← ClaudeAtlas

qa-visuallisted

Build, repair, run, or review Token Tab visual regression coverage without installing the app. Use whenever a request mentions visual QA, screenshot or golden-image tests, UI regressions, rendered app previews, visual diffs, menu-bar appearance, SwiftUI snapshot testing, or asks whether a UI change looks correct. Prefer the existing fixture-driven off-screen renderer; use real menu-bar automation only for interaction behavior that static rendering cannot prove.
YiftachCohen/token-tab · ★ 2 · Testing & QA · score 66
Install: claude install-skill YiftachCohen/token-tab
# Token Tab visual QA Treat a visual test as a regression contract for a named UI state, not as a marketing screenshot or a substitute for behavior tests. The target is reliable evidence from the real shipping SwiftUI views, without installing Token Tab, reading a person's local logs, or adding a snapshot-test dependency. ## Operating boundaries - Read `AGENTS.md` before editing. Its trust audit, zero-dependency rule, and design-lint ratchet still apply to UI-test work. - Do not package, install, launch at login, or register the app just to perform visual QA. Build and run the Swift test bundle instead. - Do not use a real `~/.claude` or `~/.codex` directory for a visual fixture. Fixture data makes every capture safe to share and reproducible. - Do not add an npm package, SwiftPM dependency, or a generic screenshot library. Use XCTest, AppKit, and the repository's existing renderer. - Do not silently replace or bless a baseline. A changed baseline is a reviewed product decision; show the diff and wait for explicit approval before updating it. ## First classify the request 1. **Static visual state** — a dropdown tab, light/dark appearance, empty/error state, metric, or provider combination. Use the off-screen renderer and a baseline comparison. 2. **Shell interaction** — status-item opening, dismissal, menu-bar placement, keyboard focus, or settings interactions. Static images cannot prove these. If available, use macOS UI automation for a focused sm