feat(harness): agent-drivable dev harness and CI that gates

A coding agent could develop this repo's Go packages and could not
develop the application: every path to running YellowJacket ended in a
blocking GTK window, so 265 bound methods, 46 events, 33 component
directories and 13 stores had exactly one form of verification
available — `tsc --noEmit`.

The unlock is that `wails dev`'s dev server on :34115 serves the real
frontend with the real generated bindings against the same Go backend a
desktop window attaches to, so a plain Chromium under Xvfb gets a fully
functional app. Four test tiers now exist, cheapest first:

- `make ui-test` — 313 Vitest tests in a real browser in ~2 s, no app,
  no backend, no display. Works because `frontend/wailsjs/` is a pure
  passthrough to `window.go`/`window.runtime`, so faking just those two
  globals runs the real bindings and the real store code.
- `make test` — services in-process, asserting on the payload the
  frontend would receive, via a new `events.Emit` wrapper.
- `make dev-headless` + `playwright-cli` — the real app, driven
  interactively, with an event bridge on `window.__yjEvents` and a
  dev-only control surface at `/__test/`.
- `make e2e` — 19 of those flows frozen as Playwright specs.

`events.Emit(ctx, …)` replaces all 35 direct `runtime.EventsEmit` call
sites: wails' `getEvents` `log.Fatalf`s on any context without its
runtime, so those paths could not run under test and a background
worker could take the app down. Four packages had each hand-rolled the
same guard; nine more guarded on `ctx != nil`, which does not help.
`TestNoDirectRuntimeEmits` fails the build on a new one.

Fixtures are generated, not committed (`make testdata`), and seeds are
built by *running the app* — never by hand-writing config and DB rows,
which would be a second description of a valid YJ_HOME.

`.gitea/workflows/ci.yml` is the first workflow here that tests
anything; the other three only package, so `gitea_ci` reported only
packaging jobs and misled anyone asking whether a push was healthy.
Both jobs were prototyped to green in a bare ubuntu:24.04 container
before the YAML was written, which immediately caught `make lint`
linting three configurations that nothing builds: all three passes
omitted `webkit2_41`, so wails resolved webkit2gtk-4.0 — which Arch
still ships and Ubuntu 24.04 dropped.

Operational instructions live in `.pi/skills/yellowjacket-dev/`,
measured discoveries in `.planning/NOTES.md`, and architecture in
`CLAUDE.md` — split by tense, not by topic, because a topical split
gives every new fact two plausible homes. `make skill-check` fails a
commit if the skill cites a make target that does not exist.

e2e/specs/harness.spec.ts

@@ -0,0 +1,89 @@
+import {
+  test,
+  expect,
+  callBinding,
+  resetEvents,
+  waitForEvent,
+} from '../support/fixtures.js';
+
+/**
+ * The harness testing itself.
+ *
+ * If these fail, every other spec's failure is uninterpretable — a
+ * missing event could mean a broken feature or a broken recorder, and
+ * telling those apart afterwards is expensive.
+ */
+test.describe('harness', () => {
+  test('the app is the real app, not a mock', async ({ app }) => {
+    // All 11 bound services land on window.go through the dev server.
+    const services = await app.evaluate(() => Object.keys(window.go));
+
+    expect(services).toEqual(
+      expect.arrayContaining(['queue', 'player', 'library', 'explore']),
+    );
+
+    const state = await callBinding<{ tracks: unknown[] }>(
+      app,
+      'queue.Queue.GetState',
+    );
+
+    expect(state).toHaveProperty('tracks');
+  });
+
+  test('backend events are recorded, in order, with payloads', async ({
+    app,
+  }) => {
+    await resetEvents(app);
+    await callBinding(app, 'player.Player.SetVolume', [37]);
+
+    const ev = await waitForEvent(app, 'VolumeChanged');
+
+    expect(ev.data).toEqual([37]);
+    expect(ev.dir).toBe('in');
+  });
+
+  test('exactly one recorder is installed', async ({ app }) => {
+    // Listeners registered by one evaluate survive into the next, so a
+    // recorder that re-registers counts every event twice.  This is the
+    // regression test for that.
+    await resetEvents(app);
+    await callBinding(app, 'player.Player.SetVolume', [41]);
+    await waitForEvent(app, 'VolumeChanged');
+
+    const count = await app.evaluate(() =>
+      window.__yjEvents.count('VolumeChanged'),
+    );
+
+    expect(count).toBe(1);
+  });
+
+  test('a binding called with wrong types fails fast', async ({ app }) => {
+    // player.UserVolume is an int.  Passing a float makes the backend
+    // log "error parsing arguments" and never fire the callback; without
+    // a timeout the promise never settles and the spec hangs until the
+    // suite gives up.
+    const failure = await app.evaluate(async () => {
+      try {
+        await window.__yjEvents.call(
+          'player.Player.SetVolume',
+          [0.42],
+          2_000,
+        );
+
+        return 'settled';
+      } catch (err) {
+        return (err as Error).message;
+      }
+    });
+
+    expect(failure).toContain('did not settle');
+  });
+
+  test('the control surface is mounted and seeded', async ({ testctl }) => {
+    const health = await testctl.health();
+
+    expect(health.ok).toBe(true);
+    expect(health.libraries.length).toBeGreaterThan(0);
+    expect(health.counts.tracks).toBeGreaterThan(0);
+  });
+});

