← Ship a FeatureArrows or space · F for fullscreen · Print to saveTalks

CLAUDE CODE BUILD NIGHT · PART 4

Ship a
Feature

One feature, tested, reviewed and merged in a night.

Ship AI · TalksShip a Feature01/20

Who's hosting

How you do anything is how you do everything.

Santos Hernandez
Santos HernandezFounder, Ship AI
$25k+/moggbucks, built with agents
$0 to $12MARR at ZBD
EU firstMiCAR approval
THE COMMUNITY

Ship AI

Free workshops and talks in Phoenix.

shipai.club
THE STUDIO

Ship AI Studio

AI built into your business, in two weeks.

shipaistudio.com

Ship AI · TalksShip a Feature02/20

Join the room

SHIP AI DISCORD

Scan it now.
Prompts later.

Open the Discord invite

Ship AI · TalksShip a Feature03/20

THE IDEA

Shipping once is easy.
Shipping again is the job.

Tonight we change it without breaking it.

Ship AI · TalksShip a Feature04/20

Our partner

Ship AIWorkuityAI Exchange

Building Phoenix's AI community, together. Workuity hosts tonight, and we present Founder Weekend side by side.

Ship AI · TalksShip a Feature05/20

Write down done first

PROMPT AND HOPE

"Add favorites."

TESTS FIRST

Done, written as checks.
Then prove it.

Ship AI · TalksShip a Feature06/20

Tonight

01

Feedback

02

Spec

03

Tests

04

Ship

Ship AI · TalksShip a Feature07/20

Tonight's feature

6:02
What's for DinnerM

What's in your fridge?

eggsspinachfetaonion
Find dinners

Favorites · 3 saved

Shakshuka25 min · one pan
Saved
Spinach and feta omelette15 min · easy
Saved
Greek frittata20 min · oven
Saved
SHAKSHUKA · SAVED TUEOne tap, every time
  1. Soften the onion in oil
  2. Wilt in the spinach
  3. Make four wells
  4. Crack in the eggs, cover
  5. Crumble feta, serve

Save the dinners you'd make again. Find them in one tap.

One featureVitestPlaywright

Ship AI · TalksShip a Feature08/20

The feature spec

SPEC.md · Feature: Favorites
Feedback

"I had to scroll my history to find it."

The one job

Keep the dinners I'd make again one tap away.

Acceptance criteria

Saved dinner shows at the top of Favorites

Saving twice keeps one copy

Another account can't see mine

Test plan

Vitest for the logic, Playwright for the flow.

Ship AI · TalksShip a Feature09/20

Let feedback pick the feature

claude · ~/whats-for-dinner

youHere are eleven notes from testers.

claudeThree ask to save dinners. One asks for meal plans.

claudeFavorites fits the one job. Meal plans can wait.

✓ Pick: Favorites

Copy · pick the feature
Read SPEC.md and FEEDBACK.md, group the notes, check each group against the one job, and recommend one feature. No code yet.

Full prompt

Copy this entire prompt into Claude Code.

Read SPEC.md, then read FEEDBACK.md, where I pasted what testers told me, one note per line.

1. Group the notes by what the person was trying to do, and count each group.
2. For each group, say in one line whether it fits The one job in SPEC.md or something under Out of scope.
3. Recommend one feature we can build and ship tonight: the biggest group that fits the one job. Say what you'd leave out.

Don't write any code. Stop and wait for me to pick.

Ship AI · TalksShip a Feature10/20

Write the feature spec

claude · ~/whats-for-dinner

claudeShould saving twice keep two copies?

youNo. One copy.

claudeCan someone favorite from their history?

youYes, every card gets a heart.

✓ Added Feature: Favorites to SPEC.md

Copy · the feature spec
Interview me about Favorites, then add a Feature section to SPEC.md with acceptance criteria and a test plan. Wait for my OK.

Full prompt

Copy this entire prompt into Claude Code.

We're adding one feature: Favorites. Read SPEC.md first.

Interview me about it one question at a time and wait for each answer. Cover what the person taps, what they see, what happens on a second tap, what data we keep, and who can see it.

Then add a section to the end of SPEC.md called "Feature: Favorites" with: Feedback, The one job, Out of scope, Screen, Data, Acceptance criteria (checkboxes, each one something a test can check), Test plan (which checks are Vitest unit tests for logic, which are Playwright end-to-end tests at phone width, and which I check by hand), and a Ship checklist.

