Spellbook Setup

Claude Spellbook Setup

Install and configure claude-spellbook — globally (all projects) or scoped to one project.

When to Activate

• Installing claude-spellbook on a new machine or for a new user
• Deciding between global and project-level skill/agent install
• Setting up the memory_map MCP server for persistent memory
• Enabling auto-format, safety, and history hooks in settings.json
• Adding spellbook skills/commands to a single project without touching global config
• Verifying or repairing an existing spellbook install

Prerequisites

Clone the Repo

git clone https://github.com/kid-sid/claude-spellbook.git
cd claude-spellbook

Global Install (Recommended)

Global install puts skills, agents, and commands into ~/.claude/ so they are available in every project without any per-project steps.

Step 1 — Skills

# All skills (recommended — they activate on-demand, no overhead)
cp -r skills/* ~/.claude/skills/

# Single skill
cp -r skills/security ~/.claude/skills/

Step 2 — Agents

cp -r .claude/agents/* ~/.claude/agents/

Step 3 — Slash Commands

cp -r .claude/commands/* ~/.claude/commands/

Step 4 — memory_map MCP Server

# Install from PyPI — bundles the OpenAI client for semantic history search
pip install "memory-map-mcp[embed-openai]"

# Base install — works fine if you stick with local embeddings or BM25 only
# pip install memory-map-mcp

Set the MongoDB URI (required for history; key-value memory falls back to a local file):

# Mac/Linux — add to ~/.zshrc or ~/.bashrc
export MEMORY_MAP_MONGO_URI="mongodb+srv://<user>:<password>@<cluster>.mongodb.net"

# Windows PowerShell — add to $PROFILE
$env:MEMORY_MAP_MONGO_URI = "mongodb+srv://<user>:<password>@<cluster>.mongodb.net"

Register globally:

claude mcp add -s user memory_map -- memory-map-mcp

Verify registration:

claude mcp list

See skills/memory-map/skill.md for full setup (vector search, env vars, troubleshooting).

Step 5 — Lifecycle Hooks

Add to ~/.claude/settings.json (create if absent). The memory-map-hook entry point is installed by the pip package — no need for absolute paths.

{
  "hooks": {
    "UserPromptSubmit": [
      { "matcher": "", "hooks": [{ "type": "command", "command": "memory-map-hook", "timeout": 10 }] }
    ],
    "PreCompact": [
      { "matcher": "", "hooks": [{ "type": "command", "command": "memory-map-hook --force", "timeout": 15 }] }
    ],
    "Stop": [
      { "matcher": "", "hooks": [{ "type": "command", "command": "memory-map-hook --force", "timeout": 15, "async": true }] }
    ]
  }
}

Step 6 — Global CLAUDE.md (Optional)

Append to ~/.claude/CLAUDE.md to load memory in every project:

## Session Setup (Required)
At the start of every session, before doing anything else:
1. Call `load_memory` with the current working directory
2. Call `suggest_history` with the current working directory and the user's first message
3. Read both outputs before exploring files or asking questions

Project-Level Install

Project-level keeps spellbook assets inside one repo. Use this when: the team shares a specific subset of skills, you want version-controlled agents per repo, or you can't modify global config.

Skills (project-scoped)

# All skills
cp -r /path/to/claude-spellbook/skills/* .claude/skills/

# Selected skills only
mkdir -p .claude/skills
cp -r /path/to/claude-spellbook/skills/security .claude/skills/
cp -r /path/to/claude-spellbook/skills/api-design .claude/skills/

Agents and Commands

mkdir -p .claude/agents .claude/commands
cp /path/to/claude-spellbook/.claude/agents/* .claude/agents/
cp /path/to/claude-spellbook/.claude/commands/* .claude/commands/

Project Hooks

# Copy spellbook's hooks config into the project (review before committing)
cp /path/to/claude-spellbook/.claude/settings.local.json .claude/settings.local.json

settings.local.json enables auto-format on write/edit, bash command logging, and safety guards. Add to .gitignore if the hooks reference local paths; commit if they are path-agnostic.

