llm-codebase-wiki

KO | EN

gitlite — search

#python #hacktoberfest #llm #android #machine-learning #docker #macos #java #windows #awesome-list #pytorch #openai #ios #open-source

llm-codebase-wiki

(★ 10)

Karpathy-style llm-wiki for multi-service codebases: ingest, query, lint, PR review, implement

llm-codebase-wiki 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# llm-codebase-wiki

Karpathy-style **llm-wiki** for multi-service codebases.

An agent reads immutable source trees under `raw/`, compiles durable markdown into `wiki/`,
and uses that wiki as the design brief for queries, lint, PR review, and implementation.

Inspired by [Karpathy's llm-wiki](https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f).

## Why

RAG re-searches raw docs on every question. An llm-wiki **compiles** knowledge once and keeps
it consistent. This package specializes the pattern for software platforms: services, DB schemas,
patterns, drift, and wiki-grounded implement / PR-review workflows.

## Package vs your instance

| This repo (public) | Your private instance |
|---|---|
| `WIKI.md` + `schema/` | `config/instance.yaml` (your services) |
| `config/instance.example.yaml` | `raw/github/**` (clones — never commit secrets) |
| `templates/`, `demo/` | `wiki/**` (compiled product knowledge) |

**Rule:** Product and repository names live in `config/instance.yaml` or `wiki/`, never in `schema/`.

## Layout

```
llm-codebase-wiki/
├── WIKI.md                 # operating manual
├── SKILL.md                # agent skill entrypoint
├── schema/                 # ingest | query | lint | pr-review | implement
├── config/
│   └── instance.example.yaml
├── templates/wiki/         # empty wiki bootstrap
└── demo/                   # two toy services + config (try ingest here)
```

## Quick start

1. Copy this repo into an agent workspace (Cursor, Claude Code, Codex, OpenClaw, …).
2. `cp config/instance.example.yaml config/instance.yaml` and list your services.
3. Clone each `services[].repo` into `raw/github/<repo>/`.
4. Tell the agent: **Ingest all services**, then **Run a lint**.

Or try the included demo:

```bash
cd demo
# point your agent at this directory as the wiki root
# then: "Ingest all services" / "What does demo-api do?"
```

## Five operations

| Op | Trigger | Behavior |
|---|---|---|
| **INGEST** | `Ingest raw/github/<repo>` | Compile raw → wiki pages; update index + log |
| **QUERY** | Architecture / service / patterns questions | Answer from wiki; aliases from config |
| **LINT** | `Run a lint` | L1–L9 coverage, links, drift |
| **PR REVIEW** | `Review PR #N` | Code skill + wiki domain checks |
| **IMPLEMENT** | `Implement X` | Wiki as design brief → code → PR → wiki update |

## Instance git policy (recommended)

In **your** private instance repo:

- **Track:** `wiki/`, `schema/` (or submodule this package), `config/instance.yaml`, `WIKI.md`
- **Ignore:** `raw/**` (keep empty dirs with `.gitkeep` if you want)
- **Do not publish** your product `wiki/` — that is proprietary compiled knowledge

## License

MIT. Pattern attribution: Andrej Karpathy's llm-wiki idea.

llm-codebase-wiki

llm-codebase-wiki 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

llm-codebase-wiki

llm-codebase-wiki 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README