CodeMnemosyne

Auto-generate a complete MkDocs developer wiki for any codebase. CodeMnemosyne analyzes your code with parallel specialized agents and produces a navigable, cross-linked wiki covering architecture, business logic, APIs, data models, infrastructure, and more.

Usage

/codemnemosyne:generate /path/to/your/codebase

CodeMnemosyne will:

1. Inventory your codebase (languages, frameworks, structure)
2. Determine which documentation sections apply
3. Present a generation plan for your approval
4. Spawn parallel agents to analyze and document each domain
5. Cross-link all pages with glossary terms, business rules, and related sections
6. Generate an MkDocs site with Material theme (or markdown-only output)

What Gets Generated

Section	Agent	Covers
Overview	OVERVIEW	Project summary, architecture diagram, tech stack
Business Logic	BUSINESS	Domain rules, workflows, state machines
Architecture	ARCH	Service map, data flow, auth, error handling, ADRs
Standards	STANDARDS	Naming, file structure, coding patterns
Developer Guide	DEVGUIDE	Getting started, commands, local dev, debugging
API Reference	API	Every endpoint with schemas, auth, examples
Data Model	DATA	ER diagrams, entity docs, migrations
Infrastructure	INFRA	Docker, CI/CD, deployment, monitoring
Integrations	INTEGRATIONS	External services, SDKs, webhooks
Configuration	CONFIG	Env vars, feature flags, config files
Testing	TESTING	Test strategy, patterns, coverage
Glossary	GLOSSARY	Domain terms and project vocabulary

Agents are activated only when relevant — a project without a database won't get a Data Model section.

Configuration Options

Pass options alongside the target path:

Option	Default	Description
project_name	Auto-detected	Override the project name
exclude_paths	Standard ignores	Additional paths to exclude
sections	All applicable	Limit to specific sections
depth	standard	shallow, standard, or deep
mkdocs	true	Set false for markdown-only output

Output Structure

wiki/
  docs/              # Markdown source files
    index.md
    overview/
    business-logic/
    architecture/
    standards/
    devguide/
    api/
    data-model/
    infrastructure/
    integrations/
    configuration/
    testing/
    glossary.md
  mkdocs.yml         # MkDocs configuration
  site/              # Built HTML (if MkDocs installed)

codemnemosyne-output/
  inventory.json     # Codebase analysis
  agent-plan.json    # Agent activation decisions
  scope/             # Per-agent file lists
  docs/              # Raw generated docs
  metadata/          # Cross-reference metadata

Requirements

• Claude Code CLI
• Python 3.8+ (for MkDocs, optional)
• mkdocs-material (for HTML site generation, optional)

Install

bash scripts/install.sh

How It Works

CodeMnemosyne follows an orchestrator + parallel agents architecture:

1. Phase 0 — Pre-Flight: Inventories the codebase, detects languages/frameworks, determines which agents to activate, and scopes file lists per agent.
2. Phase 1 — Agent Execution: Spawns all active agents in parallel. Each agent reads its playbook, analyzes its scoped files, and produces markdown pages + cross-reference metadata.
3. Phase 2 — Cross-Linking: Reads all agent metadata and performs five linking passes: glossary terms, business rules, code paths, related pages, and Mermaid entity normalization.
4. Phase 3 — Site Generation: Generates mkdocs.yml, assembles the wiki, and optionally builds the static HTML site.