fix(tui): recompute stale context usage ratio from status update token counts (#3164)

v2 engine status events carry contextTokens/maxContextTokens but never
contextUsage, so appState.contextUsage was only refreshed by getStatus
pulls and then went stale while the token counts kept updating live.
The /usage panel and footer render the ratio as a bar but recompute the
percentage text from the counts, so a stale ratio showed as a bar that
disagreed with the percentage (e.g. bar ~74% next to "18% (180k / 1M)"
after compaction or a model switch).

Recompute the ratio from the post-patch token counts whenever a status
update touches contextTokens or maxContextTokens without carrying an
explicit contextUsage. v1 events carry the ratio and are unaffected.

Co-authored-by: kimi-agent-bot <kimi-agent-bot@users.noreply.github.com>

.changeset/stale-context-usage-ratio.md

@@ -0,0 +1,5 @@
+---
+"@moonshot-ai/kimi-code": patch
+---
+
+Fix the context usage bar in /usage and the footer showing a stale percentage after the context size or model changes.

apps/kimi-code/src/tui/controllers/session-event-handler.ts

@@ -712,9 +712,20 @@ export class SessionEventHandler {
       this.host.state.appState.swarmMode &&
       this.host.state.swarmModeEntry === 'task';
     const patch: Partial<AppState> = {};
-    if (event.contextUsage !== undefined) patch.contextUsage = event.contextUsage;
     if (event.contextTokens !== undefined) patch.contextTokens = event.contextTokens;
     if (event.maxContextTokens !== undefined) patch.maxContextTokens = event.maxContextTokens;
+    if (event.contextUsage !== undefined) {
+      patch.contextUsage = event.contextUsage;
+    } else if (event.contextTokens !== undefined || event.maxContextTokens !== undefined) {
+      // v2 status events carry contextTokens/maxContextTokens but never
+      // contextUsage. Recompute the ratio from the post-patch token counts so
+      // it cannot go stale and drift from them — the footer and the /usage
+      // panel bar render this ratio while their texts recompute from the
+      // counts, so a stale ratio shows as a bar/percentage mismatch.
+      const tokens = patch.contextTokens ?? this.host.state.appState.contextTokens;
+      const max = patch.maxContextTokens ?? this.host.state.appState.maxContextTokens;
+      patch.contextUsage = max > 0 ? tokens / max : 0;
+    }
     if (event.planMode !== undefined) patch.planMode = event.planMode;
     if (event.swarmMode !== undefined) patch.swarmMode = event.swarmMode;
     if (event.towerMode !== undefined) patch.towerMode = event.towerMode;

apps/kimi-code/test/tui/kimi-tui-message-flow.test.ts

@@ -5596,6 +5596,89 @@ command = "vim"
     expect(stripSgr(renderTranscript(driver))).toContain('LLM not set');
   });
 
+  it('recomputes contextUsage when a status update carries contextTokens without it', async () => {
+    const { driver } = await makeDriver();
+    driver.state.appState.contextTokens = 0;
+    driver.state.appState.maxContextTokens = 1_000_000;
+    driver.state.appState.contextUsage = 0.74;
+
+    // v2 token-counting events carry contextTokens only; the ratio must be
+    // recomputed or the footer and /usage bar keep showing the stale value.
+    driver.sessionEventHandler.handleEvent(
+      {
+        type: 'agent.status.updated',
+        agentId: 'main',
+        sessionId: 'ses-1',
+        contextTokens: 180_000,
+      } as Event,
+      vi.fn(),
+    );
+
+    expect(driver.state.appState.contextTokens).toBe(180_000);
+    expect(driver.state.appState.contextUsage).toBeCloseTo(0.18);
+  });
+
+  it('recomputes contextUsage when a status update carries maxContextTokens without it', async () => {
+    const { driver } = await makeDriver();
+    driver.state.appState.contextTokens = 180_000;
+    driver.state.appState.maxContextTokens = 256_000;
+    driver.state.appState.contextUsage = 180_000 / 256_000;
+
+    // v2 profile events carry maxContextTokens only (e.g. a model switch).
+    driver.sessionEventHandler.handleEvent(
+      {
+        type: 'agent.status.updated',
+        agentId: 'main',
+        sessionId: 'ses-1',
+        maxContextTokens: 1_000_000,
+      } as Event,
+      vi.fn(),
+    );
+
+    expect(driver.state.appState.maxContextTokens).toBe(1_000_000);
+    expect(driver.state.appState.contextUsage).toBeCloseTo(0.18);
+  });
+
+  it('keeps an explicit contextUsage from status updates instead of recomputing', async () => {
+    const { driver } = await makeDriver();
+    driver.state.appState.contextTokens = 100;
+    driver.state.appState.maxContextTokens = 1_000_000;
+    driver.state.appState.contextUsage = 0;
+
+    driver.sessionEventHandler.handleEvent(
+      {
+        type: 'agent.status.updated',
+        agentId: 'main',
+        sessionId: 'ses-1',
+        contextTokens: 180_000,
+        maxContextTokens: 1_000_000,
+        contextUsage: 0.42,
+      } as Event,
+      vi.fn(),
+    );
+
+    expect(driver.state.appState.contextUsage).toBe(0.42);
+  });
+
+  it('zeroes contextUsage when a recomputation has no known context window', async () => {
+    const { driver } = await makeDriver();
+    driver.state.appState.contextTokens = 180_000;
+    driver.state.appState.maxContextTokens = 0;
+    driver.state.appState.contextUsage = 0.74;
+
+    driver.sessionEventHandler.handleEvent(
+      {
+        type: 'agent.status.updated',
+        agentId: 'main',
+        sessionId: 'ses-1',
+        contextTokens: 190_000,
+      } as Event,
+      vi.fn(),
+    );
+
+    expect(driver.state.appState.contextUsage).toBe(0);
+  });
+
   it('applies the effective thinking effort from status updates', async () => {
     const { driver } = await makeDriver();