fix(computer-use-mcp): bound waitForElement frame timeouts (#1856)

## Summary\n- Pass the remaining waitForElement budget into each
frame-level CU_ACTION send so unresponsive frames cannot consume the
fixed 8s sendMessage timeout.\n- Use the remaining deadline for each
poll and stop polling immediately when the budget is exhausted.\n-
Reduce the bridge-side waitForElement grace from the legacy 9.5s buffer
to a small transport grace.\n- Add a regression test covering the
hanging-extension case.\n\n## Validation\n- pnpm -C
services/computer-use-mcp exec vitest run --config ./vitest.config.ts
src/browser-dom/extension-bridge.test.ts

---------

Co-authored-by: Neko <neko@ayaka.moe>
Co-authored-by: 刘梓恒 <160735726+3361559784@users.noreply.github.com>
Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>

Author: 4 people

packages/i18n/src/locales/es/server/auth.yaml

@@ -66,7 +66,7 @@ signIn:
   footer:
     prefix: Al continuar, aceptas nuestros
     terms: Términos
-    and: "y"
+    and: 'y'
     privacy: Política de Privacidad
 verifyEmail:
   title:

packages/i18n/src/locales/ru/settings.yaml

@@ -743,7 +743,7 @@ pages:
         empty: Здесь пока ничего нет. Добавьте одно ниже!
       add:
         title: Новый
-        description: "Заполните новый сервер, затем нажмите кнопку «Сохранить и перезапустить» — он будет перемещен в «Конфигурация» выше."
+        description: 'Заполните новый сервер, затем нажмите кнопку «Сохранить и перезапустить» — он будет перемещен в «Конфигурация» выше.'
         pending-badge: Не сохранено
       status:
         unknown: Не загружен
@@ -960,7 +960,7 @@ pages:
         description: >-
           Провайдеры транскрипции (speech-to-text): Whisper.cpp, OpenAI, Azure Speech
       artistry:
-        title: Artistry 
+        title: Artistry
         description: Поставщики моделей генерации и создания изображений, например ComfyUI, Replicate.
         items:
           comfyui:

packages/i18n/src/locales/zh-Hans/settings.yaml

@@ -743,7 +743,7 @@ pages:
         empty: 这里什么都还没有哦，在下面添加一个！
       add:
         title: 新建
-        description: "填写新的服务器配置，然后点击「保存并重启」，完成后将会更新至上方的「已配置」"
+        description: '填写新的服务器配置，然后点击「保存并重启」，完成后将会更新至上方的「已配置」'
         pending-badge: 未保存
       status:
         unknown: 未加载

services/computer-use-mcp/chrome-extension/background.js

@@ -180,11 +180,19 @@ async function getActiveTab() {
  * Send a CU_ACTION message to a specific tab + frame.
  * msg_bridge.js (ISOLATED world) receives → postMessage → content.js (MAIN world)
  */
-async function sendCUAction(tabId, frameId, method, args) {
+function resolveActionTimeoutMs(timeoutMs) {
+  const numericTimeout = Number(timeoutMs)
+  if (!Number.isFinite(numericTimeout) || numericTimeout <= 0)
+    return SEND_CU_ACTION_TIMEOUT_MS
+
+  return Math.max(1, Math.min(Math.ceil(numericTimeout), SEND_CU_ACTION_TIMEOUT_MS))
+}
+
+async function sendCUAction(tabId, frameId, method, args, options = {}) {
   return new Promise((resolve) => {
     const timeout = setTimeout(() => {
       resolve({ success: false, error: 'sendMessage timeout' })
-    }, SEND_CU_ACTION_TIMEOUT_MS)
+    }, resolveActionTimeoutMs(options.timeoutMs))
 
     try {
       chrome.tabs.sendMessage(
@@ -213,7 +221,7 @@ async function sendCUAction(tabId, frameId, method, args) {
  * Run a CU_ACTION across all frames (or specified frames) in a tab.
  * Returns [{frameId, result}]
  */
-async function runCUAction(tabId, frameIds, method, args) {
+async function runCUAction(tabId, frameIds, method, args, options = {}) {
   let targets = frameIds
   if (!targets || (Array.isArray(targets) && targets.length === 0)) {
     const frames = await chrome.webNavigation.getAllFrames({ tabId })
@@ -225,7 +233,7 @@ async function runCUAction(tabId, frameIds, method, args) {
 
   return Promise.all(
     targets.map(async (fid) => {
-      const result = await sendCUAction(tabId, fid, method, args)
+      const result = await sendCUAction(tabId, fid, method, args, options)
       return { frameId: fid, result }
     }),
   )
@@ -465,9 +473,51 @@ async function handleCommand(cmd) {
         let lastFrameError = ''
 
         result = await new Promise((resolve) => {
+          async function resolveTimeout() {
+            if (lastFrames.length === 0) {
+              let frameIds = []
+              if (Array.isArray(cmd.frameIds) && cmd.frameIds.length > 0) {
+                frameIds = cmd.frameIds
+              }
+              else if (typeof cmd.frameIds === 'number') {
+                frameIds = [cmd.frameIds]
+              }
+              else {
+                try {
+                  const frames = await chrome.webNavigation.getAllFrames({ tabId })
+                  frameIds = frames.map(frame => frame.frameId)
+                }
+                catch {
+                  frameIds = [0]
+                }
+              }
+              lastFrames = frameIds.map(frameId => ({ frameId }))
+            }
+
+            const lastError = lastPollError || lastFrameError || undefined
+            resolve(lastFrames.map(entry => ({
+              frameId: entry.frameId,
+              result: {
+                success: false,
+                error: `timed out waiting for selector "${selector}"`,
+                selector,
+                timeoutMs,
+                ...(lastError ? { lastError } : {}),
+              },
+            })))
+          }
+
           async function poll() {
+            const remainingMs = deadline - Date.now()
+            if (remainingMs <= 0) {
+              await resolveTimeout()
+              return
+            }
+
             try {
-              const frames = await runCUAction(tabId, cmd.frameIds || null, 'findElements', [selector, 1])
+              const frames = await runCUAction(tabId, cmd.frameIds || null, 'findElements', [selector, 1], {
+                timeoutMs: remainingMs,
+              })
               lastFrames = frames
               const frameErrors = frames
                 .map(entry => unwrapBridgePayload(entry.result))
@@ -490,41 +540,12 @@ async function handleCommand(cmd) {
               lastPollError = e?.message || String(e)
             }
 
