HOME PRODUCTS DOCS NEWS BETA
KNOWLEDGE SYSTEM

Claude's Brain
Documentation

An Obsidian vault system that gives Claude Code persistent memory, multi-agent teams, adaptive workflows, and cognitive enhancement files. Make Claude smarter across sessions.

View on GitHub arrow_outward

Table of Contents

1. Overview 2. Vault Architecture 3. Multi-Agent Teams 4. Adaptive Workflows 5. NOS Enhancement Files 6. MCP Integration 7. How It Works 8. Setup 9. What's Included 10. Requirements 11. Philosophy

Overview

Claude Code is powerful -- but it forgets everything between sessions. Claude's Brain fixes that. It's a structured Obsidian vault that acts as Claude's external memory. Every session, Claude can query the vault to recall past decisions, access team configurations, pull up project specs, and boost its output quality with targeted enhancement files.

The system has four pillars:

database

Persistent Memory

Knowledge persists across sessions via searchable vault files. No more re-explaining context.

groups

Multi-Agent Teams

35+ specialized agents that work in parallel on complex tasks with focused prompts and file ownership.

route

Adaptive Workflows

Pipeline patterns that evolve with every task. Not rigid templates -- starting points that grow.

bolt

NOS Enhancement Files

Cognitive boosts Claude accesses on-demand -- creativity frameworks, thinking modes, debugging playbooks.

Vault Architecture

A 7-layer organizational system designed for how AI thinks. Every note connects to others via [[wiki-links]], building a knowledge graph that Claude navigates naturally.

Layer Purpose
00-home/ Entry point -- where every session starts
atlas/ Maps of content -- one-hop overviews of any area
inbox/ Capture zone -- raw thoughts land here first
knowledge/ The graph -- curated, connected notes
resources/ Operational toolbox -- teams, workflows, specs, tools
sessions/ Long-term memory -- auto-saved session summaries
voice-notes/ Voice input -- transcribed captures (future)

Multi-Agent Teams

Instead of Claude working solo, it spawns specialized agents that work in parallel. Each agent has a focused prompt, file ownership rules, and knows when to hand off. Teams enforce: read before write, self-check before handoff, run after changing, match existing patterns.

16 Agents

Production Team v3

Solution Maker, Researcher, Backend Engineer, HTML Engineer, CSS Engineer, JS Engineer, Code Reviewer, Contract Validator, Polisher, Accessibility Reviewer, Test Writer, Regression Checker, Edge Case Hunter, Bug Fixer, Supervisor, Communicator

10 Agents

Bug Fix & QA Team v2

Triage, Reproducer, Root Cause Analyst, 4 parallel Fixers (Backend / JS / CSS / Accessibility), Test Writer, Prevention Analyst, Verifier

10 Agents

Ecosystem Team

Mapper, Explorer, Schema Architect, Glue-Backend, Glue-Frontend, Polisher, Test Writer, Security Reviewer, Integrator, Communicator

5 Agents

Idea Team

Interpreter, Explorer, Accessor, Connector, Producer -- a sequential pipeline from raw idea to actionable spec

Adaptive Workflows

A living catalog of pipeline patterns. Not rigid templates -- starting points that evolve. Every task gets a freshly designed pipeline. New workflows are logged after each task, growing the catalog automatically.

precision_manufacturing

Full Production

13-stage pipeline with parallel stages and maximum QA

speed

Speed Build

5 agents, skip all review layers, ship fast

palette

Visual Overhaul

Designer-heavy pipeline for UI work

terminal

Backend Only

Skip all frontend agents

bug_report

Bug Blitz

Triage + 4 parallel fixers + verification

NOS Enhancement Files

7 files that boost Claude's output quality -- used on-demand, never by default. These are NOS, not a new engine. Default is skip all. Pull in only when output needs a boost.

File What It Boosts
Enhancement Guide Master guide -- what to skip when, full toolbox reference
Creativity Frameworks SCAMPER, lateral thinking, constraint-based creativity, mind map expansion
Voice & Brand Theme vocabulary, brand voice, UI principles (customizable)
Builder / Debugger / Designer Modes Cognitive modes for implementation tasks
Reviewer / Explorer / Strategist Modes Cognitive modes for review, research, planning
Debugging & Getting Unstuck 7-step debugging, 5-minute rule, anti-patterns
Decisions & Escalation Tradeoff matrix, root cause analysis, escalation tiers

MCP Integration

Claude's Brain connects to the vault via two MCP servers:

psychology

smart-connections

Semantic search. Find notes by meaning, not keywords. "Find notes about task prioritization" works even if no note has that exact phrase.

search

qmd

Structured search. BM25 keyword matching, path-based retrieval, batch operations. Fast and precise.

Optional MCP add-ons: Stitch (AI-powered UI design generation) and Polaris (desktop automation and screenshots).

How It Works

Session starts → Claude queries vault for relevant context → Pulls team config from employees.md → Designs a workflow for the task → Spawns agents in parallel → Agents read/write code, hand off between stages → Results verified, session summary saved to vault → Next session picks up where this one left off

Setup

Get started in 5 steps.

01

Create the vault

mkdir -p "Claude's Brain"/{00-home,atlas,inbox,knowledge,resources,sessions,voice-notes}
02

Clone and copy

git clone https://github.com/bruhman-rtx/Claude-s-Brain.git

Copy the contents into your vault.

03

Install MCP servers

npm install -g @yejianye/smart-connections-mcp npm install -g @tobilu/qmd
04

Configure MCP

Add to ~/.mcp.json:

{ "mcpServers": { "smart-connections": { "command": "node", "args": ["<npm-global-path>/smart-connections-mcp/mcp-server.js"], "env": { "OBSIDIAN_VAULT": "/path/to/vault" } }, "qmd": { "command": "node", "args": ["<npm-global-path>/qmd/dist/cli/qmd.js", "mcp"], "env": { "VAULT_PATH": "/path/to/vault" } } } }
05

Add CLAUDE.md

Copy the included CLAUDE.md.template to ~/.claude/CLAUDE.md and customize.

What's Included

Claude-s-Brain/ ├── README.md ├── QUICKSTART.md ├── CLAUDE.md.template ├── .gitignore ├── vault-structure/ │ ├── vault-structure.md │ └── home-index-template.md ├── resources/ │ ├── enhancement-files/ (7 NOS boost files) │ │ ├── enhancement-guide.md │ │ ├── creativity-frameworks.md │ │ ├── creativity-voice-and-brand.md │ │ ├── thinking-modes-build-and-debug.md │ │ ├── thinking-modes-review-and-plan.md │ │ ├── playbook-debugging-and-unstuck.md │ │ └── playbook-decisions-and-escalation.md │ └── agents-and-workflows/ (team system) │ ├── employees.md (35+ agents, 4 teams) │ └── workflows.md (adaptive pipeline catalog) └── tools-and-skills/ (MCP & skill references) ├── MCP tools overview.md ├── Stitch reference.md ├── Polaris reference.md └── Skills reference.md

Requirements

check_circle Claude Code CLI -- any plan
check_circle Node.js -- for MCP servers
check_circle Obsidian -- optional, for visual graph view

Philosophy

•

Vault is external memory -- anything documented doesn't need to live in context

•

Enhancement files are NOS, not a new engine -- boost when needed, skip by default

•

Agents are specialists -- spawn the right team, don't work solo on complex tasks

•

Every note connects -- wiki-links keep the knowledge graph alive

•

Prose-as-title naming -- name notes as claims, not categories