> ## Documentation Index
> Fetch the complete documentation index at: https://josephfajen.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Book of Archon

> Official 10-chapter onboarding guide for a 21k+ star open-source harness builder for AI coding

## The Project

[Archon](https://archon.diy) is an open-source harness-building system for AI coding agents with 21k+ GitHub stars. The project needed an onboarding guide that could help new users get productive quickly.

I wrote the Book of Archon — a 10-chapter guide that was submitted via PR, accepted by the maintainers, and now serves as the project's official user guide.

***

## What I Delivered

### 1. 10-Chapter Onboarding Guide

User guide covering installation through advanced agent configuration. Structured to get users productive quickly while providing depth for advanced use cases.

**Result:** [archon.diy](https://archon.diy)

### 2. Adversarial Quality Workflow

Built this documentation using a custom Writer Agent vs. Attacker Agent pipeline:

* The Attacker Agent challenged every factual claim
* The Writer Agent had to defend or revise
* Quality gates set at 7/10+ before content could advance

This adversarial approach strengthens the documentation to withstand scrutiny.

<CardGroup cols={2}>
  <Card title="10 Chapters" icon="book">
    Complete onboarding guide accepted as the project's official documentation.
  </Card>

  <Card title="Adversarial QA" icon="swords">
    Writer vs. Attacker workflow with 7/10+ quality gates. Every claim tested.
  </Card>

  <Card title="20k+ Stars" icon="star">
    Official docs for a major open source project.
  </Card>

  <Card title="Merged via PR" icon="check">
    Submitted, reviewed, and accepted by the Archon maintainers.
  </Card>
</CardGroup>

***

## The Methodology

For the Book of Archon, I developed a full methodology for AI-assisted documentation production. The approach:

1. Human writes the PRD — detailed spec with structure, voice, terminology, acceptance criteria
2. Workflow executes against spec — 31-node Directed Acyclic Graph (DAG) with per-chapter validation gates
3. AI drafts each section — consistent execution against the blueprint
4. Human edits and approves — judgment, polish, nuance

**Output:** 10 chapters, 11,000 words, 6 hours total. Traditional estimate: 32-40 hours.

> The same approach works for other large documentation projects. Constrain the AI with detailed specs, then let it generate against those constraints.

***

## Tech Stack

`Claude Agents SDK` · `YAML Workflows` · `Pydantic AI` · `Markdown` · `Git/GitHub`