feat(floating): live "working" status + a fading "done" nod when driving roro

docs/VERIFICATION.md

@@ -248,7 +248,7 @@ proves recall degrades to recency when embeddings are unavailable. `npm run veri
 narration containing that value.
 `npm run verify:packaged-natural-memory-turn` also keeps live Ollama enabled, teaches a stated preference through
 packaged `turnRun`, waits for the extracted profile fact, relaunches, asks about it through a later turn, and asserts
-the floating Ask shows NO success banner after recall (the receipt state is empty) while a memory-status beat proves the
+the floating Ask shows a brief `done` nod after recall (the receipt state reads `done`) while a memory-status beat proves the
 recall used memory. The natural-language packaged smoke uses the floating Ask smoke hook to inspect receipt state, but it
 does not launch the packaged app in floating-window mode or prove the failure receipt is visibly rendered; use
 `npm run verify:floating-live-turn` for the visible floating Ask path.
@@ -408,16 +408,17 @@ It is opt-in (needs a display + a vite build) and not in CI. Checks: `#floating-
 `collapsed`; pill reads "Ask Roro…"; `#floating-stop` exists and is not `armed`; clicking the pill →
 `expanded` with the input actually visible; the memory profile bridge responds before teardown; Escape
 → `collapsed`; smoke submit → `tasked` with trimmed pill copy and visible Stop before `run.started`;
-pre-run Stop calls the no-id cancel path and shows neutral `Stopped.` copy; universal `runEnd`
-collapses answer/clarify turns with no `run.started` and leaves NO receipt on success (the surface stays clean); `run.started` keeps Stop visibly armed and
+pre-run Stop calls the no-id cancel path and shows neutral `Stopped.` copy; a tasked turn shows a live
+`thinking…` progress line; universal `runEnd` collapses answer/clarify turns with no `run.started` and shows a
+brief `done` nod that fades; `run.started` keeps Stop visibly armed and
 targets the captured run id; real `run.failed` disarms Stop, shows actionable error copy, hides raw
 spawn text, and the error remains visible after `runEnd` collapse until the next summon clears it.
 
 `scripts/smoke-floating-live-turn.mjs` is the opt-in live counterpart. It launches the dev app in
 floating mode, leaves `RORO_FLOATING_SMOKE` and `RORO_DEBUG_BRIDGE` off, drives the visible Ask form,
 and verifies real `window.companion.turnRun` answer and executor turns over the public push stream.
 The answer turn must emit `ActionEvent`s, reach `runEnd`, avoid the coding executor, and collapse the
-floating Ask with NO success receipt (a completed turn shows no banner; memory use is proven by the status beat, not a receipt). In the
+floating Ask with a brief `done` nod (an answer turn changes no files; memory use is proven by the status beat). In the
 deterministic fake-Ollama path, the answer narration must include the requested phrase exactly enough
 to prove the public event stream carried the model answer. In the optional real-Ollama path, the
 narration check accepts a substantive non-placeholder answer because that mode validates the local
@@ -429,8 +430,8 @@ never complete/write the aborted file, collapse from `runEnd`, and show neutral
 following executor success turn must launch Codex with the expected
 `exec --json --skip-git-repo-check -s workspace-write -C <project>` shape, keep Stop armed for the
 accepted/running turn, emit a completed `file_change` and `run.completed`, write inside the selected
-test project (disposable by default), disarm Stop, collapse from `runEnd`, and show NO success receipt
-(the surface stays clean on success). By default it uses tiny
+test project (disposable by default), disarm Stop, collapse from `runEnd`, and show a brief `done · N files`
+nod that fades. By default it uses tiny
 local fake Ollama and Codex servers/binaries so the product-loop proof is deterministic; set
 `RORO_FLOATING_LIVE_USE_REAL_OLLAMA=1` when you specifically want to validate the local model path.
 Ignored-`SIGTERM` slot retention remains covered by the unit gate above, not this live smoke. Keep it out
@@ -446,10 +447,10 @@ Start the app (`ollama serve` first if you want a real turn): `npm start`.
 | 2 | Click the pill (or ⌘⇧Space) | Input expands and is focused; pill chrome hidden |
 | 3 | Press Enter on an **empty** input | Nothing happens — no thinking flash, stays expanded |
 | 4 | Type "add a logout route" + Enter | After readiness accepts the turn, the cat snaps to *thinking*, the pill shows `tasked: add a logout route`, and Stop appears before any executor event |
-| 5 | While a `run_agent` turn runs | `#floating-stop` remains visible with the `armed` class |
+| 5 | While a `run_agent` turn runs | `#floating-stop` stays visible with the `armed` class; the status line reads a calm `working…` the whole time (never blank) |
 | 6 | Click Stop | Run cancels; Stop shows `Stopping...`, then disarms/hides; Ask collapses and shows neutral `Stopped.` copy |
-| 7 | An answer/clarify turn (no executor run) | Ask still collapses when the turn ends (universal `runEnd`); no success banner is shown (the cat's animation conveys "done") |
-| 8 | A successful executor turn completes | Ask collapses and Stop disarms; no success banner is shown. Only a failure (sticky red) or a Stop (neutral `Stopped.`) leaves a receipt |
+| 7 | An answer/clarify turn (no executor run) | Ask collapses when the turn ends (universal `runEnd`) and shows a brief `done` nod that fades after a few seconds |
+| 8 | A successful executor turn completes | Ask collapses and Stop disarms; a brief `done · N files` nod appears, then fades. Only a failure (sticky red) lingers |
 | 9 | Press Esc while expanded | Collapses back to the "Ask Roro…" pill |
 
