Split humanoid AvatarState into idleActive/idlePassive by default

Ported animation systems (from an Unreal Engine avatar app) key idle
behavior directly off this distinction, so it moves from an unbuilt
extension point to a core state. The pure state.ts reducer stays
timer-free (idleActive -> idlePassive rides in on an explicit
idle_timeout signal); deriveAvatarState in the client wrapper owns the
actual clock via a configurable idlePassiveTimeoutMs (default 30s).

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docs/HUMANOID.md

@@ -7,18 +7,18 @@ This package ships the **contract** future animation libraries (eye look-at, bli
 ## `$lib/humanoid/state.ts` — avatar state
 
 ```ts
-type AvatarState = 'idle' | 'listening' | 'thinking' | 'talking';
+type AvatarState = 'idleActive' | 'idlePassive' | 'listening' | 'thinking' | 'talking';
 ```
 
-Deliberately just four states, not the `IdleActive`/`IdlePassive` split a humanoid rig's own animation design (e.g. a future eye look-at library) may eventually want — that distinction needs a user-presence signal (webcam/face-tracking) this package has no opinion on. A project with that signal can layer its own finer idle split on top; `idle` here just means "no signal that the user or the model is currently doing anything."
+`idle` is split into `idleActive`/`idlePassive` from the start, since ported animation systems (this one's lineage traces to an Unreal Engine avatar app) commonly key their idle behavior off exactly this distinction: `idleActive` right after a turn ends (still "engaged"), settling into `idlePassive` once nothing has happened for a while (bored/ambient idle animation). The reducer itself has no timers — pure and isomorphic — so the active→passive transition rides in on an explicit `idle_timeout` signal; the client wrapper (`deriveAvatarState`, below) owns the actual clock, same as it owns `setListening`.
 
-`nextAvatarState(current, signal)` is a pure reducer — no WS/browser dependency, directly unit-tested in `state.spec.ts`. The signal set is small and closed (`listening_started`/`listening_stopped`, `response_requested`, `talking_started`/`talking_stopped`, `reset`).
+`nextAvatarState(current, signal)` is a pure reducer — no WS/browser dependency, directly unit-tested in `state.spec.ts`. The signal set is small and closed (`listening_started`/`listening_stopped`, `response_requested`, `talking_started`/`talking_stopped`, `idle_timeout`, `reset`).
 
-Note `listening_stopped` moves to `thinking`, not `idle` — ending listening almost always means a turn just completed and a response is expected next, so the avatar shouldn't visibly relax for the split second before the model's response starts arriving.
+Note `listening_stopped` moves to `thinking`, not back to idle — ending listening almost always means a turn just completed and a response is expected next, so the avatar shouldn't visibly relax for the split second before the model's response starts arriving. `talking_stopped` and `reset` both land on `idleActive`, never straight to `idlePassive` — passive idle is only ever reached via `idle_timeout`.
 
 ## `$lib/client/humanoid/avatar-state.ts` — deriving it live
 
-`deriveAvatarState(source)` takes a `RealtimeClient` or `ResilientRealtimeClient` (accepted structurally, via an `AvatarStateSource` interface — this package doesn't care which transport variant is in use) and folds its `ServerEvent`s through `nextAvatarState`:
+`deriveAvatarState(source, options?)` takes a `RealtimeClient` or `ResilientRealtimeClient` (accepted structurally, via an `AvatarStateSource` interface — this package doesn't care which transport variant is in use) and folds its `ServerEvent`s through `nextAvatarState`:
 
 | `ServerEvent`                         | Signal            |
 | ------------------------------------- | ----------------- |
@@ -33,12 +33,14 @@ Two signals can't come from the event stream alone, so the returned store expose
 - **`setListening(active)`** — call from whatever mic-capture code a project uses. That pipeline doesn't exist in `core` yet (see "Known limitations" below), and the client is the only thing that currently knows when the user starts/stops talking: the protocol doesn't surface OpenAI's `input_audio_buffer.speech_*` VAD events — see the "not part of our normalized vocabulary" comment in `server/core/providers/openai.ts`'s `fromOpenAiEvent`.
 - **`notifyResponseRequested()`** — call right when sending a typed `input_text.send` (the voice path already gets this for free from `setListening(false)`).
 
