Sandbox
@Belkins/ai-dive-deep

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.

364 stars1 forksAstroUpdated 6d ago
Who it's for

Builders who want to publish living playbooks, reusable prompts, and agent workflows with Claude Code.

What it delivers

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

  1. 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

Edition   Live   License: code MIT, content CC BY-NC-SA 4.0


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.

SurfaceWhat's thereLive link
All 48 chaptersThe whole book, MDX-rendered, with TL;DRs, glossary tooltips, and the receipts behind every claim/sections
RadarA 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 workflowThe 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 paintSix agents, six identical black windows — statusline, OSC escape codes, and a watcher that colors every terminal by project/terminal-setup
Tier listFive 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 filesOpus 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-izationThe flagship — thesis + embedded live artifacts (AFC pitch deck, AFC robot stable, sanitized client deliverability audit)/html-first
CAD-as-codeClaude 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 mathHow 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
LearnNew to any of this? The official free courses in the right order, then the book/learn
Cheat sheetSlash commands, settings keys, hook JSON shape, permission syntax, model routing — printable/cheat-sheet
Glossary89 terms, A–Z, linked inline throughout the chapters/glossary
ResourcesCopy-paste templates: CLAUDE.md skeletons, .mcp.json, hooks, SKILL.md, subagent .md, reusable prompts/resources
Research notes13 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-*.py and the prebuild chain in package.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 in src/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)

  1. Astro 5 static site generator. Server-rendered by default; React 18 islands only where interaction needs them.
  2. MDX for chapter content with custom Astro components (<Callout>, <PullQuote>, <TLDR>, <GlossaryTooltip>, …).
  3. Tailwind 3 with project design tokens (rgb(var(--accent)), --paper, --line, dark/light themes).
  4. GitHub Pages auto-deploy via .github/workflows/deploy.yml — push to main, live in ~90 seconds at dive.vladyslavpodoliako.com.
  5. 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

  1. Chapters: drop an MDX file into src/content/chapters/ matching src/content/config.ts, add it to src/lib/chapters.ts — the dynamic route picks it up.
  2. 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-maintained llms.txt site 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.

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

Repository payload26 top-level entries
  • .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 it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More tutorials & guides

shareAI-lab/
learn-claude-code

Bash is all you need - A nano claude code–like 「agent harness」, built from 0 to 1

77k
luongnv89/
claude-howto
luongnv89/claude-howtoTutorials & Guides

A visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value.

41k
agentskills/
agentskills
agentskills/agentskillsTutorials & Guides

Specification and documentation for Agent Skills

25k