 A CSS regression shows up as: a state that doesn't visually change (e.g. input stays hidden when

scripts/smoke-floating-ask.mjs

@@ -208,11 +208,12 @@ try {
   await evalJs(`window.__roroFloatingAskSmoke.startTask('answer without an executor')`);
   await sleep(100);
   check('#floating-stop arms for accepted answer/clarify turn', await evalJs(`document.getElementById('floating-stop').classList.contains('armed')`));
+  check('#floating-error shows a live "thinking…" progress line while tasked', await evalJs(`document.getElementById('floating-error').textContent === 'thinking…' && document.getElementById('floating-error').classList.contains('progress') && !document.getElementById('floating-error').hidden`));
   await evalJs(`window.__roroFloatingAskSmoke.runEnd()`);
   await sleep(100);
   check('#floating-ask collapses on successful runEnd without run.started', await evalJs(`document.getElementById('floating-ask').classList.contains('collapsed')`));
   check('#floating-stop hides after successful runEnd without run.started', !(await isVisible('#floating-stop')));
-  check('#floating-error stays HIDDEN after successful runEnd (success shows no banner)', await evalJs(`document.getElementById('floating-error').hidden === true && !document.getElementById('floating-error').classList.contains('success')`));
+  check('#floating-error shows a brief "done" nod after successful runEnd', await evalJs(`document.getElementById('floating-error').textContent === 'done' && document.getElementById('floating-error').classList.contains('success') && !document.getElementById('floating-error').hidden`));
 
   console.log('[smoke] asserting executor Stop targets run id after run.started…');
   await evalJs(`document.getElementById('ask-pill').click()`);

scripts/smoke-floating-live-turn.mjs

@@ -1007,9 +1007,8 @@ try {
     const el = document.getElementById('floating-error');
     return { hidden: el?.hidden, text: el?.textContent ?? '', success: el?.classList.contains('success') ?? false };
   })()`, {}, 'answer receipt check');
-  check('floating receipt stays HIDDEN for successful answer turn (no success banner)', answerReceipt.hidden === true, JSON.stringify(answerReceipt));
-  check('floating receipt has no success tone for answer turn', answerReceipt.success === false, JSON.stringify(answerReceipt));
-  check('floating receipt shows no text for successful answer turn', answerReceipt.text === '', JSON.stringify(answerReceipt));
+  check('floating Ask shows a brief "done" nod for successful answer turn', answerReceipt.hidden === false && answerReceipt.success === true, JSON.stringify(answerReceipt));
+  check('floating answer nod reads "done" (an answer turn changes no files)', answerReceipt.text === 'done', JSON.stringify(answerReceipt));
 