+This wrapper also owns the `idleActive` → `idlePassive` clock: the store starts (and every `reset`/`talking_stopped` returns to) `idleActive`, and a timer — `options.idlePassiveTimeoutMs`, default `DEFAULT_IDLE_PASSIVE_TIMEOUT_MS` (30s) — fires `idle_timeout` if nothing else happens first. Any other signal re-arms it on the next `idleActive` entry; `destroy()` cancels it.
+
 The returned `AvatarStateStore.subscribe()` follows the Svelte store contract (calls `run` immediately with the current value, then on every change), so it's usable directly as `$store` in a `.svelte` file without a hard Svelte dependency in this package itself. See `/demo/humanoid` (`src/routes/demo/humanoid/+page.svelte`) for the reference wiring, including the manual "listening" toggle standing in for real mic-capture UI.
 
 ### Known limitations
 
 - **`thinking` is a heuristic**, not a precise reflection of server-side activity — the time between a request and the assistant's first output. The browser never sees server-executed tool calls at all (see [TOOLS.md](TOOLS.md)), so "the model is running a tool" isn't observable from here.
-- **No presence-based idle split** (see above) — a project with webcam/face-tracking data can extend on top.
+- **The idle split is timer-based, not presence-based** — `idlePassive` is reached purely by elapsed time with no other signal, not by knowing whether the user actually left. A project with webcam/face-tracking data can layer a real presence signal on top (e.g. calling back into `idleActive` early, or suppressing the timeout) — that stays project-owned.
 
 ## `$lib/humanoid/animation.ts` — animation-channel composition
 

src/lib/client/humanoid/avatar-state.spec.ts

@@ -1,4 +1,4 @@
-import { describe, expect, it, vi } from 'vitest';
+import { afterEach, beforeEach, describe, expect, it, vi } from 'vitest';
 import type { ServerEvent } from '../../core/protocol.js';
 import { deriveAvatarState, type AvatarStateSource } from './avatar-state.js';
 
@@ -19,15 +19,23 @@ class FakeSource implements AvatarStateSource {
 }
 
 describe('deriveAvatarState', () => {
+	beforeEach(() => {
+		vi.useFakeTimers();
+	});
+
+	afterEach(() => {
+		vi.useRealTimers();
+	});
+
 	it('subscribe() emits the current state immediately', () => {
 		const store = deriveAvatarState(new FakeSource());
 		const run = vi.fn();
 		store.subscribe(run);
 		expect(run).toHaveBeenCalledTimes(1);
-		expect(run).toHaveBeenCalledWith('idle');
+		expect(run).toHaveBeenCalledWith('idleActive');
 	});
 
-	it('walks listening -> thinking -> talking -> idle across a full voice turn', () => {
+	it('walks listening -> thinking -> talking -> idleActive across a full voice turn', () => {
 		const source = new FakeSource();
 		const store = deriveAvatarState(source);
 		const seen: string[] = [];
@@ -38,7 +46,7 @@ describe('deriveAvatarState', () => {
 		source.emit({ type: 'output_audio.chunk', audio: '' });
 		source.emit({ type: 'output_audio.done' });
 
-		expect(seen).toEqual(['idle', 'listening', 'thinking', 'talking', 'idle']);
+		expect(seen).toEqual(['idleActive', 'listening', 'thinking', 'talking', 'idleActive']);
 	});
 
 	it('notifyResponseRequested() drives thinking for the typed-text path', () => {
@@ -51,7 +59,7 @@ describe('deriveAvatarState', () => {
 		source.emit({ type: 'transcript.delta', role: 'assistant', text: 'hi' });
 		source.emit({ type: 'transcript.done', role: 'assistant' });
 
-		expect(seen).toEqual(['idle', 'thinking', 'talking', 'idle']);
+		expect(seen).toEqual(['idleActive', 'thinking', 'talking', 'idleActive']);
 	});
 
 	it('ignores user transcript events (no signal either way)', () => {
@@ -62,10 +70,10 @@ describe('deriveAvatarState', () => {
 
 		source.emit({ type: 'transcript.delta', role: 'user', text: 'hello' });
 
-		expect(seen).toEqual(['idle']);
+		expect(seen).toEqual(['idleActive']);
 	});
 
-	it('session.ready and session.closed reset to idle from any state', () => {
+	it('session.ready and session.closed reset to idleActive from any state', () => {
 		const source = new FakeSource();
 		const store = deriveAvatarState(source);
 		const seen: string[] = [];
@@ -74,7 +82,7 @@ describe('deriveAvatarState', () => {
 		source.emit({ type: 'output_audio.chunk', audio: '' }); // -> talking
 		source.emit({ type: 'session.closed', reason: 'client_requested' });
 
