Skip to content

Yellow Belt: The First Builds

Status: drafted · Time: 12 min · Audience: daily-builder Outcome: Understand the Yellow Belt path, daily AI-building claim, quests, boss fight, and evidence chain.

Graded artefacts for this belt: 🎮 Quest Y-1: Ship the Stand-up Bot · 🎮 Quest Y-2: The 30-day 2-minute challenge · 🏁 Yellow Belt boss fight · 🏆 Yellow Belt badge template

Promise. By the end of Yellow Belt, Claude Code is part of your daily rhythm. You can choose the right tool, give it useful context, use connectors safely, read unfamiliar code, and ship a real bug fix in a surface you care about.

Yellow Belt is where the playbook stops being “setup and first PR” and becomes a working habit. White Belt proved you can ship code once. Yellow Belt proves you can find small useful work, bring the right context to the AI, and close the loop with review.

The claim is:

“I build with AI daily. I find and fix things in the surfaces I care about.”


Start here if:

  • White Belt is complete;
  • you have a GREEN environment;
  • you can open a PR without needing the whole room to stop;
  • you want Claude Code to become a daily teammate instead of an event-day novelty;
  • you are ready to use Slack, docs, design files, tickets, and git history as context sources.

If you already use AI tools every day, skim the first two modules and spend your energy on Y03, Y11, Y12, Y13, and the boss fight. That is where Yellow Belt earns its name.


Yellow Belt proves four things:

  1. Tool judgement. You know when to use Claude Code, Codex, a provisioned Claude workspace, Cursor, Copilot, Slash, or a connector surface.
  2. Prompt judgement. You can state intent, constraints, context, and success criteria without turning every prompt into an essay.
  3. Context judgement. You can give Claude the right repo files, threads, docs, and design context without leaking sensitive material or flooding the window.
  4. Shipping judgement. You can notice a real bug, triage it with AI, propose a small fix, and get the PR reviewed.

White Belt asks: can you ship one tiny change?

Yellow Belt asks: can you repeat the loop with judgement?


You need:

  • White Belt awarded or equivalent evidence;
  • a working Claude Code setup;
  • a repo you are allowed to inspect;
  • access to the approved connector surfaces for your role;
  • a team or program route for review.

Design-track readers should also complete the design environment add-on before Y09. That add-on is not repeated here; Yellow Belt uses it.


Yellow Belt is designed for 6-8 hours over two weeks, plus the time it takes for PR review.

BlockTimeWhat happens
Y01-Y02~45 minTool choice and decision flow
Y03-Y05~85 minPrompt quality and context foundations
Y06-Y08~65 minReading code, permissions, model path vocabulary
Y09-Y10~60 minDesign, Slack, and workspace connectors
Y11-Y14~90 minBug hunting, debugging, PR craft, staying current
Quests + boss fightvariableHabit proof, utility proof, and one real shipped fix

The modules are meant to be read in order the first time. After that, they become a reference.


SectionChapterPillarTime
Y.1The Tool AtlasContext30 min
Y.2When to reach for which toolHarness15 min
Y.3Prompt quality, deep divePrompt40 min
Y.4What Claude can see and what it cannotContext20 min
Y.5CLAUDE.md primerContext25 min
Y.6Reading unfamiliar code with ClaudePrompt30 min
Y.7Permissions, hooks, slash commandsHarness20 min
Y.8LiteLLM and Claude workspace accessHarness15 min
Y.9Figma MCP for non-engineersContext30 min
Y.10Slack MCP + Google Workspace MCPContext30 min
Y.11Bug hunting with AIPrompt30 min
Y.12Debugging with ClaudePrompt30 min
Y.13PR craftHarness20 min
Y.14Staying currentMeta10 min

Each module follows the same rhythm you saw in White Belt: promise, short version, mental model, worked example, common failures, colour self-check, and what you can say after.


Yellow Belt is awarded when the evidence chain closes:

EvidenceFileWhat it proves
White Belt awardedWhite badge rowYou have the baseline setup and first PR loop.
Y-1 working utilityQuest Y-1You can build a small useful tool with Claude Code.
Y-2 daily practice logQuest Y-2You can sustain the habit in normal work.
Y-B merged or review-active PRBoss Fight Y-BYou can find, triage, fix, and ship one real bug.
Claim templateBadgeYour evidence can be tracked and certified.

Yellow evidence includes one thing White did not: a reflection on prompt, context, and harness decisions. The point is not just that you shipped. The point is that you can explain the judgement that made shipping possible.


If you are a designer, Y09 is the bridge from design artefact to running code. Before that chapter, make sure your design-track setup is GREEN:

  • approved design connector available;
  • design-system connector available where applicable;
  • local package install path verified;
  • sandbox repo ready;
  • permission boundaries clear.

Do not debug design connector setup inside Y09. Use Appendix B routing first.


Do not optimize for the largest prompt, the most tools, the most dramatic demo, or the longest streak screenshot.

Optimize for:

  • the smallest useful tool for Quest Y-1;
  • two minutes of real daily practice for Quest Y-2;
  • a small but real bug for the boss fight;
  • PR descriptions that show your triage;
  • reflections that name the context you gave the AI.

Yellow Belt is not about proving you can make AI do everything. It is about proving you know when and how to use it.


“I build with AI daily. I find and fix things in the surfaces I care about.”


Previous: White Belt badge - Next: Y.1 The Tool Atlas