e2e/specs/library.spec.ts

@@ -0,0 +1,68 @@
+import { test, expect } from '../support/fixtures.js';
+
+/**
+ * The library views, against the generated fixture library
+ * (`make testdata`): 31 tracks chosen to cover the cases the app has
+ * code for — unicode and RTL titles, missing tags, a deliberately
+ * absurd artist name for truncation, duplicates.
+ */
+test.describe('library views', () => {
+  test('lands in the app, not the first-run wizard', async ({ app }) => {
+    // A fresh YJ_HOME puts <first-run-wizard> over everything and it
+    // intercepts every pointer event, so "the click did nothing" is the
+    // symptom of an unseeded sandbox rather than a broken control.
+    //
+    // Asserted by clicking rather than by inspecting the wizard element:
+    // the element is always in the DOM and merely renders nothing once a
+    // library exists, so its presence proves nothing.  Playwright's own
+    // actionability check fails a covered click with "intercepts pointer
+    // events", which is exactly the condition worth catching.
+    await expect(app.getByTestId('track-row').first()).toBeVisible();
+    await app.getByTestId('nav-artists').click({ timeout: 5_000 });
+
+    await expect(app.getByTestId('main-content')).toHaveAttribute(
+      'data-active-view',
+      'artists',
+    );
+  });
+
+  test('renders every fixture track, unicode included', async ({
+    app,
+    testctl,
+  }) => {
+    const health = await testctl.health();
+    const rows = app.getByTestId('track-row');
+
+    await expect(rows).toHaveCount(health.counts.tracks);
+
+    // Non-Latin scripts survive the tag reader, the database and the
+    // renderer.  These titles exist in the fixtures for this reason.
+    await expect(app.getByText('Привет мир')).toBeVisible();
+    await expect(app.getByText('さくら')).toBeVisible();
+    await expect(app.getByText('مرحبا بالعالم')).toBeVisible();
+  });
+
+  test('the sidebar navigates between primary views', async ({ app }) => {
+    const main = app.getByTestId('main-content');
+
+    for (const view of ['artists', 'genres', 'albums', 'playlists', 'tracks']) {
+      await app.getByTestId(`nav-${view}`).click();
+      await expect(main).toHaveAttribute('data-active-view', view);
+      await expect(app.getByTestId(`nav-${view}`)).toHaveAttribute(
+        'aria-current',
+        'page',
+      );
+    }
+  });
+
+  test('the artists view shows the fixture artists', async ({ app }) => {
+    await app.getByTestId('nav-artists').click();
+    await expect(app.getByTestId('main-content')).toHaveAttribute(
+      'data-active-view',
+      'artists',
+    );
+
+    await expect(app.getByText('Aurora Fields').first()).toBeVisible();
+    await expect(app.getByText('Pale Circuit').first()).toBeVisible();
+  });
+});

