obsidian-host-test

name	obsidian-host-test
description	Verify Bases Paginator changes in a real, running Obsidian app through the official CLI and CDP. Use after user-visible, data-flow, lifecycle, rendering, accessibility, CSS, or mobile changes need host evidence beyond Vitest, or when replacing Computer Use with a repeatable disposable-vault test.

Obsidian Host Test

Boundary

• Target only an explicitly named disposable vault. Never infer or use the active or primary vault.
• Keep note, frontmatter, Canvas, and .base content byte-for-byte unchanged. The runner may replace only this plugin's installed package and Obsidian-owned workspace/plugin settings.
• Exercise the real plugin view. A successful base:query alone does not verify pagination.
• Prefer CLI DOM, CDP input/accessibility, console capture, and screenshots. Use Computer Use only for OS dialogs or window chrome; leave real screen-reader and physical-device checks manual.

Workflow

1. Run obsidian help and confirm the installed CLI exposes the commands checked by the bundled runner, including base:query, plugin lifecycle, reload, debug/error capture, DOM, screenshot, mobile, CDP, and eval commands.

2. Run the headless CI gate first: npm ci, npm run fmt:check, npm run lint, npm test, and npm run build.

3. Prepare or refresh the tracked fixture only in an explicitly named disposable vault:

   npm run test:host:setup -- --target "/absolute/path/to/obsidian-test-vault" --adopt
   

4. Run the authoritative host entry against the same registered vault:

   npm run test:host -- --vault obsidian-test-vault --scenarios tests/fixtures/host-vault/scenarios.json
   

5. Treat any nonzero exit as a failed host gate. Inspect the reported stage, error buffers, and screenshots under .oma/obsidian-host-test/; fix and repeat. The legacy --base/--embed form is a smoke test, not the full gate.

6. Record the Obsidian version/channel, platform, fixture ID, JSON summary, performance evidence, and relevant residual manual checks.

Pass contract

The bundled runner must prove all of the following before exiting zero:

• The CLI resolved the explicit test vault through its guarded fallbacks, native base:query totals matched every tracked scenario, and fixture paths were fully covered where page walks are requested.
• The packaged plugin survived disable/re-enable and repeated reloads without duplicate active views.
• Empty, one-row, exact-boundary, limited, grouped, malformed legacy, native-value, custom page-size, and 1,200-row views have bounded rows and coherent status text.
• Direct and embedded views, CDP mouse and keyboard transitions, invalid custom input, desktop/390 px mobile renders, and focus visibility passed.
• The accessibility tree exposes named navigation and table landmarks, buttons, combobox, and live status roles.
• Obsidian captured no errors or error-level console messages.
• Every vault file outside the Obsidian config directory retained the same content hash.

The runner reports timing and heap data as evidence, not a hard performance budget. It does not prove real NVDA, VoiceOver, or TalkBack speech; OS-level window focus; hover timing outside the renderer; visual quality in every theme; or physical iOS/Android behavior. Execute only the residual checks touched by the change.