feat: add Claude Code skills and automated QA CI pipeline ## Summary Automated QA pipeline that reproduces reported bugs using Playwright E2E tests. Claude Sonnet 4.6 writes tests, runs them, and records video evidence. Badges show `via E2E test` / `via video` to indicate how bugs were proven. ## How It Works
Test focus: Verify that the frontend UI remains fully functional and unaffected by the CI/CD additions. Prerequisites: Load default workflow Steps: Open the main menu → Hover over the File menu → Take a screenshot of the UI to confirm stability
# linux QA Video Report
- Generated at: 2026-04-02T14:25:55.312Z
- Model: `gemini-3-flash-preview`
- Target: https://github.com/Comfy-Org/ComfyUI_frontend/pull/9430
- Before video: `./qa-artifacts/qa-report-Linux-23905050701/qa-before-session.mp4` (2.6 MB)
- After video: `./qa-artifacts/qa-report-Linux-23905050701/qa-session.mp4` (2.5 MB)
- Mode: **Comparative (before/after)**
## AI Review
## Summary
This PR introduces a robust automated QA pipeline and Claude Code skills for issue reproduction and regression testing. The PR branch includes scripts for test generation, video recording, and CI integration.
The provided videos demonstrate the automated QA system (using the newly added `qa-record.ts` infrastructure) executing a smoke test on both the **BEFORE** (main) and **AFTER** (PR) versions of the application.
- The test covers: User login, opening/browsing the "Templates" modal, and navigating the main "File" menu.
- **Comparison**: The behavior and UI are identical in both videos. This confirms that the new QA infrastructure is compatible with the existing frontend and that no regressions were introduced to these core functional areas.
- **Bug Fixes**: While the PR description mentions fixing several specific issues (e.g., #10688 regarding preview images on tab switch), the provided videos do not demonstrate those specific scenarios (e.g., no tab switching occurs). Therefore, the individual bug fixes cannot be verified from these videos alone.
## Behavior Changes
| Behavior | Before (main) | After (PR) | Verdict |
| :--- | :--- | :--- | :--- |
| **Login Flow** | Successfully logs in with user "qa-ci". | Successfully logs in with user "qa-ci". | `No Change` |
| **Template Browser** | Opens modal; preview images and videos for "Getting Started" templates load correctly. | Identical behavior; templates load and display correctly. | `No Change` |
| **File Menu** | Displays options: New, Open, Save, Save As, Export, Export (API). | Identical menu structure and item availability. | `No Change` |
| **Nodes 2.0 BETA Badge** | Visible in the main menu with blue styling. | Visible with identical styling. | `No Change` |
| **Automated Recording** | Automated cursor movement and sequence execution functional. | Automated cursor movement and sequence execution functional. | `No Change` |
## Timeline Comparison
| Time | Type | Severity | Before (main) | After (PR) |
| :--- | :--- | :--- | :--- | :--- |
| 0:00-0:03 | State | None | Login screen presented and "Next" clicked. | Identical login sequence. |
| 0:05-0:08 | Visual | None | Templates modal opens; images and video thumbnails for "1.2 Starter" load/play. | Identical loading and playback behavior. |
| 0:12-0:14 | Menu | None | Main "File" menu navigated via automation. | Identical menu navigation and layout. |
## Confirmed Issues
No issues confirmed. The PR branch behaves identically to the main branch in the tested paths.
---
## Possible Issues (Needs Human Verification)
- **Bug Fix Verification**: The PR claims to fix issues like #10688 (Preview images cleared on tab switch) and #10708 (Collapsed subgraph text widget). These behaviors were not exercised in the provided smoke test videos. A targeted test for these specific bugs should be verified separately.
## Overall Risk
**Low**. The PR primarily adds developer tooling (scripts, CI workflows, and Claude skills). The small changes to frontend components (`LazyImage.vue`, `dialogStore.ts`) do not appear to impact core stability or the appearance of the template browser/menus. The consistency between the BEFORE and AFTER videos suggests high stability for the new infrastructure.
## Verdict
{"verdict": "INCONCLUSIVE", "risk": "low", "confidence": "high"}