-		expect(seen).toEqual(['idle', 'talking', 'idle']);
+		expect(seen).toEqual(['idleActive', 'talking', 'idleActive']);
 	});
 
 	it('response.done is a fallback that stops talking even without output_audio.done', () => {
@@ -86,7 +94,7 @@ describe('deriveAvatarState', () => {
 		source.emit({ type: 'output_audio.chunk', audio: '' });
 		source.emit({ type: 'response.done' });
 
-		expect(seen).toEqual(['idle', 'talking', 'idle']);
+		expect(seen).toEqual(['idleActive', 'talking', 'idleActive']);
 	});
 
 	it('destroy() stops reacting to further events', () => {
@@ -98,6 +106,46 @@ describe('deriveAvatarState', () => {
 		store.destroy();
 		source.emit({ type: 'output_audio.chunk', audio: '' });
 
-		expect(seen).toEqual(['idle']);
+		expect(seen).toEqual(['idleActive']);
+	});
+
+	it('settles idleActive into idlePassive after the configured timeout with no other signal', () => {
+		const source = new FakeSource();
+		const store = deriveAvatarState(source, { idlePassiveTimeoutMs: 5_000 });
+		const seen: string[] = [];
+		store.subscribe((s) => seen.push(s));
+
+		vi.advanceTimersByTime(5_000);
+
+		expect(seen).toEqual(['idleActive', 'idlePassive']);
+	});
+
+	it('does not fire idle_timeout while something else is happening', () => {
+		const source = new FakeSource();
+		const store = deriveAvatarState(source, { idlePassiveTimeoutMs: 5_000 });
+		const seen: string[] = [];
+		store.subscribe((s) => seen.push(s));
+
+		vi.advanceTimersByTime(4_000);
+		store.setListening(true); // -> listening, cancels/re-arms the idle timer
+		vi.advanceTimersByTime(4_000); // < 5s since re-arm — still listening
+		store.setListening(false); // -> thinking
+
+		expect(seen).toEqual(['idleActive', 'listening', 'thinking']);
+	});
+
+	it('re-arms the idle timer once back in idleActive, and destroy() cancels it', () => {
+		const source = new FakeSource();
+		const store = deriveAvatarState(source, { idlePassiveTimeoutMs: 5_000 });
+		const seen: string[] = [];
+		store.subscribe((s) => seen.push(s));
+
+		source.emit({ type: 'output_audio.chunk', audio: '' }); // -> talking
+		source.emit({ type: 'output_audio.done' }); // -> idleActive, re-arms timer
+
+		store.destroy();
+		vi.advanceTimersByTime(5_000);
+
+		expect(seen).toEqual(['idleActive', 'talking', 'idleActive']);
 	});
 });

src/lib/client/humanoid/avatar-state.ts

@@ -29,10 +29,24 @@ export interface AvatarStateStore {
 	 * gets this for free from `setListening(false)`).
 	 */
 	notifyResponseRequested(): void;
-	/** Stops listening to the source and releases all subscribers. */
+	/** Stops listening to the source, cancels the idle timer, and releases all subscribers. */
 	destroy(): void;
 }
 