   if (!USE_REAL_OLLAMA) {
     const stoppedTranscript = `${STOP_TRANSCRIPT}. Start a coding task that should be stopped before the executor starts.`;
@@ -1310,9 +1309,8 @@ try {
     const el = document.getElementById('floating-error');
     return { hidden: el?.hidden, text: el?.textContent ?? '', success: el?.classList.contains('success') ?? false };
   })()`, {}, 'executor receipt check');
-  check('floating receipt stays HIDDEN for successful executor turn (no success banner)', executorReceipt.hidden === true, JSON.stringify(executorReceipt));
-  check('floating receipt has no success tone for executor turn', executorReceipt.success === false, JSON.stringify(executorReceipt));
-  check('floating receipt shows no text for successful executor turn', executorReceipt.text === '', JSON.stringify(executorReceipt));
+  check('floating Ask shows a "done" nod for successful executor turn', executorReceipt.hidden === false && executorReceipt.success === true, JSON.stringify(executorReceipt));
+  check('floating executor nod reports the changed-file count', /^done · \d+ files?$/.test(executorReceipt.text), JSON.stringify(executorReceipt));
   const codexInvocations = await readFile(fakeCodexArgsFile, 'utf8')
     .then((text) => JSON.parse(text))
     .catch(() => []);

scripts/smoke-packaged-memory.mjs

@@ -1028,8 +1028,8 @@ try {
     check('natural-language recall turn completed with runEnd', natural?.ok === true && Boolean(natural.runEnd), natural?.message);
     check('natural-language recall runEnd matches turnRun result', natural?.runEnd?.runId === natural?.turnResult?.runId);
     check(
-      'natural-language floating Ask shows NO success banner after recall (memory use proven by the status beat below)',
-      natural?.receiptText === '',
+      'natural-language floating Ask shows a brief "done" nod after recall (memory use proven by the status beat below)',
+      natural?.receiptText === 'done',
       natural?.receiptText || JSON.stringify(natural?.floatingState ?? {}),
     );
     check('natural-language recall emitted a memory status beat', Boolean(memoryStatus), JSON.stringify(naturalEvents));

src/index.css

@@ -393,6 +393,13 @@ body.floating-window #floating-error.success {
   background: rgba(18, 78, 48, 0.9);
   border-color: rgba(126, 230, 164, 0.5);
 }
+/* Live "roro is working" status — calm slate, clearly neither success (green) nor failure (red). */
+body.floating-window #floating-error.progress {
+  color: #eef2fb;
+  background: rgba(28, 36, 56, 0.86);
+  border-color: rgba(150, 175, 230, 0.4);
+  opacity: 0.92;
+}
 body.floating-window #floating-stop.armed + #floating-error:not([hidden]) {
   bottom: 88px;
 }

src/renderer/ask/floatingAsk.test.ts

@@ -179,43 +179,40 @@ describe('floatingAsk shell (jsdom)', () => {
     expect(h.stop.classList.contains('armed')).toBe(false);
   });
 
-  it('shows NO receipt after a successful answer turn (the cat conveys "done", not a banner)', async () => {
+  it('shows "thinking…" while planning, then a brief "done" nod after a successful answer turn', async () => {
     h.pill.click();
     h.input.value = 'what did we decide?';
     submit(h.form);
     await flush();
+    expect(h.error.classList.contains('progress')).toBe(true); // live "I'm on it" while it works
+    expect(h.error.textContent).toBe('thinking…');
     h.fireAction(memoryUsed);
     h.fireRunEnd();
     expect(h.form.classList.contains('collapsed')).toBe(true);
-    // Success leaves the surface clean — no lingering "Done." banner over the user's screen.
-    expect(h.error.hidden).toBe(true);
-    expect(h.error.classList.contains('success')).toBe(false);
-    expect(h.error.textContent).toBe('');
+    // A nod (not a lingering banner) — an answer turn changes no files.
+    expect(h.error.hidden).toBe(false);
+    expect(h.error.classList.contains('success')).toBe(true);
+    expect(h.error.classList.contains('progress')).toBe(false);
+    expect(h.error.textContent).toBe('done');
   });
 