-            if (Date.now() >= deadline) {
-              if (lastFrames.length === 0) {
-                let frameIds = []
-                if (Array.isArray(cmd.frameIds) && cmd.frameIds.length > 0) {
-                  frameIds = cmd.frameIds
-                }
-                else if (typeof cmd.frameIds === 'number') {
-                  frameIds = [cmd.frameIds]
-                }
-                else {
-                  try {
-                    const frames = await chrome.webNavigation.getAllFrames({ tabId })
-                    frameIds = frames.map(frame => frame.frameId)
-                  }
-                  catch {
-                    frameIds = [0]
-                  }
-                }
-                lastFrames = frameIds.map(frameId => ({ frameId }))
-              }
-
-              const lastError = lastPollError || lastFrameError || undefined
-              resolve(lastFrames.map(entry => ({
-                frameId: entry.frameId,
-                result: {
-                  success: false,
-                  error: `timed out waiting for selector "${selector}"`,
-                  selector,
-                  timeoutMs,
-                  ...(lastError ? { lastError } : {}),
-                },
-              })))
+            const nextDelayMs = Math.min(WAIT_FOR_ELEMENT_POLL_INTERVAL_MS, Math.max(0, deadline - Date.now()))
+            if (nextDelayMs <= 0) {
+              await resolveTimeout()
               return
             }
-            setTimeout(poll, WAIT_FOR_ELEMENT_POLL_INTERVAL_MS)
+            setTimeout(poll, nextDelayMs)
           }
           poll()
         })

services/computer-use-mcp/src/browser-dom/extension-bridge.test.ts

@@ -237,6 +237,31 @@ describe('browserDomExtensionBridge', () => {
     ])
   })
 
+  it('waitForElement does not keep the legacy full send-timeout buffer when the extension hangs', async () => {
+    const result = await createConnectedBridge({ requestTimeoutMs: 5_000 })
+    bridge = result.bridge
+    client = result.client
+
+    // The mock extension deliberately does not answer waitForElement. The
+    // bridge should only keep a small transport grace on top of the requested
+    // wait budget, not the old 9.5s background send-message buffer.
+    client.on('message', (raw) => {
+      const data = JSON.parse(String(raw)) as Record<string, unknown>
+      if (data.action !== 'waitForElement')
+        return
+
+      expect(data.timeoutMs).toBe(100)
+    })
+
+    const startedAt = Date.now()
+    await expect(bridge.waitForElement({
+      selector: '#never-appears',
+      timeoutMs: 100,
+    })).rejects.toThrow('browser dom bridge timed out waiting for waitForElement')
+
+    expect(Date.now() - startedAt).toBeLessThan(2_500)
+  })
+
   it('waitForElement uses the default requestTimeoutMs for extension-side polling when no timeoutMs is provided', async () => {
     const result = await createConnectedBridge({ requestTimeoutMs: 200 })
     bridge = result.bridge

services/computer-use-mcp/src/browser-dom/extension-bridge.ts

@@ -23,7 +23,7 @@ const SUPPORTED_ACTIONS = new Set([
   'getComputedStyles',
   'waitForElement',
 ])
-const WAIT_FOR_ELEMENT_BRIDGE_TIMEOUT_BUFFER_MS = 9_500
+const WAIT_FOR_ELEMENT_BRIDGE_TIMEOUT_GRACE_MS = 1_000
 
 interface PendingBridgeRequest {
   reject: (error: Error) => void
@@ -375,17 +375,17 @@ export class BrowserDomExtensionBridge {
     frameIds?: number[]
   }) {
     const effectiveTimeout = params.timeoutMs ?? this.config.requestTimeoutMs
-    // NOTICE: The bridge-level timeout must exceed the background-level polling
-    // timeout, otherwise the bridge rejects before the extension finishes polling.
-    // The extension can overrun by one full frame send timeout (8s) plus the
-    // polling interval (500ms), so keep headroom for slow or unresponsive frames.
+    // NOTICE: The bridge-level timeout only needs a small transport grace: the
+    // extension now passes the remaining waitForElement budget into each
+    // frame-level send, so slow or unresponsive frames no longer require an
+    // extra full send-message timeout on top of the requested poll budget.
     return await this.callAction<Array<BrowserDomFrameResult<Record<string, unknown>>>>('waitForElement', {
       selector: params.selector,
       timeoutMs: effectiveTimeout,
       tabId: params.tabId,
       frameIds: params.frameIds,
     }, {
-      timeoutMs: effectiveTimeout + WAIT_FOR_ELEMENT_BRIDGE_TIMEOUT_BUFFER_MS,
+      timeoutMs: effectiveTimeout + WAIT_FOR_ELEMENT_BRIDGE_TIMEOUT_GRACE_MS,
     })
   }