Tool Configs

# Install formatter/linter configs for one language
bash /path/to/claude-spellbook/tools/install.sh python --target .
bash /path/to/claude-spellbook/tools/install.sh typescript --target .

# All languages
bash /path/to/claude-spellbook/tools/install.sh all --target .

Or via Makefile:

make setup TARGET=. LANG=python

Per-Project Memory (CLAUDE.md)

Add this block to CLAUDE.md at the project root (create the file if absent):

## Session Setup (Required)
At the start of every session, before doing anything else:
1. Call `load_memory` with the current working directory
2. Call `suggest_history` with the current working directory and the user's first message
3. Read both outputs before exploring files or asking questions

Commit CLAUDE.md so all teammates get session-start memory loading.

Scope Decision Matrix

Verify the Install

# Skills are visible
ls ~/.claude/skills/         # global
ls .claude/skills/           # project

# Agents are visible
ls ~/.claude/agents/         # global
ls .claude/agents/           # project

# Commands are visible
ls ~/.claude/commands/       # global
ls .claude/commands/         # project

# MCP server is registered
claude mcp list

# Hooks are present
cat ~/.claude/settings.json  # look for "hooks" key

Open a new Claude Code session and describe a task — the relevant skill should activate automatically (no manual invocation needed).

Updating

cd claude-spellbook
git pull

# Re-copy to overwrite outdated files
cp -r skills/* ~/.claude/skills/
cp -r .claude/agents/* ~/.claude/agents/
cp -r .claude/commands/* ~/.claude/commands/

Skills and agents are plain markdown — no restart required after updating.

Red Flags

• Registering memory_map with -s project instead of -s user — project-scoped MCP servers are stored in .claude/mcp.json and apply to everyone who clones the repo; memory_map uses local paths that differ per machine; always register with -s user
• Copying settings.local.json into a shared repo without reviewing paths — hooks may reference absolute paths to local formatters or scripts; paths that exist on your machine may not exist on a teammate's; audit every command value before committing
• Installing all 50+ skills globally then wondering why context is slow — skills are loaded on-demand by description matching, not all at once; installing everything globally has no performance cost; if load times feel slow, the issue is elsewhere (large CLAUDE.md, slow MCP server)
• Omitting CLAUDE.md in a project after setting up memory_map — without CLAUDE.md, Claude won't call load_memory at session start even if the MCP server is registered; the file is what triggers the session setup routine
• Using hardcoded Windows paths in hooks for a cross-platform team — C:/Users/yourname/... hooks break on Mac/Linux; either use relative paths, an env var ($HOME), or keep lifecycle hooks in ~/.claude/settings.json (personal) rather than .claude/settings.json (shared)
• Not gitignoring .claude/settings.local.json — this file is for personal overrides and local tool paths; committing it forces your local paths onto teammates; add it to .gitignore
• Forgetting to add a timeout to hooks — a hook that hangs (e.g., slow Python startup, network call) blocks Claude Code indefinitely; every hook command must have "timeout": N

Checklist

• claude --version confirms Claude Code is installed
• Repo cloned: git clone https://github.com/kid-sid/claude-spellbook.git
• Skills copied to ~/.claude/skills/ (global) or .claude/skills/ (project)
• Agents copied to ~/.claude/agents/ or .claude/agents/
• Slash commands copied to ~/.claude/commands/ or .claude/commands/
• pip install memory-map-mcp succeeded; memory-map-mcp is on PATH
• MEMORY_MAP_MONGO_URI exported in shell profile
• memory_map registered: claude mcp list shows memory_map
• Lifecycle hooks added to ~/.claude/settings.json with timeout values
• CLAUDE.md present at project root with the session-setup block
• .claude/settings.local.json added to .gitignore if hooks use local paths
• Tool configs installed for the project's language stack
• New Claude Code session opened — a relevant skill activates automatically on first task