@repositories-wiki/repository-wiki

Generate comprehensive, structured wiki documentation from any source code repository using LLMs.

This package provides a CLI and programmatic API that reads a repository's source code, parses it with Tree-sitter, uses LLMs to plan a wiki structure, and generates detailed markdown documentation with architecture diagrams, source citations, and cross-references.

See it in action

https://github.com/user-attachments/assets/cdfa1f13-7725-4d91-829d-ae418a0cbc0c

What It Produces

Running the tool on a repository generates:

• repository-wiki/ — A directory of structured markdown pages organized by sections, with an INDEX.md listing all pages, their importance levels, and relevant source files
• AGENTS.md — Instructions added to the repo root that tell AI coding agents how to consult the wiki and when to update it

Installation

npm install -g @repositories-wiki/repository-wiki

CLI Usage

repository-wiki [options]

Required Options

| Flag | Description | |------|-------------| | --provider-id <id> | LLM provider ID (openai, anthropic, azure_openai, google-genai, bedrock, sap-ai-core) | | --planer-model <id> | Model for the planning LLM (recommended: Opus-class) | | --exploration-model <id> | Model for the exploration LLM (recommended: Haiku-class) | | --builder-model <id> | Model for the builder LLM (recommended: Sonnet-class) |

You must also provide one of:

| Flag | Description | |------|-------------| | --repo-url <url> | GitHub repository URL to clone (requires --github-token) | | --local-repo-path <path> | Path to a local git repository |

Optional Flags

| Flag | Description | Default | |------|-------------|---------| | --github-token <token> | GitHub token (required with --repo-url or --push-to-github) | — | | --commit-id <id> | Specific commit ID to generate wiki from | HEAD | | --output-dir <path> | Output directory for generated wiki files | repository-wiki | | --push-to-github | Push generated wiki to a GitHub branch | false | | --wiki-branch <branch> | Branch name for the wiki push | repository-wiki-<timestamp> |

Examples

Generate wiki from a local repository using Anthropic:

export ANTHROPIC_API_KEY=sk-...

repository-wiki \
    --provider-id anthropic \
    --planer-model claude-opus-4-6 \
    --exploration-model claude-haiku-4-5 \
    --builder-model claude-sonnet-4-6 \
    --local-repo-path /path/to/my-project

Generate wiki from a GitHub repository URL:

export ANTHROPIC_API_KEY=sk-...

repository-wiki \
    --provider-id anthropic \
    --planer-model claude-opus-4-6 \
    --exploration-model claude-haiku-4-5 \
    --builder-model claude-sonnet-4-6 \
    --repo-url https://github.com/owner/repo \
    --github-token ghp_xxxxxxxxxxxx

Programmatic API

The package also exports a programmatic API for use in your own tools:

import { WikiGeneratorPipeline } from "@repositories-wiki/repository-wiki";

const pipeline = WikiGeneratorPipeline.create();
const result = await pipeline.execute({
  localRepoPath: "/path/to/my-project",
  providerConfig: { providerID: "anthropic" },
  llmPlaner: { modelID: "claude-opus-4-6" },
  llmExploration: { modelID: "claude-haiku-4-5" },
  llmBuilder: { modelID: "claude-sonnet-4-6" },
});

console.log(result.wikiStructure.title);
console.log(result.wikiStructure.sections);
console.log(result.commitId);

Exports

| Export | Description | |--------|-------------| | WikiGeneratorPipeline | The main pipeline class — call .create() for the default pipeline, or build a custom one with .addStep() |

How the Pipeline Works

flowchart TD
    A["<b>Setup Repository</b> <br>Clone from URL or<br>alidate local repo"] --> B["<b>Infer Files</b><br>Tree-sitter parsing +<br>exploration LLM"]
    B --> C["<b>Generate Structure</b><br>Planner LLM designs<br>sections & pages"]
    C --> D["<b>Generate Pages</b><br>Builder LLM writes each<br>page with source citations"]
    D --> E["<b>Write to Local</b><br>Markdown files +<br>INDEX.md + AGENTS.md"]
    E --> F["<b>Push to GitHub</b><br>Optional: commit &<br>push to a branch"]

| Step | What It Does | |------|--------------| | Setup Repository | Clones from a URL (with optional token/commit) or validates a local git repo | | Infer Files | Walks the repo file tree, reads the README, uses Tree-sitter to extract code signatures, and asks the exploration LLM to identify the most important files | | Generate Structure | Sends the file tree and enriched file data to the planner LLM, which designs the full wiki structure — sections, pages, importance levels, and relevant source files | | Generate Pages | Generates all wiki pages concurrently (up to 100 in parallel) using the builder LLM, with pre-loaded source files injected into each prompt for accurate citations | | Write to Local | Writes markdown files organized by section, generates INDEX.md, and creates/updates AGENTS.md at the repo root | | Push to GitHub | If enabled, creates a new branch, commits the wiki and AGENTS.md, and pushes to the remote |

Supported LLM Providers

Before running, export the environment variables required by your LLM provider:

| Provider | Env Variable | Setup Guide | |----------|-------------|-------------| | anthropic | ANTHROPIC_API_KEY | LangChain docs | | openai | OPENAI_API_KEY | LangChain docs | | azure_openai | AZURE_OPENAI_API_KEY | LangChain docs | | google-genai | GOOGLE_API_KEY | LangChain docs | | bedrock | AWS credentials | LangChain docs | | sap-ai-core | SAP AI Core config | SAP AI SDK docs |

Tree-sitter Language Support

The exploration step uses Tree-sitter to extract code signatures (functions, classes, interfaces) for richer context. Currently supported languages:

| Language | Extensions | |----------|-----------| | TypeScript | .ts | | TSX | .tsx | | JavaScript | .js, .jsx, .mjs, .cjs | | Python | .py, .pyw | | Java | .java | | Go | .go |

Files in other languages are still included in the wiki — they're just processed as plain text without signature extraction.

Want better results for your language? You can add your own Tree-sitter grammar by implementing a language query in src/tree-sitter/language-queries/ and placing the corresponding .wasm grammar file in assets/grammars/. See the existing language queries for reference.

Keeping the Wiki Updated

The wiki generation is a one-time step. To keep it current as your code evolves, install the update-wiki agent skill included in the monorepo. The generated AGENTS.md tells your agent when to update the wiki; the skill teaches it how.

Claude Code:

cp -r skills/update-wiki ~/.claude/skills/update-wiki

OpenCode:

cp -r skills/update-wiki ~/.config/opencode/skills/update-wiki

Cursor:

cp -r skills/update-wiki ~/.cursor/skills/update-wiki

Once installed, your AI agent will automatically update affected wiki pages whenever it modifies code that's referenced in the wiki — no manual re-generation needed.

Related

• @repositories-wiki/mcp — MCP server that serves generated wikis to AI coding tools
• Repositories Wiki — Monorepo with examples, the update-wiki agent skill, and full documentation

License

MIT