Quay 2

Appearance

Swarm 2: TDD Swarm kata ​

Build a password-strength checker the test-first way with Quay's TDD Swarm: one Claude Code agent writes the tests, a second makes them pass, a PowerShell pane re-runs them on every save, and a third agent reviews the result. You drive the loop yourself, one pane at a time.

  • Time not measured yet
  • Tokens not measured yet
  • Agents 3 × Claude + 1 × PowerShell
  • Level Starter

You'll learn: how to launch a four-pane preset in a 2x2 grid, move between its panes, zoom one pane to read it (Ctrl+B Z), and run a red, green, review loop by hand. A2A 3 automates this same loop.

Quay's window has two parts: the rail, the panel down the left that lists your projects and their terminals, and the stage to its right, where the terminal panes open. When the stage is split, each pane has a header along its top with the terminal's name and, for an agent, its status. Every Ctrl+B shortcut is two key presses: press Ctrl+B, let go, then press the next key within a second and a half. Each prompt the page gives you goes into the pane it names: click inside that pane, paste the prompt, and press Enter to run it or, in a Claude Code pane, to send it.

Before you start ​

1. Get the starter project ​

Do

  1. Download tdd-swarm-kata.zip and unzip it into a new, empty folder.
  2. In Quay, click Add at the top of the rail, click Browse..., pick that folder and click Add Project.
  3. Press Ctrl+B N, choose PowerShell Console and click Start Terminal (Enter).

Paste into PowerShell pane

text
git init -b main; git add -A; git commit -m Starter; npm install

You should see

The PowerShell pane in the project folder: git makes the Starter commit, then npm install finishes with no vulnerabilities.

Must match

  • git commit prints Starter as the message.
  • npm install ends with found 0 vulnerabilities.

May differ

  • The commit id, and the package counts and timings npm install prints.
If not: node or npm is not recognised

Node.js 24 or later isn't on your PATH. Install it as Node.js and npm shows, close and reopen Quay so its terminals see it, then paste the line again.

If not: git commit says Author identity unknown

Git doesn't know your name and email yet. Set them as Git shows, then paste the line again.

2. Launch the TDD Swarm ​

Do

  1. Press Ctrl+B S. The Swarm Launcher opens with Dev & Watcher selected.
  2. Click the TDD Swarm card.
  3. Click Launch Swarm Now.
  4. Wait until each of the three Claude Code panes has confirmed its role in one line.

You should see

The launcher closes and four panes open in a 2x2 grid: test-architect top left, feature-impl top right, test-runner bottom left at a PowerShell prompt, and reviewer bottom right. test-architect has confirmed its role in one line, feature-impl and reviewer are still starting, and a red banner across the top of the stage names reviewer.

Must match

  • Four panes in a 2x2 grid: test-architect top left, feature-impl top right, test-runner bottom left and reviewer bottom right.
  • The rail lists all four under the project, beside the PowerShell pane from step 1.
  • test-architect, feature-impl and reviewer each confirm their role in one line, then wait for your task.
  • test-runner sits at a PowerShell prompt.
  • While a Claude Code pane has focus, a red DANGEROUSLY SKIPPING PERMISSIONS banner runs across the top of the stage and names that pane. That is expected: swarm agents run in Skip mode.

May differ

  • The wording of each one-line confirmation, and how long each pane takes to start.
  • The port number in the rail header.
  • Which pane the red banner names, and whether it shows at all while test-runner has focus.
If not: a Claude Code pane asks you to accept Bypass Permissions mode

Quay starts a swarm's Claude Code agents in Skip mode, which Claude Code calls Bypass Permissions mode, and Claude Code asks once before its first session in that mode. Click into the pane that asks, choose Yes, I accept and press Enter. It then confirms its role. Do the same in any other pane that asks.

If not: a Claude Code pane shows the plugin hint

Quay isn't hearing from the event-bridge plugin. Install it as the hint says, then run /reload-plugins in that pane. The hint goes away as soon as Quay hears from the plugin.

