Learn it. Build it. Ship it for others.
Interactive operator playbook for Claude Code
This repo is a live book about building agent-led workflows and HTML-first artifacts. It combines MDX chapters, interactive widgets, data pages, and workflow files that show how the system is run.
Builders who want to publish living playbooks, reusable prompts, and agent workflows with Claude Code.
You can study the chapters and copy the same workflow into your own living docs or agent-run project.
What it does
Forty-eight chapter playbook
MDX chapters present the operator manual with linked sections, glossary terms, and navigation across the book.
Interactive widgets
The site includes reusable components such as leaderboards, calculators, and visual tools embedded inside the chapters and pages.
Self-updating radar
A radar page republishes current AI ecosystem signals on a schedule without manual editing.
Agent workflow files
`CLAUDE.md`, `AGENTS.md`, and `.claude/workflow-kit.json` define the repo's Claude Code workflow and review flow.
Reusable starter vault
`vault-starter/` provides template folders and a Claude Code setup for organizing notes and projects.
Repo checks and guards
Scripts and tests validate links, numbers, SEO, glossary terms, and other page integrity before build and release.
How to get it
- 1Built and maintained by Claude Code agents — content, widgets, data refreshes, review…
npm install npm run dev # http://localhost:4321 npm run build # prebuild guards → astro build → postbuild checks → dist/ npm run preview # serve dist/ locally
README
Vlad's Playbook
The Ultimate AI Dive Deep
A 48-chapter operator field manual where every artifact is live, clickable, and forwardable — and the repo runs the discipline the book teaches.
→ Read it: dive.vladyslavpodoliako.com ←
The thesis: HTML-ization
Stop sending dead files. Every report, pitch, audit, deck, and analysis ships as a live interactive HTML artifact on a deployed link — not a PDF attachment that started rotting the second it was exported. The link is current because the repo is. People forward links; they archive PDFs unread.
This repo is the proof. The book about building this kind of artifact is, itself, this kind of artifact — a repo deployed to a public link, updated by commit, with real interactive case studies embedded and clickable inside it, and a radar page that re-publishes itself every hour without a human in the loop.
→ Read the thesis + click the embeds: dive.vladyslavpodoliako.com/html-first
Read it (the site is the book)
The repo is how it's built. The site is what it IS.
| Surface | What's there | Live link |
|---|---|---|
| All 48 chapters | The whole book, MDX-rendered, with TL;DRs, glossary tooltips, and the receipts behind every claim | /sections |
| Radar | A self-updating index of what's moving through the AI ecosystem, ranked by lead time — recomputed hourly by a cron pipeline, receipts on every row | /radar |
| Agent workflow | The queue that feeds the swarm: boards, issue discipline, three slash commands, six path-routed review agents — the system this very repo runs on | /agent-workflow |
| Fleet paint | Six agents, six identical black windows — statusline, OSC escape codes, and a watcher that colors every terminal by project | /terminal-setup |
| Tier list | Five readings of the same models: the Arena crowd boards (hand-verified snapshot, cross-checked against a community mirror), Artificial Analysis with $/task economics + its Agentic Index, the labs' launch decks discounted on arrival, Arena's Agent board, and the operator ranking you can drag and share | /tier-list |
| Model files | Opus 5 (the effort dial, the hierarchy question) and Fable 5 / Mythos 5 (a hub + 9 spokes, the system card read honestly) | /opus-5 · /fable-5 |
| HTML-ization | The flagship — thesis + embedded live artifacts (AFC pitch deck, AFC robot stable, sanitized client deliverability audit) | /html-first |
| CAD-as-code | Claude designs a 3D-printable museum frame as 267 lines of Python — STL + STEP as build artifacts, zero grams printed until the calipers agree | /cad-as-code |
| Music is math | How AI actually writes a song, and where the same recipe conquered proteins, robots and the weather — with the five claims the first draft got wrong kept on the page | /music-is-math |
| Learn | New to any of this? The official free courses in the right order, then the book | /learn |
| Cheat sheet | Slash commands, settings keys, hook JSON shape, permission syntax, model routing — printable | /cheat-sheet |
| Glossary | 89 terms, A–Z, linked inline throughout the chapters | /glossary |
| Resources | Copy-paste templates: CLAUDE.md skeletons, .mcp.json, hooks, SKILL.md, subagent .md, reusable prompts | /resources |
| Research notes | 13 dated external signals that shift what an operator does Monday — sourced, signal-vs-receipt discipline | /research-notes |
Press ⌘K anywhere on the site — search every chapter, page, section anchor, glossary term, and research note from one box. The changelog drives a dismissible "what's new" bar on every page, and homepage tiles carry auto-expiring NEW badges derived from each page's git ship date — nothing on the site claims to be new by hand.