e2e/specs/playback.spec.ts

@@ -0,0 +1,114 @@
+import {
+  test,
+  expect,
+  callBinding,
+  resetEvents,
+  waitForEvent,
+  LONG_TRACK,
+} from '../support/fixtures.js';
+
+/** The one fixture long enough to still be playing on the next line. */
+const longRow = (app: import('@playwright/test').Page) =>
+  app.getByTestId('track-row').filter({ hasText: LONG_TRACK }).first();
+
+/**
+ * Playback and the queue, driven through the UI and asserted on the
+ * events the backend actually emits.
+ *
+ * Audio really is initialised here: under `dbus-run-session` + Xvfb the
+ * PulseAudio socket in /run/user is untouched, so InitSpeaker succeeds
+ * and these tracks genuinely play.  A CI container without /run/user
+ * needs a null sink; everything except the audio itself still works
+ * without one.
+ */
+test.describe('playback', () => {
+  test.beforeEach(async ({ app }) => {
+    await callBinding(app, 'queue.Queue.Clear').catch(() => {
+      /* older builds may not expose Clear; the specs below do not need it */
+    });
+    await resetEvents(app);
+  });
+
+  test('double-clicking a track plays it', async ({ app }) => {
+    await longRow(app).dblclick();
+
+    const changed = await waitForEvent(app, 'TrackChanged');
+
+    expect(changed.data[0]).toBeTruthy();
+
+    // The transport flips to Pause, which is the only place the UI
+    // states "we are playing" in a way a user can see.  `exact` is not
+    // optional: "Add queue to playlist" also matches /play/i.
+    await expect(
+      app.getByRole('button', { name: 'Pause', exact: true }),
+    ).toBeVisible();
+
+    await expect(app.getByTestId('now-playing-title')).toContainText(
+      LONG_TRACK,
+    );
+  });
+
+  test('the elapsed time advances', async ({ app }) => {
+    await longRow(app).dblclick();
+    await waitForEvent(app, 'TrackChanged');
+
+    // Not a fixed sleep on a fixed value: assert the observable
+    // outcome, which is that the clock is no longer at zero.
+    await expect(app.getByTestId('elapsed-time')).not.toHaveText('--:--');
+    await expect(app.getByTestId('elapsed-time')).not.toHaveText('00:00', {
+      timeout: 15_000,
+    });
+  });
+
+  test('pause and play round-trip through the backend', async ({ app }) => {
+    await longRow(app).dblclick();
+    await waitForEvent(app, 'TrackChanged');
+
+    await resetEvents(app);
+    await app.getByRole('button', { name: 'Pause', exact: true }).click();
+    await waitForEvent(app, 'PlaybackStateChanged');
+
+    await expect(
+      app.getByRole('button', { name: 'Play', exact: true }),
+    ).toBeVisible();
+  });
+
+  test('volume changes are pushed back from Go', async ({ app }) => {
+    await resetEvents(app);
+    await callBinding(app, 'player.Player.SetVolume', [55]);
+
+    const ev = await waitForEvent(app, 'VolumeChanged');
+
+    expect(ev.data).toEqual([55]);
+  });
+});
+
+test.describe('queue', () => {
+  test('playing a track populates the queue panel', async ({ app }) => {
+    await resetEvents(app);
+    await longRow(app).dblclick();
+    await waitForEvent(app, 'QueueChanged');
+
+    await expect(app.getByTestId('queue-row')).toHaveCount(1);
+
+    const state = await callBinding<{ tracks: unknown[] }>(
+      app,
+      'queue.Queue.GetState',
+    );
+
+    expect(state.tracks).toHaveLength(1);
+  });
+
+  test('shuffle and repeat toggles report their state', async ({ app }) => {
+    const shuffle = app.getByRole('button', { name: 'Shuffle' });
+
+    await resetEvents(app);
+    await shuffle.click();
+    await waitForEvent(app, 'QueueModeChanged');
+
+    await expect(shuffle).toHaveAttribute('aria-pressed', 'true');
+
+    await shuffle.click();
+    await expect(shuffle).toHaveAttribute('aria-pressed', 'false');
+  });
+});

e2e/specs/testctl.spec.ts

@@ -0,0 +1,68 @@
+import { test, expect, resetEvents, waitForEvent } from '../support/fixtures.js';
+
+/**
+ * The dev-only control surface (backend/testctl), which exists for the
+ * things a browser genuinely cannot do.
+ */
+test.describe('control surface', () => {
+  test('database snapshot and restore round-trip', async ({ testctl }) => {
+    // VACUUM INTO copies the whole file and the restore copies every
+    // row back; on a database carrying an explore catalog that is tens
+    // of seconds, not the default 30s budget for a whole test.
+    test.setTimeout(180_000);
+
+    const before = (await testctl.health()).counts.tracks;
+
+    await testctl.snapshot('e2e-pristine');
+    await testctl.sql('DELETE FROM audio_files');
+
+    expect((await testctl.health()).counts.tracks).toBe(0);
+
+    // Restore copies rows rather than files, because the app holds the
+    // database open across two connection pools and cannot be made to
+    // reopen it from here.
+    await testctl.restore('e2e-pristine');
+
+    expect((await testctl.health()).counts.tracks).toBe(before);
+  });
+
+  test('a forced backend event reaches the browser', async ({
+    app,
+    testctl,
+  }) => {
+    // LibraryScanProgress normally only arrives during a real scan.
+    // Emitting it directly is how a push-driven view gets exercised
+    // without staging the work that would produce it.
+    await resetEvents(app);
+    await testctl.emit('LibraryScanProgress', {
+      current: 7,
+      total: 31,
+      currentFile: 'probe.mp3',
+    });
+
+    const ev = await waitForEvent(app, 'LibraryScanProgress');
+
+    expect(ev.data[0]).toMatchObject({ current: 7, total: 31 });
+  });
+
+  test('sql reads return rows, writes return a count', async ({ testctl }) => {
+    const read = await testctl.sql(
+      'SELECT COUNT(*) AS n FROM audio_files',
+    );
+
+    expect(read.rows[0].n).toBeGreaterThan(0);
+
+    const write = await testctl.sql(
+      'UPDATE player_state SET volume = volume',
+    );
+
+    expect(write).toHaveProperty('rowsAffected');
+  });
+
+  test('bad input is rejected with a reason, not a bare status', async ({
+    testctl,
+  }) => {
+    await expect(testctl.snapshot('../escape')).rejects.toThrow(/name must/);
+    await expect(testctl.restore('nope')).rejects.toThrow(/no such snapshot/);
+  });
+});