If not: test-architect stops with an error about its model

Your Claude Code sign-in can't use claude-sonnet-5-5, the model the TDD Swarm preset starts test-architect on. Sign in to Claude Code with an account that can use it. Then close the four swarm terminals: in the rail, hover over each of test-architect, feature-impl, test-runner and reviewer and click the × that appears, Close terminal. Launch the swarm again from the start of this step.

3. Start the test watcher ​

Do

  1. Close the PowerShell pane from step 1, which you don't need again: in the rail, hover over pwsh under the project and click the × that appears, Close terminal.
  2. Click inside test-runner, the bottom-left pane. Its border lights up.

Paste into test-runner

text
npm run test:watch

You should see

test-runner: Vitest in watch mode lists the one todo test, rule 1: length, then waits for file changes.

Must match

  • Tests 1 todo (1).
  • A Waiting for file changes... line under it.

May differ

  • How long the run takes.
If not: vitest is not recognized

npm install didn't finish in step 1. Paste npm install into test-runner, wait for it to finish, then paste npm run test:watch again.

If not: test-architect says it hasn't run npm run test:watch

You pasted the line into test-architect, which never starts the watcher itself. Find the pane whose header says test-runner, click inside it, and paste the line there.

4. Round 1, red: test-architect writes the first tests ​

Do

  1. Click inside test-architect, the top-left pane.

Paste into test-architect

text
Read AGENTS.md. In test/strength.test.ts, replace the todo with failing Vitest tests for rule 1 (length) only: an 11-character password must fail it and a 12-character password must pass it, checking both score and failed. Pick passwords that pass rules 2 and 3, so these tests still hold once every rule is written. Do not touch src/.

You should see

test-architect says the tests should fail until feature-impl replaces the "not implemented" throw in src/strength.ts, and that it didn't touch src/. test-runner has re-run on its own and shows both tests failing.

Must match

  • test-runner re-ran on its own when test-architect saved the file.
  • Every test fails and none passes: the Tests line shows only a failed count.
  • test-architect names test/strength.test.ts as the file it changed, and nothing under src/.

May differ

  • How many tests test-architect writes, and their names.
  • The failure text, as long as every test fails.
  • test-architect's wording, and how long it takes.
  • test-architect may message feature-impl on its own, and feature-impl may answer that it's ready and waiting for you. Its rail line then starts with [Peer agent. Nothing changes in src/ until your prompt in step 5.
  • Grey text in a Claude Code pane's input box, such as Tell feature-impl the tests are ready, is a suggestion Claude Code offers. Ignore it and paste the next prompt as usual.
If not: a test already passes

That test doesn't call checkStrength, which still throws. Paste Every test must fail until src/strength.ts is written. Fix the ones that pass. Do not touch src/. into test-architect.

If not: test-runner never re-runs

Vitest re-runs when a file under the project changes. Check that test-architect edited test/strength.test.ts and didn't write a new file somewhere else.

5. Round 1, green: feature-impl makes them pass ​

Do

  1. Click inside feature-impl, the top-right pane.

Paste into feature-impl

text
Read AGENTS.md. Make every test in test/ pass with the simplest correct implementation of checkStrength in src/strength.ts, covering rule 1 only for now. Do not edit test/.

You should see

feature-impl says it left test/ alone and is ready for the next rule, and test-runner has turned green: both tests pass and none fails.

Must match

  • test-runner shows every test passing: the Tests line shows only a passed count.
  • feature-impl names src/strength.ts as the file it changed, and nothing under test/.

May differ

  • How feature-impl writes the code, and its wording.
  • How many times test-runner re-runs on the way.

6. Review round 1 ​

Do

  1. Click inside reviewer, the bottom-right pane, and paste the prompt below.
  2. When reviewer has replied, press Ctrl+B Z to zoom it to the whole stage, and read its findings.
  3. Press Ctrl+B Z again to put the grid back.

Paste into reviewer

text
Read AGENTS.md, src/strength.ts and test/strength.test.ts. Review them for correctness, missed edge cases and readability. Reply with a numbered list of findings, most important first. Do not edit any file.

You should see

reviewer fills the whole stage, showing a numbered list of findings.

Must match

  • reviewer fills the stage while zoomed.
  • Its reply is a numbered list.
  • Back in the grid, test-runner hasn't re-run since step 5, because reviewer changed no file.

