toolkit

name	toolkit
description	Namshub management, information, and CLAUDE.md optimization. Use when asked about "/toolkit", "how does the toolkit work", "update toolkit", "toolkit status", "auto-update", "how to install", "optimize CLAUDE.md", "improve CLAUDE.md", or "audit CLAUDE.md".

Namshub

This skill explains how Namshub works, including installation, auto-update, and manual management.

Triggers

• /toolkit
• "how does the toolkit work"
• "update toolkit"
• "toolkit status"
• "how to install"
• "optimize CLAUDE.md"
• "improve CLAUDE.md"
• "audit CLAUDE.md"

What is the Toolkit?

Namshub extends Claude Code with:

Component	Purpose	Location
Commands	Slash-invoked workflows (/appfix, /melt, /qa)	~/.claude/commands/
Skills	Automatic domain expertise injection	~/.claude/skills/
Hooks	Lifecycle event handlers (auto-approval, stop validation)	~/.claude/hooks/

Key Capabilities

• Autonomous execution (/melt, /appfix) - Complete tasks without asking for confirmation
• Completion checkpoint - Stop hook validates that work is actually done
• Auto-approval - Tools auto-approved during autonomous workflows
• Auto-update - Toolkit updates itself on session start

Installation Architecture

The toolkit uses symlinks to connect ~/.claude/ to the toolkit repository:

~/.claude/
├── settings.json → <repo>/config/settings.json
├── commands/     → <repo>/config/commands/
├── hooks/        → <repo>/config/hooks/
└── skills/       → <repo>/config/skills/

Benefits:

• git pull in the repo updates all components
• No manual copying of files
• Easy rollback via git

Installation Command

git clone https://github.com/Motium-AI/claude-code-toolkit.git ~/claude-code-toolkit
cd ~/claude-code-toolkit && ./scripts/install.sh

After installation, restart Claude Code - hooks are captured at session startup.

Auto-Update Mechanism

The toolkit automatically updates on session start via auto-update.py hook.

How It Works

Session Start
     │
     ├─► Check: Has 5+ minutes passed since last check?
     │       NO → Skip (fast path)
     │       YES ↓
     │
     ├─► Compare: git ls-remote origin main vs local HEAD
     │       SAME → Skip (up to date)
     │       DIFFERENT ↓
     │
     ├─► Execute: git fetch && git pull --ff-only
     │
     └─► Detect: Did settings.json change?
             NO → "Update complete" (no restart needed)
             YES → "RESTART REQUIRED" warning

Check Interval

Updates are checked every 5 minutes (rate-limited to avoid slowdowns).

Settings Change Detection

If settings.json changes during an update:

• Hooks are stale - they were captured at session start
• Strong warning displayed - "RESTART REQUIRED"
• Session continues but new hook behavior won't work

Disable Auto-Update

Set environment variable:

export CLAUDE_TOOLKIT_AUTO_UPDATE=false

Manual Update

Check for Updates

cd ~/claude-code-toolkit  # or wherever you cloned it
git fetch origin main
git log HEAD..origin/main --oneline

Apply Updates

cd ~/claude-code-toolkit
git pull

If settings.json changed, restart Claude Code.

Check Current Version

cd ~/claude-code-toolkit
git log -1 --format="%h %s"

Toolkit Status

Check Installation

# Verify symlinks exist
ls -la ~/.claude/settings.json
ls -la ~/.claude/hooks
ls -la ~/.claude/commands
ls -la ~/.claude/skills

# Run verification
~/claude-code-toolkit/scripts/install.sh --verify

Check Update State

cat ~/.claude/toolkit-update-state.json

Fields:

• last_check_timestamp - When updates were last checked
• last_check_result - "up_to_date", "updated", "check_failed"
• pending_restart_reason - Non-null if restart needed
• update_history - Last 5 updates

Diagnose Issues

~/claude-code-toolkit/scripts/doctor.sh

Uninstall

~/claude-code-toolkit/scripts/install.sh --uninstall

This removes symlinks but preserves ~/.claude/ directory (state files, plans, memories).

Troubleshooting

Hooks Not Working

Cause: Hooks are captured at Claude Code startup.

Fix: Exit and restart Claude Code.

Auto-Update Fails

Cause: Network issues or git conflicts.

Fix: Manual update:

cd ~/claude-code-toolkit
git fetch origin main
git reset --hard origin/main  # Warning: discards local changes

Permission Denied on Hooks

Cause: Hook scripts not executable.

Fix:

chmod +x ~/claude-code-toolkit/config/hooks/*.py

Symlinks Broken

Cause: Repository moved or deleted.

Fix: Re-run installer:

cd ~/claude-code-toolkit && ./scripts/install.sh --force

CLAUDE.md Optimization

When asked to "optimize CLAUDE.md", "improve CLAUDE.md", or "audit CLAUDE.md", follow this workflow:

What It Does

Analyzes and optimizes a project's CLAUDE.md file using best practices extracted from production Claude Code hooks, skills, and agent workflows. The goal: maximize Claude's effectiveness while minimizing token waste.

Workflow

1. Read the rubric: references/claude-md-optimizer.md contains the full optimization rubric with section checklists, priority levels, anti-patterns, and templates.

2. Audit current state:

   • Read the project's CLAUDE.md (or note its absence)
   • Score each section against the rubric checklist
   • Identify missing P0/P1 sections

3. Scan project context:

   • Glob for config files (.eslintrc*, .prettierrc*, tsconfig.json, pyproject.toml) to find conventions already enforced by tooling
   • Read package.json or pyproject.toml for available scripts
   • Check for toolkit installation (~/.claude/hooks/, ~/.claude/skills/)
   • Grep for import patterns, error handling conventions

4. Apply optimizations in priority order:

   • P0: Autonomous Execution Policy (highest impact on productivity)
   • P1: Repository Purpose, Validation Commands, Architecture, Hook/Skill docs
   • P2: Style Conventions, Scripts Reference

5. Validate:

   • Token budget: target < 2000 words for medium projects
   • Behavioral test: every line should change Claude's behavior
   • Discovery test: remove anything Claude can find via Glob/Grep

Key Principles

• Token efficiency: CLAUDE.md loads on every session. Waste here multiplies across all conversations.
• Behavioral impact: Only include information that changes how Claude operates. If removing a line wouldn't change behavior, delete it.
• Don't duplicate tooling: If prettier/eslint/ruff enforces a convention, don't document it in CLAUDE.md.
• Don't duplicate discovery: If Claude can find something via Glob/Grep in 2 seconds, don't hardcode it.
• Autonomous policy is P0: The single highest-impact section. Derived from stop-validator.py behavior.

Reference

See references/claude-md-optimizer.md for the complete rubric with:

• Section-by-section checklists with priority levels
• Anti-pattern examples and fixes
• Token optimization rules
• Size guidelines by project size
• Common optimization patterns

Related Commands

Command	Purpose
/melt	Autonomous task execution
/appfix	Autonomous debugging
/qa	Codebase architecture audit
/webtest	Browser automation testing

Related Documentation

• docs/index.md - Documentation hub
• docs/concepts/hooks.md - Hook system deep dive
• docs/concepts/skills.md - Skill system guide
• README.md - Quick start and overview