LLM Wiki: The 4 Layers Every Repo Needs | Harness Engineering 101
Read the Portuguese translation →
Author: The Augmented · Watch on YouTube
Video authored by the channel shown, not by IANIA; all rights belong to the creator. Transcript and summary generated automatically; the original-language content is preserved.
Original language: English
Original video: LLM Wiki: The 4 Layers Every Repo Needs | Harness Engineering 101 · Channel: The Augmented · 00:32 · Watch on YouTube
Topics: AI agents, Software engineering with AI, RAG, search and information retrieval
Summary
Section titled “Summary”A strong LLM wiki has four layers: detailed agent notes, a clean human-facing view, carefully reviewed context and rules, and a janitor that maintains consistency. Agent notes can include verbose details that help agents resume tasks, while the human layer distills information into a dashboard or Kanban board. The curated layer holds slowly changing repo conventions, and the janitor catches conflicting rules and removes outdated material. Together, these layers support a useful agentic knowledge base for many kinds of work.
Key points
Section titled “Key points”- The wiki has four layers: agent notes, a human-facing layer, curated context and rules, and a janitor.
- Agents write detailed notes for other agents and can include hashes, branches, and commits.
- The human layer is a clean, distilled view that links to detailed agent notes when needed.
- A dashboard or Kanban board can serve as the human-facing layer.
- The human layer is for managing your own context, not simply reviewing what the agent is doing.
- The curated layer contains slowly changing skills, rules, and repo conventions that need careful review.
- The janitor identifies conflicting rules in agent notes and removes old material.
- All four layers together can support research, personal growth, project work, or other uses.
Chapters
Section titled “Chapters”| Start | Chapter | Summary |
|---|---|---|
| 00:00 | The four layers and agent notes | The video introduces the four-layer wiki and explains how agents write detailed notes for other agents. |
| 00:32 | Human view, curated context, and maintenance | It describes the dashboard-like human layer, carefully reviewed repo conventions, and a janitor that keeps the wiki consistent and useful. |
Quotes
Section titled “Quotes”A good LLM wiki needs four layers. — 00:00
You never need to read it or go here. — 00:00
And then you need the janitor. — 00:32
Concepts and entities
Section titled “Concepts and entities”LLM wiki, agent layer, human layer, curated layer, janitor, Kanban board
Full transcript (original language)
Section titled “Full transcript (original language)”00:00 A good LLM wiki needs four layers. A place where agents can write for agents, a clean surface for humans to read, a curated layer of slowly changing context and rules that you review, and a janitor that keeps it all up to date and consistent. The agent layer is written by agents for other agents. Agents can be as verbose as they need to be. These can include details that they need to resume a task later like hashes, branches, commits. You never need to read it or go here. The human layer is the clean distilled view that you link out and that goes to these detailed agent notes when necessary. Think of a
00:32 dashboard or a Kanban board. It’s less about reviewing what the agent’s up to and more about managing your own context. The curated layer, this is the slowly changing context that you need to handle with care. These are the skills and rules, the conventions of your repo. And then you need the janitor. This looks to catch conflicting rules in your agent notes before they spread. And it cleans up old material so the human layer stays useful. With all four layers, you can build a strong agentic knowledge base for research, personal growth, project work, or anything else.