+/** Default `idlePassiveTimeoutMs` — see `DeriveAvatarStateOptions`. */
+export const DEFAULT_IDLE_PASSIVE_TIMEOUT_MS = 30_000;
+
+export interface DeriveAvatarStateOptions {
+	/**
+	 * How long the avatar sits in `idleActive` with no other signal before
+	 * this wrapper fires the `idle_timeout` signal that settles it into
+	 * `idlePassive`. The pure reducer in `$lib/humanoid/state.ts` has no
+	 * timers of its own, so this wrapper is what actually owns the clock —
+	 * tune to whatever cadence a project's idle animation set expects.
+	 */
+	idlePassiveTimeoutMs?: number;
+}
+
 /**
  * Derives a reactive `AvatarState` from a live `RealtimeClient` or
  * `ResilientRealtimeClient`. Translates `ServerEvent`s (plus the imperative
@@ -44,18 +58,46 @@ export interface AvatarStateStore {
  * assistant's first output — not a precise reflection of server-side
  * activity: the browser never sees server-executed tool calls at all (see
  * docs/TOOLS.md), so "the model is running a tool" isn't observable here.
+ *
+ * Also owns the `idleActive` -> `idlePassive` clock: the store starts (and
+ * every `reset`/`talking_stopped` returns to) `idleActive`, and this
+ * function arms a timer — `options.idlePassiveTimeoutMs`, default
+ * `DEFAULT_IDLE_PASSIVE_TIMEOUT_MS` — that fires `idle_timeout` if nothing
+ * else happens first.
  */
-export function deriveAvatarState(source: AvatarStateSource): AvatarStateStore {
+export function deriveAvatarState(
-	let state: AvatarState = 'idle';
+	source: AvatarStateSource,
+	options: DeriveAvatarStateOptions = {}
+): AvatarStateStore {
+	const idlePassiveTimeoutMs = options.idlePassiveTimeoutMs ?? DEFAULT_IDLE_PASSIVE_TIMEOUT_MS;
+	let state: AvatarState = 'idleActive';
 	const subscribers = new Set<(state: AvatarState) => void>();
+	let idleTimer: ReturnType<typeof setTimeout> | null = null;
+
+	// Re-armed after every transition: scheduled only while `idleActive` (the
+	// one state `idle_timeout` acts on), cleared everywhere else so a stale
+	// timer from a prior idleActive spell can't fire `idle_timeout` while
+	// e.g. the avatar is now talking.
+	function syncIdleTimer(): void {
+		if (idleTimer !== null) {
+			clearTimeout(idleTimer);
+			idleTimer = null;
+		}
+		if (state === 'idleActive') {
+			idleTimer = setTimeout(() => apply({ type: 'idle_timeout' }), idlePassiveTimeoutMs);
+		}
+	}
 
 	function apply(signal: AvatarSignal): void {
 		const next = nextAvatarState(state, signal);
 		if (next === state) return;
 		state = next;
+		syncIdleTimer();
 		for (const run of subscribers) run(state);
 	}
 
+	syncIdleTimer();
+
 	const stopEvents = source.on((event) => {
 		switch (event.type) {
 			case 'session.ready':
@@ -93,6 +135,7 @@ export function deriveAvatarState(source: AvatarStateSource): AvatarStateStore {
 		destroy() {
 			stopEvents();
 			subscribers.clear();
+			if (idleTimer !== null) clearTimeout(idleTimer);
 		}
 	};
 }

src/lib/humanoid/state.spec.ts

@@ -3,17 +3,19 @@ import { nextAvatarState } from './state.js';
 
 describe('nextAvatarState', () => {
 	it('listening_started always moves to listening', () => {
-		expect(nextAvatarState('idle', { type: 'listening_started' })).toBe('listening');
+		expect(nextAvatarState('idleActive', { type: 'listening_started' })).toBe('listening');
+		expect(nextAvatarState('idlePassive', { type: 'listening_started' })).toBe('listening');
 		expect(nextAvatarState('talking', { type: 'listening_started' })).toBe('listening');
 	});
 
 	it('listening_stopped moves to thinking (a response is expected next), only from listening', () => {
 		expect(nextAvatarState('listening', { type: 'listening_stopped' })).toBe('thinking');
-		expect(nextAvatarState('idle', { type: 'listening_stopped' })).toBe('idle');
+		expect(nextAvatarState('idleActive', { type: 'listening_stopped' })).toBe('idleActive');
 	});
 
 	it('response_requested moves to thinking unless already talking', () => {
-		expect(nextAvatarState('idle', { type: 'response_requested' })).toBe('thinking');
+		expect(nextAvatarState('idleActive', { type: 'response_requested' })).toBe('thinking');
+		expect(nextAvatarState('idlePassive', { type: 'response_requested' })).toBe('thinking');
 		expect(nextAvatarState('listening', { type: 'response_requested' })).toBe('thinking');
 		expect(nextAvatarState('talking', { type: 'response_requested' })).toBe('talking');
 	});
@@ -22,13 +24,23 @@ describe('nextAvatarState', () => {
 		expect(nextAvatarState('thinking', { type: 'talking_started' })).toBe('talking');
 	});
 
