09 AI writing path

Goal

Produce one small, idiomatic test (or plan section) using skills and thin agent prompts, with evidence and the review rubric.

Modules 01 to 08 already assume you may use a coding agent. This module deepens the structured plan, generate, and heal loop.

Read first

• AI testing guide
• Agent house style

Path A: Explore with CLI (default)

npm run dev
# keep that process running, then in another terminal:
npx playwright cli open http://127.0.0.1:3000/ --headed
npx playwright cli snapshot

If the CLI reports ERR_CONNECTION_REFUSED, start (or restart) npm run dev so ports 3000 / 4000 are up before exploring.

Draft a short test that matches manage-lists-* style. Run it with --project="logged-in chrome" when it uses list fixtures. If it fails, open a trace before editing.

Path B: Planner, generator, heal

Keep scope to one feature slice. Example plan section:

## Scenarios

### Open share dialog

**Steps**
1. Open a seeded list on View List.
2. Click Share.
3. Expect a dialog with a share URL or copy control.

**Expected**
- Share UI is visible with a role-based locator

On a clone, use these thin prompts (names matter):

Step	Prompt
Plan	.github/prompts/playwright-test-plan.prompt.md
Generate one scenario	.github/prompts/playwright-test-generate.prompt.md
Heal	.github/prompts/playwright-test-heal.prompt.md

Rewrite toward house style with the movies-playwright skill, then heal with trace evidence. Pass the review rubric.

Done when

• Path A or Path B produced one small idiomatic test (or a scoped plan section).
• Role locators + fixtures/helpers; no Codegen / brittle waits.
• The review rubric passes for what you landed.

Extra practice: Exercise 03 — plan a section with AI.

Key takeaways

• You chose CLI explore vs test agents deliberately.
• Output uses role locators and existing helpers or fixtures.
• No waitForTimeout, force: true, networkidle, or Codegen.
• The review rubric passes for what you landed.

Bonus: 10 Sharding.