Don't change the rest of SPEC.md. Stop and wait for my OK before writing any code.

Ship AI · TalksShip a Feature11/20

Set up the tests

~/whats-for-dinner
ONE TIME$ npm install -D vitest
✓ vitest.config.mts
$ npm init playwright@latest
✓ playwright.config.ts

Two runners, one for logic and one for the whole flow.

01

Vitest

Fast checks for the logic in lib/.

02

Playwright

A real browser, sized like an iPhone.

03

A hook

Tests run after every edit Claude makes.

Copy · set up the tests
Install Vitest, have me run npm init playwright@latest, add an iPhone 13 project, then add a hook that runs the tests after every edit.

Full prompt

Copy this entire prompt into Claude Code.

Read SPEC.md, especially Feature: Favorites. Set up testing one step at a time and stop after each step.

1. Install Vitest following the Next.js Vitest guide:
npm install -D vitest @vitejs/plugin-react jsdom @testing-library/react @testing-library/dom vite-tsconfig-paths
Add vitest.config.mts and a "test" script that runs vitest.
2. Tell me to run this in my own terminal, because it asks questions:
npm init playwright@latest
Wait until I tell you it's done.
3. In playwright.config, add a project that uses the iPhone 13 device, and a webServer that starts npm run dev.
4. Add a PostToolUse hook to .claude/settings.json with the matcher Edit|Write that runs npx vitest run. If the tests fail, it exits with code 2 and writes the failure summary to stderr so you see it. Follow code.claude.com/docs/en/hooks-guide. Tell me to restart Claude Code so the hook loads, and wait.

Don't write any feature code yet.

code.claude.com/docs/en/hooks-guide

Ship AI · TalksShip a Feature12/20

Red first, with Matt Pocock's /tdd

claude · ~/whats-for-dinner

claudeSeams: lib/favorites.ts and the Favorites screen. OK?

youOK.

✗ toggles a favorite on (not implemented)

claudeRed. Ready to write the code?

Copy · failing tests first
Install mattpocock-skills if needed, use the tdd skill, agree the seams with me, then write one failing Vitest test and one failing Playwright test.

Full prompt

Copy this entire prompt into Claude Code.

If you don't have Matt Pocock's skills yet, I'll install them in my terminal:
claude plugin install mattpocock-skills@claude-plugins-official
Wait until I tell you it's done, and tell me to restart Claude Code if you can't see the tdd skill.

Read SPEC.md, especially Feature: Favorites. Use the tdd skill.

1. Propose the seams under test: lib/favorites.ts for the logic, and the Favorites flow in the browser. Wait for my OK.
2. Write the first Vitest test from the Test plan, run it, and show me it fails for the right reason.
3. Write the Playwright test for the main flow (heart a dinner, see it in Favorites, reload, still there), run it, and show me it fails.

Don't write any feature code yet. Stop when both are red.

github.com/mattpocock/skills

Ship AI · TalksShip a Feature13/20

Green, one slice at a time

claude · ~/whats-for-dinner

✓ favorites table created

✓ toggles a favorite on and off

✓ saving twice keeps one copy

claudeSlice 2 of 4: the heart on each card.

Copy · build it to green
On a feature/favorites branch, list the slices, then build each one red to green, commit, and wait for me before the next.

Full prompt

Copy this entire prompt into Claude Code.

Read SPEC.md, especially Feature: Favorites. Keep using the tdd skill.

First, create a git branch called feature/favorites. Then create the favorites table from the Data section.

List the slices that get every acceptance criterion to pass, in order, and wait for my OK. For each slice:
1. Write one failing test, run it, and show me it's red.
2. Write only enough code to make it pass, then run all the tests.
3. Tell me what to open at localhost:3000 and wait for me before the next slice.

If something in the spec is unclear, ask me instead of guessing. Don't add anything under Out of scope. Commit after each green slice.

Ship AI · TalksShip a Feature14/20

Green isn't done

WHAT THE TESTS SAY

12 passed.

WHAT YOU SEE

The heart sits under your thumb.
Check it at phone width.

Copy · verify at phone width
Run Playwright on iPhone 13, map every acceptance criterion to a test, then give me a short list to check by hand at phone width.

Full prompt

Copy this entire prompt into Claude Code.

Read the acceptance criteria under Feature: Favorites in SPEC.md.

