testing-windows-desktop — Claude Code skill: Windows E2E testing
testing-windows-desktop is a Claude Code skill for end-to-end testing of the pi-gui Electron app on Windows. It validates dev server startup, folder pickers, and terminal sessions.
MIT ★ 364 updated 2026-11-17
npx skills add raylee-ai/lancer --skill testing-windows-desktop -a claude-codeWhat testing-windows-desktop does
The skill manages end-to-end testing for the pi-gui Electron client on Windows environments. It launches the application via pnpm --filter @lancer/desktop dev, checking for clean startup without spawn pnpm ENOENT failures and confirming the renderer server at http://localhost:5173. When driving native Windows file dialogs, it pastes paths using Set-Clipboard and Ctrl+V to prevent automated typing tools from dropping colons or uppercase drive letters.
Opening the integrated node-pty terminal in the live GUI requires an active session, which is blocked when provider credentials are missing. The skill works around this by running a headless Playwright test in background mode using launchDesktop(userDataDir, { testMode: "background" }), requiring no API tokens.
All terminal assertions use cross-platform commands like echo. POSIX commands such as printf or pwd are avoided because the default Windows shell resolves to cmd.exe.
What it helps with
- run end-to-end checks on the pi-gui Electron desktop client under Windows
- catch launcher issues and
spawn pnpm ENOENTerrors during dev startup - bypass character-dropping bugs when filling Windows folder dialog paths
- verify integrated terminal functionality without configured model keys
Example prompts for the agent
- Run end-to-end tests for the pi-gui desktop application on Windows
- Test the integrated node-pty terminal in background headless mode
- Verify native folder picker handling in the @lancer/desktop Windows build
How to install testing-windows-desktop
1Skills CLI (Vercel)
npx skills add raylee-ai/lancer --skill testing-windows-desktop -a claude-code2Manually: put the files in place
~/.claude/skills/testing-windows-desktop/SKILL.mdCopy the whole skill folder; for one project use .claude/skills/testing-windows-desktop/
1Skills CLI (Vercel)
npx skills add raylee-ai/lancer --skill testing-windows-desktop -a codex2Manually: put the files in place
.agents/skills/testing-windows-desktop/SKILL.mdFor all projects use ~/.agents/skills/
1Skills CLI (Vercel)
npx skills add raylee-ai/lancer --skill testing-windows-desktop -a cursor2Manually: put the files in place
.cursor/skills/testing-windows-desktop/SKILL.mdCursor also reads .agents/skills/ and .claude/skills/
1GitHub CLI
gh skill install raylee-ai/lancer .agents/skills/testing-windows-desktop2Manually: put the files in place
.github/skills/testing-windows-desktop/SKILL.md1Skills CLI (Vercel)
npx skills add raylee-ai/lancer --skill testing-windows-desktop -a gemini-cli2Platform CLI
gemini skills install https://github.com/raylee-ai/lancer --path .agents/skills/testing-windows-desktopManually: put the files in place
.devin/skills/testing-windows-desktop/SKILL.mdThe legacy .windsurf/skills/ path still works
Claude.ai
Zip the folder and upload it in Settings → Capabilities → Skills
Step by step, with commands for each platform: How to install skills in Claude Code: five ways
Questions about testing-windows-desktop
How to install testing-windows-desktop in Claude Code?
Run npx skills add raylee-ai/lancer --skill testing-windows-desktop -a claude-code.
What does testing-windows-desktop do?
It guides automated and headless verification of the Electron desktop GUI, system folder dialogs, and node-pty terminal on Windows.
Are API credentials required to run testing-windows-desktop?
No credentials are needed for app launch, UI navigation, or headless Playwright tests. An LLM provider key is only required if you test terminal sessions directly through the live GUI.
Source text
SKILL.md · MIT
---
name: testing-windows-desktop
description: End-to-end test the pi-gui Electron desktop app on Windows. Use when verifying Windows compatibility (dev launcher, PATH, folder picker, integrated terminal / node-pty) or any apps/desktop change on Windows.
---
# Testing the pi-gui desktop app on Windows
## Setup
- Ensure pnpm is available: `corepack enable` then `corepack prepare pnpm@<version from package.json> --activate`.
- Install: `pnpm install` (repo root).
- Launch the app for GUI testing: `pnpm --filter @lancer/desktop dev`. Optionally set `PI_APP_USER_DATA_DIR` to an isolated temp dir.
- A working launch prints `starting electron app...` and the renderer dev server (`http://localhost:5173`) and must NOT print `spawn pnpm ENOENT`.
## What renders where
- Empty state: sidebar (New thread / Threads / Skills / Extensions / Settings) + "Open a folder to start".
- Navigation (Settings / Skills / Threads) is a good cheap proof the renderer is interactive.
## Integrated terminal (node-pty / ConPTY) — IMPORTANT gotcha
- The Terminal item in the Open side panel menu is disabled until there is an **active session**: `terminalAvailable={Boolean(selectedSessionKey)}` in `apps/desktop/src/app/App.tsx`.
- Creating a session through the live UI requires a **connected provider/model**. Without provider credentials you CANNOT reach the terminal via the GUI ("No models available" blocks send).
- Workaround that needs no credentials: run a **headless Playwright spec in background test mode**. Use helpers from `apps/desktop/tests/helpers/electron-app.ts`: `launchDesktop(userDataDir, { initialWorkspaces, testMode: "background" })` -> `createNamedThread(window, ...)` -> open Terminal from the Open side panel menu (or press Ctrl+J) -> type a command -> assert `.xterm-rows` text.
- Use a **cross-platform command**: `echo <marker>` works in both `cmd.exe` (Windows default shell via `defaultShellForPlatform()`) and POSIX shells. The existing `tests/core/integrated-terminal.spec.ts` uses `printf`/`pwd`, which do NOT exist in `cmd.exe` — that lane runs on Linux and macOS CI only, so don't expect it to pass as-is on Windows.
- Run a single spec: `pnpm --filter @lancer/desktop run test:e2e:runner -- apps/desktop/tests/core/<spec>.spec.ts`. Delete any temporary spec you add after the run.
## Native Windows folder picker
- "Open first folder" opens the real Windows folder dialog (itself a Windows-compat proof).
- Computer-use `type` action may DROP uppercase letters and `:` in the dialog's path field (e.g. `C:\Users\Administrator` becomes `\sers\dministrator`). Workaround: put the path on the clipboard (`Set-Clipboard -Value "<path>"`) and `Ctrl+V` into the Folder field, then Select Folder.
## Cosmetic (not functional) macOS-centric labels on Windows
- Settings shortcut hints may show `Cmd+…` and the integrated-terminal shell field placeholder may read `/bin/zsh`. Display-only; the actual default shell resolves to `cmd.exe`.
## Reporting
- Record GUI interactions with annotations (setup / test_start / assertion). Do not record shell-only runs.
- Mark the live-UI terminal test `untested` (not failed) when blocked by missing provider credentials, and note the headless spec result separately.
## Devin Secrets Needed
- None required for launch / navigation / folder picker / headless terminal test.
- A provider API key (e.g. an LLM provider configured under Settings > Providers) is required only to create a session through the **live UI** and thus exercise the terminal through the GUI.
© raylee-ai · MIT · raylee-ai/lancer