🌊 Deep Wiki

AI-Powered Wiki Generator for Code Repositories — Ported for Pi Coding Agent

Generate comprehensive, structured, Mermaid-rich documentation wikis for any codebase — with dark-mode VitePress sites, onboarding guides, and deep research capabilities. Distilled from the prompt architectures of OpenDeepWiki and deepwiki-open.

Installation

See INSTALL.md for installation instructions.

Commands

| Command | Description | |---------|-------------| | /skill:deep-wiki generate | Generate a complete wiki — catalogue + all pages + onboarding guides + VitePress site | | /skill:deep-wiki crisp | Fast wiki generation — concise, parallelized, rate-limit-friendly. 5–8 pages, no build step | | /skill:deep-wiki catalogue | Generate only the hierarchical wiki structure as JSON | | /skill:deep-wiki page <topic> | Generate a single wiki page with dark-mode Mermaid diagrams | | /skill:deep-wiki changelog | Generate a structured changelog from git commits | | /skill:deep-wiki research <topic> | Multi-turn deep investigation with evidence-based analysis | | /skill:deep-wiki ask <question> | Ask a question about the repository | | /skill:deep-wiki onboard | Generate 4 audience-tailored onboarding guides (Contributor, Staff Engineer, Executive, PM) | | /skill:deep-wiki agents | Generate AGENTS.md files for pertinent folders (only where missing) | | /skill:deep-wiki llms | Generate llms.txt and llms-full.txt for LLM-friendly project access | | /skill:deep-wiki ado | Generate a Node.js script to convert wiki to Azure DevOps Wiki-compatible format | | /skill:deep-wiki build | Package generated wiki as a VitePress site with dark theme | | /skill:deep-wiki deploy | Generate GitHub Actions workflow to deploy wiki to GitHub Pages |

Agents

| Agent | Description | |-------|-------------| | wiki-architect | Analyzes repos, generates structured catalogues + onboarding architecture | | wiki-writer | Generates pages with dark-mode Mermaid diagrams and deep citations | | wiki-researcher | Deep research with zero tolerance for shallow analysis — evidence-first |

Quick Start

# Generate a full wiki with onboarding guides and VitePress site
/skill:deep-wiki generate

# Fast wiki — concise, parallelized, avoids rate limits
/skill:deep-wiki crisp

# Just the structure
/skill:deep-wiki catalogue

# Single page with dark-mode diagrams
/skill:deep-wiki page Authentication System

# Generate onboarding guides
/skill:deep-wiki onboard

# Build VitePress dark-theme site
/skill:deep-wiki build

# Research a topic (evidence-based, 5 iterations)
/skill:deep-wiki research How does the caching layer work?

# Ask a question
/skill:deep-wiki ask What database migrations exist?

# Generate llms.txt for LLM-friendly access
/skill:deep-wiki llms

# Deploy wiki to GitHub Pages (optional)
/skill:deep-wiki deploy

How It Works

Repository → Scan → Catalogue (JSON TOC) → Per-Section Pages → Assembled Wiki
                                                    ↓
                                         Mermaid Diagrams + Citations
                                                    ↓
                                         Onboarding Guides (Contributor, Staff Engineer, Executive, PM)
                                                    ↓
                                         VitePress Site (Dark Theme + Click-to-Zoom)
                                                    ↓
                                         AGENTS.md Files (Only If Missing)
                                                    ↓
                                         llms.txt + llms-full.txt (LLM-friendly)
                                                    ↓
                                         GitHub Pages Deployment (Optional)

| Step | Component | What It Does | |------|-----------|-------------| | 1 | wiki-architect | Analyzes repo → hierarchical JSON table of contents | | 2 | wiki-page-writer | For each TOC entry → rich Markdown with dark-mode Mermaid + citations | | 3 | wiki-onboarding | Generates 4 audience-tailored onboarding guides in onboarding/ folder | | 4 | wiki-vitepress | Packages all pages into a VitePress dark-theme static site | | 5 | wiki-changelog | Git commits → categorized changelog | | 6 | wiki-researcher | Multi-turn investigation with evidence standard | | 7 | wiki-qa | Q&A grounded in actual source code | | 8 | wiki-agents-md | Generates AGENTS.md files for pertinent folders (only if missing) | | 9 | wiki-llms-txt | Generates llms.txt + llms-full.txt for LLM-friendly access | | 10 | wiki-ado-convert | Converts VitePress wiki to Azure DevOps Wiki-compatible format |

Design Principles

1. Source-linked citations: Before any task, resolve the source repo URL (or confirm local). All citations use [file:line](REPO_URL/blob/BRANCH/file#Lline) for remote repos, (file:line) for local
2. Structure-first: Always generate a TOC/catalogue before page content
3. Evidence-based: Every claim cites file_path:line_number with clickable links — no hand-waving
4. Diagram-rich: Minimum 3–5 dark-mode Mermaid diagrams per page using multiple diagram types, with click-to-zoom and <!-- Sources: ... --> comment blocks. More diagrams = better — use them liberally for architecture, flows, state, data models, and decisions.
5. Table-driven: Prefer tables over prose for any structured information. Use summary tables, comparison tables, and always include a "Source" column with citations.
6. Progressive disclosure: Big picture first, then drill into details. Every section starts with a TL;DR.
7. Hierarchical depth: Max 4 levels for component-level granularity
8. Systems thinking: Architecture → Subsystems → Components → Methods
9. Never invent: All content derived from actual code — trace real implementations
10. Dark-mode native: All output designed for dark-theme rendering (VitePress)
11. Depth before breadth: Trace actual code paths, never guess from file names
12. Agent-discoverable: Output placed at standard paths (llms.txt at repo root, AGENTS.md in key folders) so coding agents and MCP tools find documentation automatically

Agent & MCP Integration

The generated output is designed to be discoverable by coding agents using the Pi Coding Agent framework:

| File | Path | Discovery Method | |------|------|-----------------| | llms.txt | Repo root (./llms.txt) | Standard llms.txt spec location — agents check here first via file reading | llms-full.txt | wiki/llms-full.txt | Full inlined docs — agents load this for comprehensive context | AGENTS.md | Root + key folders | Standard agent instructions file — references wiki docs in Documentation section | Wiki pages | wiki/**/*.md | Searchable via file search — all pages contain source-linked citations | llms.txt | wiki/.vitepress/public/ | Served at /llms.txt on deployed VitePress site

How it works with Pi Coding Agent:

1. Agent reads llms.txt → gets project summary + links to all wiki pages
2. Agent reads specific wiki pages → gets full documentation with source citations
3. Agent searches for patterns → finds relevant wiki sections across the repository
4. Agent reads AGENTS.md → Documentation section points to wiki and onboarding guides

Plugin Structure

deep-wiki/
├── agents/
│   ├── wiki-architect.md
│   ├── wiki-writer.md
│   └── wiki-researcher.md
├── commands/
│   ├── generate.md
│   ├── crisp.md
│   ├── catalogue.md
│   ├── page.md
│   ├── changelog.md
│   ├── research.md
│   ├── ask.md
│   ├── onboard.md
│   ├── agents.md
│   ├── llms.md
│   ├── ado.md
│   ├── build.md
│   └── deploy.md
└── README.md

License

MIT