The repo runs the book
This isn't a docs folder next to a manuscript — the repository practices the operating discipline the chapters describe:
- Issues → board → branch → review fleet → PR. Work is filed from
issue templates written so an agent with no
memory of the conversation can open a mergeable PR, and every branch runs a
path-routed review fleet before its PR opens
(docs/workflow/, config in
.claude/workflow-kit.json). The system itself is
documented on
/agent-workflow. - Prebuild guards fail the build on dishonesty: stale numbers against a
ledger, broken internal links, glossary drift, nested anchors, template-literal
breakage — see
scripts/check-*.pyand theprebuildchain inpackage.json. - Data pages are dated captures, not vibes. The tier-list's Arena boards are hand-verified snapshots cross-checked against an independent community mirror; Artificial Analysis figures trace to the page's own embedded JSON-LD; every capture is date-stamped on the surface that shows it.
- The radar updates itself. A sibling pipeline pushes hourly; the site rebuilds on every push (~90s via GitHub Pages), which is also what makes the homepage's NEW badges expire without anyone editing a label.
What's in this repo (the receipts)
- 48 chapters in
src/content/chapters/*.mdx - 31 React/Astro widgets in
src/widgets/(Arena leaderboard, AA economics panel, sortable tier list, command palette, token-burn calculator, swarm visualizer, tokenizer lab, …) - 50+ standalone pages in
src/pages/— the surfaces above plus the journey, day zero, questions, showcase, vault starter, weekend builds, swarms, and more - 89 glossary terms in
src/lib/glossary.ts· 13 dated research notes insrc/lib/research-notes.ts - 4 embedded interactive artifacts in
public/artifacts/— single-file, self-contained, sandboxed-iframe-embedded - 35 shipped editions in
src/lib/changelog.ts, each with receipts
How it's built (5 bullets)
- Astro 5 static site generator. Server-rendered by default; React 18 islands only where interaction needs them.
- MDX for chapter content with custom Astro components
(
<Callout>,<PullQuote>,<TLDR>,<GlossaryTooltip>, …). - Tailwind 3 with project design tokens
(
rgb(var(--accent)),--paper,--line, dark/light themes). - GitHub Pages auto-deploy via
.github/workflows/deploy.yml— push tomain, live in ~90 seconds atdive.vladyslavpodoliako.com. - Built and maintained by Claude Code agents — content, widgets, data refreshes, review fleets, deploys. The process is the technique the book describes; the artifact IS the technique.
Run locally
npm install
npm run dev # http://localhost:4321
npm run build # prebuild guards → astro build → postbuild checks → dist/
npm run preview # serve dist/ locally
Requires Node 20+.
Add a chapter or page
- Chapters: drop an MDX file into
src/content/chapters/matchingsrc/content/config.ts, add it tosrc/lib/chapters.ts— the dynamic route picks it up. - Standalone pages need eight wiring surfaces, or they ship invisible:
page file · ⌘K index · chapter cross-link · glossary · homepage tile
(
<TileEyebrow added="…">— the NEW badge derives from the ship date) · changelog · site nav · the hand-maintainedllms.txtsite map.
See CONTRIBUTING.md for how work is filed and reviewed.
Credits
Written and built by Vlad Podoliako — Founder & CEO of Belkins, founder of Folderly, and operator across a portfolio of others.
- 📬 Vlad's Newsletter (10K+ subscribers)
- 🌐 vladyslavpodoliako.com
- 🎧 Spotify (My music side project)
The book teaches the technique. The artifact IS the technique. The repo is how it's built — but the site is what it IS.
→ dive.vladyslavpodoliako.com
License
- Code — MIT. Copy the Astro + Tailwind + widget structure into your own living-link artifacts; that's the entire point.
- Content — CC BY-NC-SA 4.0. Read it, share it, quote it, credit it; don't resell it.
See LICENSE.
Files in the repo
- .claude
- .github
- apps
- assets
- docs
- outputs
- public
- scripts
- src
- tests
- vault-starter
- .env.example
- .gitignore
- AGENTS.md
- astro.config.mjs
- CLAUDE.md
- CONTRIBUTING.md
- design-system.html
- LICENSE
- package-lock.json
- package.json
- postcss.config.cjs
- README.md
- SECURITY.md
- tailwind.config.ts
- tsconfig.json
Discussion (0)
Ask about usage, or say what you built with itSign in to join the discussion.
No comments yet. Be the first to say what this is good for.
More tutorials & guides
Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1
A visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value.
Specification and documentation for Agent Skills

💻 vibe coding 101|The first course for AI-native product builders.

End-to-end, code-first tutorials for building production-grade GenAI agents. From prototype to enterprise deployment.