# Borg SSH guidance checks These standalone checks are separate from the Python i18n suite. They do not contact a Proxmox host or generate SSH keys. ## JSX/provider regression Prerequisites: Node **22.14+** and installed `AppImage` dependencies (including React, React DOM and TypeScript). From the repository root: ```bash node tests/test_borg_ssh_guidance.cjs ``` The test extracts the actual SSH branch from `AddDestinationDialog`, renders it with React, and executes the actual provider's lookup/interpolation callback. Basic control wrappers replace shadcn components only in this fast test. It covers all shipped catalogs, unconditional missing-key English fallback, synthetic whole-message translations and reordered placeholders, initial/result/loading states, account escaping, blank/custom paths and literal public-key output. The optional `account` argument isolates the remote-account instruction. ## Real-component browser fixture Additional prerequisites: `esbuild`, `playwright`, its Chromium browser and a successful full Monitor frontend build. The validation runtime used esbuild 0.28.2 and Playwright 1.63.0. If these optional tools are not installed, install them locally without changing the project manifests/lockfile: ```bash cd AppImage npm install --no-save --package-lock=false --legacy-peer-deps esbuild@0.28.2 playwright@1.63.0 node node_modules/playwright/cli.js install chromium npm run build cd .. node tests/test_borg_ssh_browser.cjs /absolute/path/to/borg-browser-evidence ``` The fixture bundles the actual complete `AddDestinationDialog`, real shadcn components, real API helper and `I18nProvider`, with CSS from `AppImage/out` by default. Set `BORG_SSH_CSS_DIR` to the `_next/static/css` directory of an archived full build to reuse that build's CSS. A test-only in-memory export exposes the unexported dialog; production source is not rewritten. Browser locale fixtures omit the six new keys from Italian and supply expanded/reordered synthetic messages in German **in memory only**. They are not proposed translations. All requests are intercepted **before navigation**. Known static resources and read endpoints are fulfilled from fixtures; key preparation POSTs receive inert fresh/existing/error responses. All other requests, including Save, are aborted and make the run fail. No server is needed. The test checks current request payloads and characterizes unchanged last-response retention after field edits and errors; mocked fresh/existing responses are not a backend generation test. The matrix is desktop/mobile (1440×1000, 390×844), light/dark, English/forced Italian fallback/synthetic expansion. It writes initial/result screenshots and `browser-results.json`, and asserts message/button horizontal fit. The modal is scrolled to show the relevant panel; this is not whole-dashboard acceptance. ## Other gates ```bash python3 -m unittest discover -s .github/scripts/tests -v cd AppImage node node_modules/typescript/bin/tsc --noEmit --incremental false ``` Run typechecking separately: the production build skips it. Compare diagnostics against the unchanged baseline with the same dependencies; a successful build is not a clean typecheck. CONTRIBUTING's real-Proxmox deployment smoke test is a separate integration gate and is not replaced by these fixtures.