← ClaudeAtlas

debug-kitlisted

Cross-platform app debug and testing skill for Codex and Claude Code. Build, launch, screenshot, tap, type, inspect, and monitor applications across 9 platforms: Electron, iOS, macOS, Web, Flutter, React Native, Android, Tauri, Chrome Extension. Auto-detects project type from package.json, pubspec.yaml, xcodeproj, build.gradle, src-tauri/, manifest.json (MV3). TRIGGER when: user says "run the app", "test the app", "take screenshot", "tap button", "debug", "launch", "check if it works", "hot reload", "monitor console", "check logs", or is working on ANY app project and wants to build, run, test, or verify changes. Also trigger when the user asks to interact with a running app, inspect UI elements, check performance, run accessibility audits, or simulate user input.
dxiongya/debug-kit · ★ 0 · Code & Development · score 73
Install: claude install-skill dxiongya/debug-kit
# Debug Kit Cross-platform app debug toolkit. One skill to build, launch, interact with, and inspect apps across 9 platforms. ## Evidence-driven test loop Use the smallest loop that proves the requested behavior: reproduce → inspect state → change when authorized → build → focused tests → actual UI verification. A successful build is not evidence that the interaction works. - Before mutating a running app, identify the exact instance and its data store. Use disposable fixtures. For desktop apps, account for shared resources such as the clipboard, global shortcuts, login items, and sync; a separate process or bundle ID alone does not isolate them. - Write observable checks for the happy path, cancellation/undo, persistence, and neighboring behavior. For a small visual-only change, a build and focused visual check may suffice. - Read the accessibility tree to locate controls; use screenshots for layout, icons, contrast, and clipping. After an action, refresh state before choosing the next target. Prefer scoped, stable accessibility identifiers over repeated labels or coordinates. - If automation fails, distinguish a product failure from targeting, focus, permission, or runner failure. After one reasoned correction and retry, change the verification method or report the blocker instead of repeating the same failed event sequence. - Keep automated-test results and manually verified UI results separate. Record the exact build, checks, outcome, and remaining uncertainty; do not