1. Run the full Playwright suite on the iPhone 13 project and save a screenshot of Favorites with three saved dinners and of the empty state.
2. Go through every acceptance criterion and tell me which test covers it, or that it needs a check by hand.
3. Give me a short list of things to try myself in the browser at phone width: long dinner names, ten favorites, a fast double tap on the heart.

Fix anything that fails, one issue at a time, and show me each fix.

Ship AI · TalksShip a Feature15/20

A fresh pair of eyes

claude · ~/whats-for-dinner

claudeStarting a fresh subagent with only the spec and the diff.

! DELETE /api/favorites/[id] doesn't check the owner

✓ Fixed, with a test for it

Copy · review the diff
Start a fresh subagent with only SPEC.md and the diff, review it against the feature, then verify and fix each real finding test first.

Full prompt

Copy this entire prompt into Claude Code.

Review this feature before it ships.

Start a fresh subagent that hasn't seen our conversation. Give it only SPEC.md and the output of git diff main...feature/favorites. Ask it to review the diff adversarially against Feature: Favorites: missed acceptance criteria, anything under Out of scope that crept in, another account reaching my favorites, and tests that would pass even if the code were wrong. Ask for findings only, with file and line.

When it reports, check each finding against the code yourself. For each real one, write a failing test, then fix it. Tell me which findings you rejected and why. Don't push yet.

code.claude.com/docs/en/sub-agents · or type /code-review

Ship AI · TalksShip a Feature16/20

Preview, then merge

claude · ~/whats-for-dinner

✓ PR #4 Favorites opened

✓ Vercel preview ready

claudeChecks green. Open the preview on your phone, then say merge.

Copy · open the PR
Set preview env vars and Better Auth origins, run Vitest and Playwright in GitHub Actions, open the PR, check the preview, ask before merging.

Full prompt

Copy this entire prompt into Claude Code.

Read SPEC.md. Get Feature: Favorites ready to merge, and stop after each step.

1. Preview deployments use the Vercel project's Preview environment variables, not Production. Create a Neon branch called preview for the preview database, with the new favorites table. Then give me the steps to set these for Preview in the Vercel dashboard myself: DATABASE_URL (the preview branch) and BETTER_AUTH_SECRET (a new value, not the production one). Never print secrets in chat. Wait until I tell you it's done.
2. Better Auth rejects sign-ins from origins it doesn't trust, and every preview has its own URL. Read better-auth.com/docs/reference/options and update the auth config: use the dynamic baseURL with allowedHosts for this project's preview hosts only (a pattern scoped to our project name on vercel.app, never all of *.vercel.app) plus production and localhost. The docs say allowedHosts entries are added to trustedOrigins automatically; use trustedOrigins directly if our version doesn't support that.
3. Write .github/workflows/test.yml so every pull request runs the tests. Follow playwright.dev/docs/ci-intro: actions/checkout and actions/setup-node at the versions it shows, npm ci, npx vitest run, npx playwright install --with-deps, then npx playwright test. The Playwright tests sign up real accounts, so create a Neon branch called ci and tell me to add DATABASE_URL (the ci branch) and BETTER_AUTH_SECRET as GitHub Actions repository secrets myself. Wait until I tell you it's done.
4. Push feature/favorites and open a pull request with gh pr create. In the description, paste the acceptance criteria with each one ticked, and the review findings you fixed.
5. Wait for the Vercel preview deployment and give me its URL so I can sign in and check Favorites on my phone. Wait until I tell you it works.
6. Run gh pr checks --watch. The checks must include the test workflow and the Vercel deployment. If anything fails, fix it and push again.
7. When every check is green, ask me before you merge. After the merge, confirm the production deploy and tick the Ship checklist in SPEC.md.

Next talk: Digital Presence, a public face for What's for Dinner.

Ship AI · TalksShip a Feature17/20

Take it home

THE TALK PAGE

Every prompt,
and this deck.

Open the talk page

Ship AI · TalksShip a Feature18/20

Founder Weekend

FEB 6‑7, 2027 · FREE

Tonight was one hour.
Take a weekend.

An Arizona AI hackathon. Teams of 1-4.

See the hackathon

Ship AI · TalksShip a Feature19/20

SHIP AI DISCORD

Free and open.
Build with us.

Scan to join

Ship AI · TalksShip a Feature20/20
01/20