May differ

  • The findings themselves, and how many there are.
  • A row of bracketed choices Claude Code prints under the reply, such as [ Continue ]. Leave them alone.
If not: Ctrl+B Z did nothing

The Ctrl+B prefix lasts a second and a half. Press Ctrl+B, let go, then press Z straight after.

If not: Ctrl+B Z zoomed another pane

Focus was on that pane, not reviewer. Press Ctrl+B Z to put the grid back, click inside reviewer, and zoom again.

7. Round 2, red: tests for the other two rules ​

Do

  1. Click inside test-architect.

Paste into test-architect

text
Read AGENTS.md again. Add failing tests for rule 2 (classes) and rule 3 (repeats), including the order of names in failed, and one test where a password passes all three rules. Keep the rule 1 tests as they are. Do not touch src/.

You should see

test-runner shows a mix: the rule 1 tests still pass and the new tests fail.

Must match

  • The Tests line shows both a failed and a passed count.
  • test-architect names test/strength.test.ts as the file it changed, and nothing under src/.

May differ

  • How many tests there are, and their names.
  • test-architect's wording, and how long it takes.
If not: every test passes

The new tests don't yet ask for anything src/strength.ts lacks. Paste Some of the new tests must fail until rules 2 and 3 are written. Fix the ones that pass. Do not touch src/. into test-architect.

8. Round 2, green: feature-impl writes every rule ​

Do

  1. Click inside feature-impl.

Paste into feature-impl

text
Read AGENTS.md. Make every test in test/ pass, implementing all three rules in src/strength.ts, then tidy the code without changing behaviour. Do not edit test/.

You should see

test-runner is all green again, with more tests passing than in step 5.

Must match

  • test-runner shows every test passing: the Tests line shows only a passed count, higher than in step 5.
  • feature-impl names src/strength.ts as the file it changed, and nothing under test/.

May differ

  • How feature-impl writes and tidies the code, and its wording.
  • The number of tests.

9. Sign-off ​

Do

  1. Click inside reviewer.

Paste into reviewer

text
Re-read AGENTS.md, src/strength.ts and test/strength.test.ts. If all three rules are implemented and tested, reply with exactly APPROVED as your last line. Otherwise reply CHANGES REQUESTED followed by a numbered list of what's missing. Do not edit any file.

You should see

reviewer's reply ends with the line APPROVED.

Must match

  • The last line of reviewer's reply is APPROVED.

May differ

  • Anything reviewer says before that line.
If not: reviewer replies CHANGES REQUESTED

Read its list, then hand the work on one agent at a time. You don't need to copy the list: each prompt below is one line, and each agent re-reads the files itself.

  1. If any item asks for a test, paste The reviewer asked for more tests. Re-read AGENTS.md and test/strength.test.ts and add the tests the three rules still lack. Do not touch src/. into test-architect, and wait until it says it's done.
  2. Paste The reviewer asked for changes. Re-read AGENTS.md, src/strength.ts and test/strength.test.ts, then fix src/strength.ts so all three rules are implemented and every test passes. Do not edit test/. into feature-impl, and wait until test-runner shows every test passing.
  3. Paste this step's prompt into reviewer again.

10. Check it worked ​

Do

  1. Click inside test-runner and press q to stop the watcher.

Paste into test-runner

text
npm test; npm run typecheck; git status --short

You should see

test-runner: npm test passes every test, the type check prints no errors, and git status lists src/strength.ts and test/strength.test.ts as the two modified files.

Must match

  • npm test shows every test passing: the Tests line shows only a passed count.
  • npm run typecheck prints no error after its > tsc --noEmit line.
  • git status --short prints two lines, M src/strength.ts and M test/strength.test.ts.

May differ

  • The number of tests, and the times beside them.
  • The folder in the PowerShell prompt, which is wherever you unzipped the starter.

You’re done when

  • ✅ npm test passes every test, and npm run typecheck reports no error.
  • ✅ reviewer's last reply ends with APPROVED.
  • ✅ git status --short lists src/strength.ts and test/strength.test.ts as the only modified files.

Released under the GNU Affero General Public License v3.0 (AGPL-3.0).

Copyright © 2026 Sean Robert Wright