-	it('talking_stopped only returns to idle from talking', () => {
+	it('talking_stopped only returns to idleActive from talking', () => {
-		expect(nextAvatarState('talking', { type: 'talking_stopped' })).toBe('idle');
+		expect(nextAvatarState('talking', { type: 'talking_stopped' })).toBe('idleActive');
-		expect(nextAvatarState('idle', { type: 'talking_stopped' })).toBe('idle');
+		expect(nextAvatarState('idleActive', { type: 'talking_stopped' })).toBe('idleActive');
+		expect(nextAvatarState('idlePassive', { type: 'talking_stopped' })).toBe('idlePassive');
 	});
 
-	it('reset always returns to idle', () => {
+	it('idle_timeout moves idleActive to idlePassive, and is a no-op everywhere else', () => {
-		expect(nextAvatarState('talking', { type: 'reset' })).toBe('idle');
+		expect(nextAvatarState('idleActive', { type: 'idle_timeout' })).toBe('idlePassive');
-		expect(nextAvatarState('listening', { type: 'reset' })).toBe('idle');
+		expect(nextAvatarState('idlePassive', { type: 'idle_timeout' })).toBe('idlePassive');
+		expect(nextAvatarState('listening', { type: 'idle_timeout' })).toBe('listening');
+		expect(nextAvatarState('thinking', { type: 'idle_timeout' })).toBe('thinking');
+		expect(nextAvatarState('talking', { type: 'idle_timeout' })).toBe('talking');
+	});
+
+	it('reset always returns to idleActive', () => {
+		expect(nextAvatarState('talking', { type: 'reset' })).toBe('idleActive');
+		expect(nextAvatarState('listening', { type: 'reset' })).toBe('idleActive');
+		expect(nextAvatarState('idlePassive', { type: 'reset' })).toBe('idleActive');
 	});
 });

src/lib/humanoid/state.ts

@@ -6,14 +6,18 @@
  * look-at, blink, lip-sync, body/gesture) can all key off the same signal
  * without each re-deriving turn-taking themselves.
  *
- * Deliberately just four states, not the `IdleActive`/`IdlePassive` split a
+ * `idle` is split into `idleActive`/`idlePassive` from the start — ported
- * humanoid rig's own animation design may eventually want — that
+ * animation systems (this one's lineage traces to an Unreal Engine avatar
- * distinction needs a user-presence signal (webcam/face-tracking) this
+ * app) commonly key their idle behavior off exactly this distinction:
- * package has no opinion on. A project with that signal can layer its own
+ * `idleActive` right after a turn ends (still "engaged"), `idlePassive`
- * finer-grained idle split on top; `idle` here just means "no signal that
+ * once nothing has happened for a while (settle into bored/ambient idle
- * the user or the model is currently doing anything."
+ * animation). This reducer has no timers of its own — pure and isomorphic —
+ * so the active→passive transition rides in on an explicit `idle_timeout`
+ * signal rather than an internal side effect; the client wrapper
+ * (`$lib/client/humanoid/avatar-state.ts`) owns the actual timer, same as
+ * it owns `setListening`.
  */
-export type AvatarState = 'idle' | 'listening' | 'thinking' | 'talking';
+export type AvatarState = 'idleActive' | 'idlePassive' | 'listening' | 'thinking' | 'talking';
 
 export type AvatarSignal =
 	| { type: 'listening_started' }
@@ -21,6 +25,7 @@ export type AvatarSignal =
 	| { type: 'response_requested' }
 	| { type: 'talking_started' }
 	| { type: 'talking_stopped' }
+	| { type: 'idle_timeout' }
 	| { type: 'reset' };
 
 /**
@@ -45,9 +50,14 @@ export function nextAvatarState(current: AvatarState, signal: AvatarSignal): Ava
 		case 'talking_started':
 			return 'talking';
 		case 'talking_stopped':
-			return current === 'talking' ? 'idle' : current;
+			return current === 'talking' ? 'idleActive' : current;
+		// Only fires the active->passive transition; a client that emits this
+		// while something else is happening (listening/thinking/talking) is
+		// racing its own timer against real activity, so it's a no-op there.
+		case 'idle_timeout':
+			return current === 'idleActive' ? 'idlePassive' : current;
 		case 'reset':
-			return 'idle';
+			return 'idleActive';
 		default:
 			return current;
 	}

src/routes/demo/humanoid/+page.svelte

@@ -11,7 +11,7 @@
 
 	let status = $state<Status>('idle');
 	let statusMessage = $state('');
-	let avatarState = $state<AvatarState>('idle');
+	let avatarState = $state<AvatarState>('idleActive');
 	let listening = $state(false);
 
 	let client: ResilientRealtimeClient | null = null;