test-setup

Test Setup

One-command bootstrap. After it finishes, /CreateTest works immediately — no extra setup steps. Hub publishing and CI dispatch are configured later, by their own skills, when first used.

The skill never copies from a reference project. Every scaffold file lives in .claude/skills/test-setup/assets/; the skill reads from there, substitutes user-supplied values, and writes into the project.

Two Phases

1. Phase A — Scaffold. Audit which files are missing. Prompt once for project-specific values. Read each missing template from assets/, substitute placeholders, write to the project. Never overwrite existing files without explicit confirmation.
2. Phase B — Install & verify. Node ≥ 18 (hard), npm install, Playwright Chromium, Java runtime for Allure (hard), then smoke-test the runner.

Run phases sequentially. Within Phase A, ask for values once, then write every file in parallel.

At the end, print the summary defined under Finishing reply.

---

Phase A — Scaffold

A0. Audit what's present

Check each path relative to the repo root. Build the missing list — don't touch present files yet.

Path	Asset (under .claude/skills/test-setup/assets/)	Placeholders
CLAUDE.md	CLAUDE.md.template	__APP_NAME__, __APP_URL__, __ENV__, __ADMIN_USERNAME__, __ADMIN_PASSWORD__, __PROJECT_NAME__
package.json	package.json.template	__PROJECT_NAME__
playwright.config.ts	playwright.config.ts	__APP_URL__
tsconfig.json	tsconfig.json	—
.gitignore (must contain test entries)	gitignore.snippet	—
scripts/run-plan.js	scripts/run-plan.js	—
scripts/build-dashboard.js	scripts/build-dashboard.js	__APP_NAME__, __ENV__
scripts/generate-allure-results.js	scripts/generate-allure-results.js	__APP_NAME__, __APP_URL__
scripts/sync-to-hub.js	scripts/sync-to-hub.js	__PROJECT_NAME__
scripts/hub.config.example.json	scripts/hub.config.example.json	—
test-plans/RULES.md	test-plans/RULES.md	—
test-plans/ directory	(create if absent)	—
test-reports/ directory	(create if absent)	—

Print a one-line audit so the user sees what's about to happen:

Scaffold audit: present=<count>, missing=<list>. Will prompt for values, then write the missing files.

A1. Collect values (one or two prompts max — keep it quick)

If every placeholder-bearing file is already present (CLAUDE.md + package.json + playwright.config.ts + build-dashboard.js + generate-allure-results.js + sync-to-hub.js), skip this step — there's nothing to substitute.

Otherwise:

Prompt 1 — single AskUserQuestion call with these four questions:

1. Project slug (header "Project slug") — kebab-case used in package.json and the hub project folder. Derive 2–3 sensible defaults from the repo folder name (e.g. cwd c:\Projects\Acme-Portal → acme-portal, acme, acmeportal).
2. App name (header "App name") — human-readable. Suggest title-cased slug + " Admin Portal", " Web App", or just title-cased slug.
3. App URL (header "App URL") — provide one placeholder option; user picks "Other" to type it.
4. Environment (header "Environment") — options: QA, Dev, Staging, Production.

Prompt 2 — second AskUserQuestion call for admin credentials. These are separate so the user explicitly opts into providing them:

1. Admin username (header "Admin username") — defaults: admin, the slug, "Other".
2. Admin password (header "Admin password") — options: "Type a password (Other)", "Skip — I'll fill it in CLAUDE.md myself" (writes <TODO: fill in>; /CreateTest will refuse to run until it's set), "Use placeholder (CHANGE-ME)" (writes CHANGE-ME, warned in the summary).

Never invent or echo back real credential values. Capture-only.

If CLAUDE.md is present and would otherwise need rescaffolding, ask once:

CLAUDE.md already exists. Overwrite it with the templated version? This replaces your current ## Application Under Test and ## Credentials sections.

Default No. If declined, leave CLAUDE.md alone and print the snippet in A4 so the user can paste it manually.

A2. Substitute and write

For every file in Missing, in parallel:

1. Read the corresponding asset from .claude/skills/test-setup/assets/.
2. Replace each __PLACEHOLDER__ with the captured value.
3. Write to the target path.

For directories (test-plans/, test-reports/): create them with a .gitkeep so they survive a fresh git clone.

For .gitignore: if the file exists and lacks the test entries (look for node_modules/ + allure-results/ + playwright-report/), append gitignore.snippet. If .gitignore doesn't exist, write the snippet as the whole file.

A3. Verify substitution

Grep "__[A-Z_]+__" across CLAUDE.md, package.json, playwright.config.ts, scripts/

Any hit means a placeholder leaked. Stop and tell the user which file + which placeholder; do not proceed to Phase B.

