PromptQA Recorder
Chat-driven Playwright test recorder for VS Code. Turn natural-language
instructions into Playwright tests structured as Page Object Model classes,
powered by Claude AI (BYOK).
Existing AI test-automation tools hallucinate selectors and produce flat,
brittle scripts. PromptQA Recorder does it differently:
a real headed browser executes every step before it's written to the file,
the LLM picks the intent while Playwright picks the selector
(getByRole, getByLabel, getByText), and the output is real POM code
you can edit, refactor, and own.
Highlights
- 🎙️ Record by chatting — describe actions in plain English, see them run
in a live browser, watch the spec and POM files grow line by line.
- 🤖 Robust locators — semantic resolution via Playwright's accessibility
tree. When a locator is ambiguous, the live a11y snapshot is fed back to
the LLM so it retries with a better description, not a worse selector.
- 📂 True Page Object Model — one POM per route, methods named from
intent (
clickSignInButton, fillEmailField), merged via ts-morph so
your manual edits survive.
- ✎ Edit mode after recording — close the browser and keep chatting:
read, refactor, and run tests with workspace-wide tools
(
list_files, read_file, apply_edit, run_test).
- 📎 File attachments — attach any project file via 📎 and ask Claude
to refactor it.
- 🔐 BYOK + zero telemetry — your Claude API key lives in VS Code
SecretStorage. The extension only talks to
api.anthropic.com.
Quick start
- Install the extension.
- Run PromptQA: Set Claude API Key and paste your key
(get one from the Anthropic console).
- Open the PromptQA Recorder view from the activity bar.
- Click Record (or Edit to refine existing files).
- Type instructions:
open https://playwright.dev
click the Get started link
assert the page contains 'Installation'
- Click Save & Run to verify with
npx playwright test.
Modes
| Mode |
What happens |
| Recording |
Headed Chromium open. 12 tools available: navigate, click, dblclick, hover, fill, check, select_option, press, screenshot, assert_visible / _text / _url. Every successful step is appended to the spec and POM live. |
| Editing |
No browser. 5 tools: list_files, read_file, apply_edit, run_test, resume_recording. Use it to refine, refactor, or run any workspace file. |
Type undo to revert the last recorded step. Click Stop Browser to
flip from recording to editing without losing conversation history.
Configuration
| Setting |
Default |
Description |
promptqa.model |
claude-sonnet-4-6 |
Claude model used for tool dispatch. |
promptqa.verboseExplanations |
true |
Plain-English commentary alongside generated code. |
Per-workspace config lives in .vscode/promptqa.json (multi-root supported):
{
"baseUrl": "https://staging.example.com",
"testsDir": "tests",
"pagesDir": "tests/pages",
"specsDir": "tests/specs",
"fixturesDir": "tests/fixtures"
}
Project layout it generates
tests/
├── pages/ # PageObject classes (one per route)
├── specs/ # Generated *.spec.ts files
├── fixtures/ # Test data + screenshots/
└── utils/
└── BasePage.ts
If the workspace doesn't already have Playwright, PromptQA offers to run
npm init playwright@latest (or pnpm/yarn based on your lockfile).
Privacy
- No telemetry. The extension makes network calls only to
api.anthropic.com.
- API key in SecretStorage, never written to settings.json.
- Files stay local — attachments are read by VS Code and sent only as
part of your Claude messages.
Requirements
- VS Code 1.90 or newer
- Node.js 20+
- A Claude API key (BYOK)
- A workspace folder (multi-root supported)
Known limits (v0.0.1)
- Bundle size is ~13 MB because
ts-morph carries the TypeScript compiler.
- No streaming yet — long Claude responses arrive in one go (with a typing
indicator while you wait).
- Recording produces one POM file per route; complex SPAs may want to be
split further by hand.
License
MIT — see LICENSE.
| |
