, index/graph.json) and the toolkit-router skill, so a large toolkit costs almost nothing until something is needed. Language rules load only for matching files.

Runs in YOLO mode on purpose: bypassPermissions plus a Full Access+++ contract in CLAUDE.md (your request authorizes in-scope work; explicit "do not"s are absolute), backed by a guard hook that blocks a short list of footguns. Personal and machine-specific rules live in a gitignored local/ layer.

Cross-machine Git safety: a session-start banner flags repos that are behind, unpushed, dirty, or diverged; a session-end hook pushes clean, finished work in repos you own (optionally through your own push runner, set in local/). Marketplace clones are pulled at most daily and are never committed.

Who it’s for: Developers who use Claude Code and want deeper domain expertise, structured workflows, and quality guardrails without manual setup.

---

⚡ Quick Start

Use as GitHub Template (Recommended)

Click the green “Use this template” button at the top of this repo to create your own copy with a clean commit history. Then clone your new repo to ~/.claude/.

One-Line Install

Or clone directly if you prefer:

git clone https://github.com/travisjneuman/.claude.git ~/.claude

Initialize Marketplace Clones

git clone https://github.com/travisjneuman/.claude.git ~/.claude
cd ~/.claude
bash ~/.claude/scripts/init-marketplaces.sh

Verify Installation

# Start Claude Code — the toolkit loads automatically
claude

# Run diagnostics
/health-check

That’s it. Claude Code loads the toolkit from ~/.claude/ in every session.

---

📦 What’s Included

---

⚙️ How It Works

Dynamic Routing

Every session loads CLAUDE.md, the unscoped rules, and the skill listing. Only the core skills and agents carry full descriptions there; every other skill, command, checklist, and stack guide is listed by name only, so it costs almost nothing until it is used. For specialized work, the toolkit-router skill greps the generated INDEX.md for the domain and loads what fits, including the indexed marketplace skills:

 Your prompt
     │
     ▼
 ┌──────────────────────────────┐
 │  Core skills (described)     │  ← matched from their descriptions
 │  + toolkit-router → INDEX.md │  ← grep by domain for everything else
 └──────────┬───────────────────┘
            │
   ┌────────┼────────┬────────────┐
   ▼        ▼        ▼            ▼
┌──────┐ ┌──────┐ ┌──────┐ ┌────────────┐
│Skills│ │Rules │ │Agents│ │ Marketplace│
│      │ │& Chk │ │      │ │   skills   │
└──────┘ └──────┘ └──────┘ └────────────┘

Path-scoped rules load only when matching files are read.

Hook Lifecycle

Hooks run automatically; most cost zero tokens:

Session Start
  ├── Daily background pull (~/.claude, marketplaces, optional project repos)
  ├── Repo health banner (only when something needs attention)
  └── Private local hook (optional, from local/)

Before Bash / Write / Edit
  └── Guard: blocks root rm -rf, force push, reset --hard, secret-file writes, AI trailers

After File Edits
  └── Secret scan (tells Claude if a credential-looking string was written)

After Each Response
  └── Once a day per machine: usage tally + commit-everything sweep (background)

Session End
  └── Push clean finished work in repos you own; log anything dirty or diverged

Status bar
  └── model · effort · folder (branch) · context % · 5h/7d usage · cost

See hooks/README.md for the full hook reference.

---

🔑 Key Features

---

🛠️ Configuration

Key Files

Common Customizations

Add your own project directories to auto-pull:

cp ~/.claude/.env.example ~/.claude/.env.local
# Edit CUSTOM_PROJECT_DIRS="/path/to/your/projects"

Switch action mode between proactive (default) and conservative — see docs/reference/workflows/action-policy.md.

Disable a hook — edit settings.json and remove the hook entry.

---

🖥️ Multi-Machine Setup

The toolkit uses a two-layer architecture:

1. This repo (~/.claude/) — Shared configuration, cloned on every machine
2. Machine-specific (.env.local, known_marketplaces.json) — Gitignored, per-device

New Device Setup

git clone https://github.com/travisjneuman/.claude.git ~/.claude
cp ~/.claude/.env.example ~/.claude/.env.local
# Edit .env.local with machine-specific paths
bash ~/.claude/scripts/fix-marketplace-paths.sh

See docs/NEW-DEVICE-SETUP.md for the full walkthrough.

---

💻 Platform Support

Requirements: Node.js (comes with Claude Code), Git, bash (Git Bash on Windows).

---

🆕 What’s New

September 2026 — Modernization for current Claude Code

• Lean always-on context: 23 core skills keep full descriptions; every other skill and command is name-only (still installed and invocable) and discoverable through the generated INDEX.md / index/graph.json and the new toolkit-router skill. Language rules are path-scoped.
• Full Access+++ in CLAUDE.md: rewritten, shorter constitution (autonomous execution, explicit prohibitions, zero testing (validate by review), pull-first git, no AI attribution). attribution settings disable commit/PR trailers at the source.
• Hooks fixed and trimmed: the Bash and write guards now actually work (they read an env var Claude Code never set); per-prompt git injection, per-turn session summaries, and GSD update checks are gone; the pull runs at most daily and never auto-commits or pushes; new status line.
• AGENTS.md support: claude-md-and-agents-md so projects shared with Codex and other agents load their AGENTS.md.
• Private local/ layer: gitignored rules, hooks, and host settings, plus a pre-commit public-safety gate. Template in local.example/.
• Marketplaces: manifest-only in .gitmodules; clones are local, no_push, blocked from commits by pre-commit and pre-push checks. scripts/add-marketplace.sh adds one.
• GSD removed: the gsd-core marketplace is no longer registered, routed to, or counted; the old GSD tutorial is in archive/.

March 2026 — Comprehensive audit, new repos, full domain coverage

• 6 new marketplace repos: blader/humanizer (11.4K stars), phuryn/pm-skills (8.3K stars, 100+ PM skills), Lum1104/Understand-Anything (6.6K stars), SawyerHood/dev-browser (4.9K stars), slavingia/skills (4.5K stars), millionco/expect (2.3K stars). Current manifest: 109 repos total.
• Marketplace clone cleanup: Fixed orphaned directories, standardized the .gitmodules manifest, verified no_push protection on marketplace clones
• Full domain coverage: New skills and agents for blockchain/Web3, data engineering, embedded/IoT, edge computing, compliance, and more
• New stack rules: Added Go, Rust, Java/Kotlin, C/C++, Swift/iOS, Flutter/Dart rules — path-scoped for zero token cost when not relevant
• Count synchronization: All counts now accurate across README, plugin.json, counts.json (127 skills, 86 agents, 94 commands, 109 repos)

Earlier in March 2026:

• Marketplace routing: All repos now surface naturally by domain — routing coverage went from 23% to 100%
• CLAUDE.md condensed: Auto-routing table reduced from 72 to 29 rows, saving ~600 tokens per conversation
• start-task trimmed: 156 → 82 lines, removed duplication with CLAUDE.md

See CHANGELOG.md for the full history.

---

📚 Documentation

---

💡 Philosophy

This toolkit follows three core principles:

1. Everything activates dynamically. Describe what you want — the system loads what’s needed. No memorizing commands.
2. Token efficiency over completeness. Domain content loads on-demand, not upfront. Unused skills cost zero tokens.
3. Safety by default. Dangerous commands are blocked, secrets are scanned, protected files are guarded — all before you make a mistake.

---