A nonfiction book by Justin Johnson on what knowledge work looks like when a senior leader operates through a persistent, multi-agent harness instead of a chat session.
This is the public showcase for Builder-Leader. Not the book.
It exists to demonstrate three things, and only three things:
- The book is rigorous. Visible production scaffolding (research artifacts, multi-stage critique, fact-check registries, voice polish) shows the work was built carefully, not generated.
- The framework is real and reusable. The Claude Code skills used to draft, critique, fact-check, and ship the manuscript are checked in. Clone them, point them at your own writing.
- The book is out. The preface is here in full; the remaining ten chapters are in the book.
Get it on Amazon: amazon.com/dp/B0H3LQQ4J6. More at builder-leader.com.
Every chapter passed nine sequential stages before it was allowed to be called ship-ready.
| Stage | What it does | Per-chapter artifact |
|---|---|---|
| 1 — Canon | Thesis, outline, audience, voice locked at kickoff. | canon/ |
| 2 — Research | Three-tier platform sweep before drafting. Floor: six raw research files per chapter. | research/<ch>/findings.md |
| 3 — Draft | Chapter generated against canon and findings. Post-draft scan for voice violations. | chapters/NN-<slug>.md |
| 4 — Stage 2 red team | Structural and adversarial critique. Weak arguments, unsupported load-bearing claims. | 04_red_team.md |
| 5 — Fact-check | Every factual claim extracted, matched to a source, classified. | 01_assertion_registry.md |
| 6 — Adversarial reader | Multi-persona reader simulation, including the hostile reviewer the book argues against. | 05_adversarial_reader.md |
| 7 — Voice polish | Mechanical scan for banned words, em-dashes, sentence-length monotony, plus per-hit Opus judgment. | 06_voice_polish.md |
| 8 — Stage 6 red team | Final rigor gate. Looks specifically for what slipped through six prior passes. | 07_red_team_v2.md |
| 9 — Ship | Apply blockers, resolve open questions, advance status, citation re-verify against live sources. | inline + Wayback archive |
See PRODUCTION-STATS.md for the full numbers (claim counts, citation index, words per chapter, rigor pipeline).
| Directory | What's there | What's not |
|---|---|---|
chapters/_sample/ |
The full preface as a reading sample. | The other 10 chapters. |
export/ |
Typeset PDF of the preface only. | The full manuscript PDF and EPUB. |
.claude/skills/ |
Project-local Claude Code skills used to build the book. | — |
research/_samples/ |
One chapter's full research artifacts (Ch 7 — personal harness building). | The full research corpus. |
review-work/_samples/ |
One chapter's full nine-stage rigor pipeline + manuscript-wide ship-day summary. | The other ten chapters' rigor artifacts. |
planning/_samples/ |
One ship-day session log. | All other session logs. |
canon/ |
A README explaining the canonical spine. | The actual canon files (thesis, outline, audience, voice are reserved). |
field-guide/ |
Companion exercises for each chapter. Take the fifteen-minute self-audit (01-two-groups.md). |
— |
PRODUCTION-STATS.md |
One page summarizing the whole production. | — |
- Markdown source:
chapters/_sample/00-preface.md - Typeset PDF:
export/preface-sample.pdf
The skills in .claude/skills/ are project-local Claude Code skills, built and tuned during this book's production. To use them on your own project:
- Drop the skill directory into your project's
.claude/skills/folder. - Open Claude Code from your project root — the skills are auto-discovered.
- The general patterns transfer cleanly. The author's specific banned-word lists, canon references, and chapter-slug conventions are sample data — replace with your own.
What they cover: three-tier research sweeps, draft generation grounded in canon, structural and adversarial critique (Stage 2 + Stage 6), fact-check assertion registries, multi-persona reader simulation, mechanical voice polish, manuscript export, audiobook generation.
The book argues for running Claude Code as a persistent system (skills, hooks, memory, a self-correction loop, multi-provider routing), not a chat box you visit. Two open-source repos put that whole system in the open, well beyond the three skills showcased here:
- Claudelicious: a cookbook for running Claude Code as a system, not a chat box. The why and the wiring behind a full harness, written up to copy and adapt.
- Slopless: the author's real production Claude Code config, 18 months and 50+ projects, sanitized. Every rule earned the hard way.
The book makes the case; these two repos are the working setup behind it.
Justin Johnson — built and led an AI Center of Excellence at a Fortune 500 pharmaceutical company, and now leads Applied Data and AI in R&D at an AI-driven biotech. He writes Run Data Run and the AIXplore technical blog.
- Site: builder-leader.com
- Substack: Run Data Run
- LinkedIn: justinhaywardjohnson
- X: @bioinfo
- Code and skills: MIT.
- Sample artifacts and the preface: CC-BY-NC-ND 4.0 (read, share, attribute; no commercial use, no derivatives).
- The book itself: All rights reserved.
See LICENSE for full terms.