Testing

Required checks#

pnpm run check

This runs strict contract checking, source-reference checks, skill-manifest validation, and the regression suite. Tests use Node's built-in node:test. Add focused regressions for connector, parser, cache, permission, and security-boundary changes.

Browser UI checks#

Install the WebKit test browser if needed, then run the workspace checks:

pnpm exec playwright install webkit
pnpm run check:ui

The checks use an isolated local service, fictional data, and fixtures for personal sources. They cover layouts, theme modes, window sizes, and interactions such as Notes drag-and-drop and contextual editing.

To focus a layout investigation on one workspace:

HABIBI_UI_WORKSPACE=notes pnpm run check:ui

Native checks#

After building the client, macOS can test the teleprompter with assets copied into a real app bundle:

native/smoke-test-teleprompter.sh

This catches packaged resource URL and rendering issues that a source-only browser test can miss. Other native/smoke-test-* scripts cover their named native boundaries; inspect their inputs before running a packaged-release or account-dependent check.

Safe screenshots#

Repository screenshots use the sealed fictional-data demo mode:

pnpm exec playwright install webkit
pnpm generate:demo-screenshots

To refresh selected captures, pass their filenames without the extension:

pnpm generate:demo-screenshots --only=emoji-picker,clipboard-history,focus-timer

The capture script uses a temporary data root and browser profile. Browser-only fixtures in scripts/demo-capture-fixtures.mjs provide fictional agent sessions, cluster resources, and Mac utilities over the sealed demo. No native clipboard, Reminders, timer, terminal, or cluster connection is used. Window captures have transparent corners; the documentation theme supplies their shadows.

Never record real mail, clipboard contents, chat exports, keys, or private screens into committed fixtures. Keep personal data directories outside the test service's data root.

Documentation checks#

Run pnpm run docs:check for page metadata, links, assets, and MDX compilation. Run pnpm run docs:build for a deployable static site. See Maintain the docs.

AI file workflow tests#

The live fixture suite passed with gpt-6-luna on 6 November 2026. It checked:

• Forty-file content-based renaming, actual apply, and Undo.
• Forty-file folder organization, actual apply, and Undo.
• Image compression and a valid PDF merge preview.
• Invoice CSV extraction grounded in fictional PDF text.
• Evidence-based document selection, storage explanation, and duplicate reporting.
• A seventeen-minute timer and a dated Apple Reminder proposal.
• Opted-in duplicate cleanup with a real Trash/Undo round trip.
• AI-selected PDF page order and rotation with a valid output preview.

The native fixture suite separately checks PDF page order, rotation, extraction, compressed copies, image dimensions, overwrite refusal, and a Trash/restore round trip. Browser checks cover @ selection, content-sharing controls, cancellation, real temporary-file rename/Undo, responsive layouts, and reviewed inline utility actions. These checks establish the supported paths; they do not guarantee every model response or every document format will produce a useful plan.

Run the live suite after building and configuring an OpenAI key in Habibi:

HABIBI_TEST_MODEL=gpt-6-luna node scripts/check-file-tools-live.mjs

It uses generated fixtures only. The key stays in Keychain, is never printed, and is never written to the fixture configuration.