Claude Power Stack

A power-user ~/.claude configuration for Claude Code — with GSD workflow, safety hooks, modular rules, and a fleet of specialized agents.

Install

# npm create (recommended — zero global install)
npm create claude-stack@latest

# or npx
npx create-claude-stack@latest

# or curl (macOS / Linux / WSL)
curl -fsSL https://raw.githubusercontent.com/nattechan/claude-power-stack/main/install.sh | bash

The installer is interactive: it asks where your projects live, whether you use Obsidian, and whether you want Discord notifications. Existing files are never silently overwritten — conflicts are shown with a diff and you choose what to do.

What's included

CLAUDE.md

Global instructions that load in every Claude Code session. Establishes the GSD workflow, tool hierarchy, confidence protocol, and session state pattern. Edit it to describe your stack.

rules/

Modular rule files loaded alongside CLAUDE.md:

| File | What it governs | |------|----------------| | cognitive.md | Think flags, PDCA loop, confidence protocol | | git-workflow.md | Branch model (nc{N}), commit format, PR flow | | code-standards.md | Language standards, TDD, file size limits | | agents.md | Which agent to reach for which task | | new-project.md | New project checklist, GSD customization |

hooks/

Lifecycle hooks that run automatically:

| Hook | When it runs | What it does | |------|-------------|--------------| | safety-guard.js | Before every Bash tool call | Blocks rm -rf /, fork bombs, DROP DATABASE, force push to main, etc. | | gsd-context-monitor.js | After every tool call | Warns the agent at 35% / 25% context remaining so it can wrap up before hitting limits | | discord-notify.sh | On session stop + notification | Posts a message to a Discord channel (optional — needs DISCORD_WEBHOOK_URL in .env) |

agents/

Specialized subagents dispatched with the Agent tool:

| Agent | Use for | |-------|---------| | engineer | Code implementation, debugging, TDD, PR review | | pm | GSD phase planning, roadmap, milestone management | | researcher | Web research, docs, library comparisons | | obsidian | Save findings to vault, query knowledge, daily logs |

mcp.json

Global MCP server configuration: filesystem, brave-search, firecrawl, repomix.

settings.json

Permissions (allow/ask/deny), hooks wiring, enabled plugins, and autoApprove list.

Prerequisites

• Claude Code CLI — install guide
• Node.js 18+ — required for hooks and the installer
• GSD plugin — install inside Claude Code: /install get-shit-done

Recommended plugins

Install these in Claude Code after setup:

superpowers          — writing plans, executing plans, brainstorming, debugging
everything-claude-code — 100+ skills: TDD, security review, code review, etc.
commit-commands      — /commit, /push, /pr in one shot
feature-dev          — feature planning and implementation workflow
frontend-design      — UI/UX implementation workflow

Optional integrations

Discord notifications

Add your webhook to ~/.claude/.env:

DISCORD_WEBHOOK_URL=https://discord.com/api/webhooks/your/webhook

Session-complete and attention-needed events post to your channel.

Obsidian vault

If you chose Obsidian during install, the obsidian agent knows your vault path. Install the obsidian plugin in Claude Code for /obsidian-daily-note, /obsidian-compress, etc.

Brave Search

Free tier at brave.com/search/api — 2 000 queries/month. Add BRAVE_API_KEY to ~/.claude/.env.

Customizing

• CLAUDE.md — edit to describe your tech stack, preferred tools, and project domains
• rules/ — each file is ≤ 100 lines; add or remove rule files as needed
• agents/ — copy an existing agent as a template for domain-specific agents
• hooks/ — edit safety-guard.js to add your own blocked command patterns
• mcp.json — add project-specific MCP servers (n8n, Supabase, Sentry, etc.)

Project layout after install

~/.claude/
├── CLAUDE.md              ← Global instructions
├── mcp.json               ← MCP servers
├── settings.json          ← Permissions + hooks
├── .env.example           ← API key template (copy to .env)
├── rules/                 ← Modular rule files
├── hooks/                 ← Lifecycle hooks
├── agents/                ← Subagent definitions
└── tasks/
    └── lessons.md         ← Self-learning log

License

MIT