| # Task-list clipping regression | |
| From `apps/desktop`, after the locked workspace dependencies and Playwright Chromium are installed: | |
| ```sh | |
| npm run test:task-scroll | |
| ``` | |
| The two tests mount the production `ComposerStatusStack` (including status grouping, | |
| section collapse, task rows and the production stylesheet) with 14 and 20 synthetic | |
| task records. Real Chromium wheel input must expose and hit-test the final row | |
| inside the capped viewport, above a composer boundary. Collapse and re-expand must | |
| preserve scrolling. No inline CSS treatment or `scrollIntoView` repairs the page. | |
| The fixture uses no backend, model, native Electron bridge or user session data; | |
| it is a renderer-component regression, not a native-app end-to-end test. | |
| The runner serves only loopback port 18120 and cleans up its browser and Vite | |
| server. Its Vite cache is private. Set `TASK_SCROLL_OUTPUT` to retain the JSON | |
| geometry and screenshots in a chosen directory. Otherwise these are retained in | |
| the printed test run's temporary directory. In shared campaigns wrap the command | |
| with the campaign's `flock tests.lock`, and isolate `HOME` and `HERMES_HOME`. | |