A4. If CLAUDE.md was kept, print the snippet for manual paste

When the user declined to overwrite an existing CLAUDE.md, print the substituted sections they need to add:

## Application Under Test
| Key | Value |
|-----|-------|
| App | <APP_NAME> |
| URL | <APP_URL> |
| Environment | <ENV> |

## Credentials
| Role | Username | Password |
|------|----------|----------|
| Admin | <USERNAME> | <PASSWORD> |

Tell them: "/CreateTest reads these sections at authoring time. Paste them into CLAUDE.md before invoking it."

---

Phase B — Install & verify

Run these sequentially.

B1. Node ≥ 18 (hard requirement)

node --version

• Healthy: v18.x or higher.
• Hard missing: stop and tell the user:

  Node 18 or newer is required. Install from https://nodejs.org/ (LTS), restart this shell, then re-run /test-setup.

B2. npm dependencies (soft missing — installable)

test -d node_modules/@playwright/test && test -d node_modules/allure-commandline && echo OK

• Healthy: OK.
• Soft missing: ask once Install npm dependencies now? (npm install). On yes, run npm install. On no, mark skipped and continue — Phase B3/B4/B5 will fail loudly.

B3. Playwright Chromium (soft missing)

npx playwright install --dry-run chromium 2>&1

If the output contains is already installed, the browser is present. Otherwise ask once, then run npx playwright install chromium.

B4. Java runtime for Allure CLI (hard requirement)

java -version

• Healthy: a version string was printed (Java prints to stderr but exits 0).
• Hard missing: stop and tell the user:

  Allure CLI needs a Java runtime (JRE 11 or newer). Install one of:

  • Eclipse Temurin: https://adoptium.net/temurin/releases/
  • Microsoft OpenJDK: https://learn.microsoft.com/java/openjdk/download

  After install, open a new shell so java is on PATH, then re-run /test-setup.

Do not attempt to install Java via winget, choco, or any other package manager.

B5. Playwright MCP server (soft missing — required for /CreateTest)

/CreateTest records selectors live by driving a browser through the Playwright MCP server. The server must be registered with Claude Code before any plan can be authored.

Check whether it's already registered:

claude mcp list 2>&1

Look for a line beginning with playwright. If present, mark this step ✓ and move on.

If absent, ask once:

Register the Playwright MCP server now? Runs claude mcp add playwright -- npx -y @playwright/mcp@latest --headless. Required for /CreateTest. The --headless flag keeps the recording browser hidden so it never steals focus while you work.

On yes, run:

claude mcp add playwright -- npx -y @playwright/mcp@latest --headless

Then re-run claude mcp list to confirm playwright is now listed. If the add command failed, surface the stderr verbatim — common causes are claude CLI not on PATH or a stale lockfile in ~/.claude/. Do not retry silently.

On no, mark this step skipped and warn the user in the finishing summary that /CreateTest will fail until they register the server.

B6. Smoke-test the runner

node scripts/run-plan.js --help 2>&1 | head -20

• Healthy: the runner prints usage or exits cleanly on the missing argument.
• Failed: surface the error verbatim. Usually means B2 was skipped or npm install partially failed — re-run it.

---

Finishing reply

Pick the line that matches what actually happened. Do not claim "complete" if any hard step failed.

If Phase A scaffolded files and Phase B passed:

Setup complete. Scaffolded <n> file(s). Prerequisites verified (Node ✓, deps ✓, Chromium ✓, Java ✓, Playwright MCP ✓, runner ✓). Next: /CreateTest to author a test.

If Phase A found nothing missing and Phase B passed:

Setup verified — nothing scaffolded. Prerequisites ✓. Ready for /CreateTest.

If something blocking failed:

Setup blocked at Phase <A|B>, step : . . . Re-run /test-setup once it's resolved.

If the user picked the CHANGE-ME password placeholder in A1:

⚠️ The Admin password in CLAUDE.md is CHANGE-ME. Edit it before running /CreateTest.

If the user picked Skip on the admin password:

⚠️ The Admin password in CLAUDE.md is <TODO: fill in>. Replace it before running /CreateTest.

If the user declined to register the Playwright MCP server in B5:

⚠️ Playwright MCP server not registered. /CreateTest will fail until you run claude mcp add playwright -- npx -y @playwright/mcp@latest --headless.

Do not mention hub config or CI prereqs in the summary. Those are configured separately by /submit-test-results and /Run-test-remote when first invoked.

---

Idempotence guarantees

• Never overwrite a present file without explicit user confirmation.
• Never write real credentials that weren't typed by the user in this session.
• Re-running on a healthy project produces "Setup verified — nothing scaffolded" and exits in seconds.
• Re-running after a partial setup picks up only the missing files.