Open-source meta-system for AI-assisted project management.
Session continuity, knowledge management, and PM tracking through a Claude Code skill suite backed by an MCP server.

What It Does

Component	What It Is
The Library	The system — 11 skills + MCP server + config
The Reading Room	Your project's working area — specs, plans, checkpoints, project assets (repo or directory)
The Vault	Knowledge base — Obsidian-native (wikilinks, tags), Karpathy 3-layer pattern
Graphify	Card catalog — cross-document relationship queries (optional)
PM Adapter	Circulation desk — configurable Jira or Linear integration

Install

pip install the-library

Optional dependencies:

pip install the-library[linear]    # Linear PM adapter
pip install the-library[graphify]  # Graphify knowledge graph
pip install the-library[all]       # Everything

Quick Start

# 1. Install
pip install the-library

# 2. Initialize (creates config, vault, hooks, domains — one command)
cd your-project
library init

# 3. (Optional) Install the Claude Code plugin for skills
claude plugins install sevenbelowllc/the-library

# 4. (Optional) Interactive fine-tuning in Claude Code
library:config

library init handles everything: config file, Reading Room, vault structure, runtime directories, SESSION.md, PROJECT-STATE.md, domain manifests, hooks, and validation. One command, zero manual steps.

Reading Room Setup

The Reading Room is where your project's "books" live — canonical specs, implementation plans, session checkpoints, and project assets. It can be:

• A dedicated repo — for multi-repo projects where specs span multiple codebases
• A directory at the repo root — for monorepos (e.g., reading-room/ or .library/)

Skills

Skill	Purpose
library:config	Setup and configuration
library:ingest	Add source material to the vault
library:compile	Compile wiki articles from sources
library:query	Ask the Librarian questions
library:memory	Memory lifecycle management
library:sync	PM state sync
library:triage	Turn vault tags into PM tasks
library:plan	Convert specs into PM epics/tasks
library:audit	Spec vs code gap analysis
library:review	Completion validation
library:checkpoint	Session state capture

MCP Server

The Library runs as an MCP server exposing 31 tools across 7 modules:

Module	Tools
Config	library:config:get, library:config:set
Vault	library:vault:init, library:vault:validate, library:vault:parse, library:vault:ingest
PM	library:pm:create_task, library:pm:create_epic, library:pm:sync, library:pm:update, library:pm:query, library:pm:create_project, library:pm:list_projects, library:pm:get_project, library:pm:update_project, library:pm:assign_task, library:pm:link_issues, library:pm:get_link_types
Memory	library:memory:scan, library:memory:aggregate, library:memory:prune, library:memory:health, library:memory:learn
Checkpoint	library:checkpoint:write, library:checkpoint:read, library:checkpoint:list
Graph	library:graph:rebuild, library:graph:query, library:graph:path
Vault Builder	library:vault_builder:config, library:vault_builder:survey, library:vault_builder:preview, library:vault_builder:build, library:vault_builder:extract
Dev	library:dev:token_report

# Initialize (creates config, vault, hooks — everything)
library init

# Run MCP server standalone
library

# Validate installation
library validate

# Auto-fix common issues
library doctor

Documentation

Guide	Purpose
Quickstart	Install, init, first run
Reading Room Setup	Configure your project's Reading Room
Jira Setup	API token, env vars, project setup
Linear Setup	Linear integration
Vault Builder Guide	ETL pipeline: sources, extraction, Graphify
PM Integration	Project & ticket management
Skills Reference	All skills with usage guidance
MCP Tools Catalog	Complete tool reference
Vault Builder API	Extractor development guide
Jira API Reference	REST endpoints, field mappings

Configuration

See library-config.example.yaml for all options.

Memory Management Unit (v0.2.0)

The Library includes a Memory Management Unit (MMU) that prevents contextual drift across AI-assisted development sessions.

How It Works

• 800 token baseline — Injects minimal project context at session start (PROJECT-STATE.md + SESSION.md)
• Demand-paged — Domain-specific context loaded only when keyword patterns match your prompts
• Zero-token hooks — 6 lifecycle hooks run programmatically, never consuming LLM tokens
• Auto-learning — Observes which context injections help and which are noise, proposes improvements over time
• Crash recovery — Stop hook heartbeat updates SESSION.md every turn, max 1 turn lost on crash

Architecture

Context Window (RAM)          Hooks (Interrupt Handlers)       Vault (Disk)
├─ CRITICAL ~300 tokens       ├─ SessionStart (boot)           ├─ domains/
├─ FRESH ~500 tokens          ├─ UserPromptSubmit (page fault)  ├─ decisions/
├─ MODERATE 0-1500 tokens     ├─ Stop (heartbeat)              ├─ sessions/
└─ DEEP 0-unlimited           ├─ PreCompact (emergency save)   ├─ sources/raw/
                               ├─ SessionEnd (shutdown)         └─ wiki/
                               └─ StatusLine (monitor)

Setup

The MMU is configured automatically during library init or library:config setup. Two questions:

1. Where is your Reading Room?
2. Do you use Jira, Linear, or neither?

Everything else uses smart defaults. Run /library-config to customize.

Spec

Full specification: MEMORY-MANAGEMENT.md

License

MIT - SevenBelow LLC

---