-  it('shows NO receipt after a successful executor turn (even with changed files + memory)', async () => {
+  it('shows a persistent "working…" line through a multi-event task (never a frozen gap)', async () => {
     h.pill.click();
-    h.input.value = 'edit it';
+    h.input.value = 'build it';
     submit(h.form);
     await flush();
+    expect(h.error.textContent).toBe('thinking…');
     h.fireAction(started);
+    expect(h.error.textContent).toBe('working…'); // run.started → sustained "working" feedback
+    // mid-task events keep the working line visible — the bug was it going blank for 30–90s.
+    h.fireAction({ kind: 'file_change', runId: 'r1', itemId: 'f1', status: 'completed', files: [{ path: 'a.ts', op: 'update' }], ts: 2 });
     h.fireAction(memoryUsed);
-    h.fireAction({
-      kind: 'file_change',
-      runId: 'r1',
-      itemId: 'file-1',
-      status: 'completed',
-      files: [{ path: 'src/app.ts', op: 'update' }],
-      ts: 2,
-    });
-    h.fireAction({ kind: 'run.completed', runId: 'r1', ok: true, finalText: 'done', ts: 3 });
-    h.fireRunEnd();
-    expect(h.error.hidden).toBe(true);
-    expect(h.error.classList.contains('success')).toBe(false);
-    expect(h.error.textContent).toBe('');
+    expect(h.error.hidden).toBe(false);
+    expect(h.error.classList.contains('progress')).toBe(true);
+    expect(h.error.textContent).toBe('working…');
   });
 
-  it('leaves no lingering receipt across two consecutive successful turns', async () => {
+  it('shows "done · 1 file" after a successful executor turn', async () => {
     h.pill.click();
     h.input.value = 'edit it';
     submit(h.form);
@@ -231,20 +228,31 @@ describe('floatingAsk shell (jsdom)', () => {
       ts: 2,
     });
     h.fireAction({ kind: 'run.completed', runId: 'r1', ok: true, finalText: 'done', ts: 3 });
-    h.fireRunEnd('r1');
-    expect(h.error.hidden).toBe(true); // first success: clean surface
-
-    h.pill.click();
-    h.input.value = 'what now?';
-    h.turnRun.mockResolvedValueOnce({ runId: 'r2' });
-    submit(h.form);
-    await flush();
-    h.fireRunEnd('r2');
+    h.fireRunEnd();
+    expect(h.error.hidden).toBe(false);
+    expect(h.error.classList.contains('success')).toBe(true);
+    expect(h.error.textContent).toBe('done · 1 file');
+  });
 
-    // Second success also leaves nothing — no banner from either turn lingers over the screen.
-    expect(h.error.hidden).toBe(true);
-    expect(h.error.classList.contains('success')).toBe(false);
-    expect(h.error.textContent).toBe('');
+  it('fades the "done" nod after a few seconds (a nod, not a lingering banner)', async () => {
+    vi.useFakeTimers();
+    try {
+      h.pill.click();
+      h.input.value = 'edit it';
+      submit(h.form);
+      await vi.advanceTimersByTimeAsync(5); // resolve the async submit (turnRun mock + flush)
+      h.fireAction(started);
+      h.fireAction({ kind: 'file_change', runId: 'r1', itemId: 'file-1', status: 'completed', files: [{ path: 'src/app.ts', op: 'update' }], ts: 2 });
+      h.fireAction({ kind: 'run.completed', runId: 'r1', ok: true, finalText: 'done', ts: 3 });
+      h.fireRunEnd('r1');
+      expect(h.error.textContent).toBe('done · 1 file'); // nod shown…
+      expect(h.error.hidden).toBe(false);
+      await vi.advanceTimersByTimeAsync(3500); // past DONE_NOD_MS
+      expect(h.error.hidden).toBe(true); // …then it fades on its own
+      expect(h.error.textContent).toBe('');
+    } finally {
+      vi.useRealTimers();
+    }
   });
 
   it('ignores unrelated floating events and runEnd signals once the accepted run id is known', async () => {
@@ -262,7 +270,9 @@ describe('floatingAsk shell (jsdom)', () => {
     h.fireRunEnd('r1');
 
     expect(h.form.classList.contains('collapsed')).toBe(true);
-    expect(h.error.hidden).toBe(true); // success → no banner; the unrelated 'other-run' status never surfaced
+    // Success nod for the real run; the unrelated 'other-run' run never drove the surface.
+    expect(h.error.classList.contains('success')).toBe(true);
+    expect(h.error.textContent).toBe('done'); // no file_change for r1 → plain 'done'
   });
 
   it('keeps actionable failure copy visible after runEnd collapses the Ask', async () => {