Coding agents can write frontend code but can't easily see what they built. Layout QA is a small local runner that closes that loop:
1. The agent wires deterministic API/auth fixtures into your app behind a local env flag (the README includes a setup prompt for this), so runs don't depend on live services.
2. The agent declares flows in a small JSON manifest — clicks, fills, text checks, and layout assertions like "no horizontal overflow" or "this button is in the viewport" — and runs them via the CLI at desktop/tablet/mobile viewports.
3. Each run captures screenshots and writes a local HTML report the agent can read to verify and fix its own work. Exit code 0/1, so the same command works in CI.
It's a thin layer on Playwright: the goal isn't to replace it, but to make the see-act-verify loop simple and deterministic enough that an agent can set it up and iterate on it unattended. No account, no hosted service, no Storybook. Example output in the README.
1. The agent wires deterministic API/auth fixtures into your app behind a local env flag (the README includes a setup prompt for this), so runs don't depend on live services. 2. The agent declares flows in a small JSON manifest — clicks, fills, text checks, and layout assertions like "no horizontal overflow" or "this button is in the viewport" — and runs them via the CLI at desktop/tablet/mobile viewports. 3. Each run captures screenshots and writes a local HTML report the agent can read to verify and fix its own work. Exit code 0/1, so the same command works in CI.
It's a thin layer on Playwright: the goal isn't to replace it, but to make the see-act-verify loop simple and deterministic enough that an agent can set it up and iterate on it unattended. No account, no hosted service, no Storybook. Example output in the README.