fix(CS-11647): hide manual approval UI for auto-executed commands

packages/host/app/components/matrix/room-message-command.gts

@@ -26,6 +26,7 @@ import {
 
 import type { CommandRequest } from '@cardstack/runtime-common/commands';
 
+import { isAutoExecutableCommand } from '@cardstack/host/lib/command-auto-execute';
 import type MessageCommand from '@cardstack/host/lib/matrix-classes/message-command';
 
 import type { RoomResource } from '@cardstack/host/resources/room';
@@ -77,7 +78,33 @@ export default class RoomMessageCommand extends Component<Signature> {
     if (this.didFailCorrectnessCheck) {
       return 'applied-with-error';
     }
-    return this.args.messageCommand?.status ?? 'ready';
+    let status = this.args.messageCommand?.status;
+    // Mirror the Accept All bar fix: for any command the host will
+    // auto-execute (checkCorrectness, requiresApproval=false, LLM mode
+    // 'act'), present the applying spinner immediately on message-landed
+    // instead of the clickable Run button. Without this, the per-command
+    // Apply button flashes through 'ready' for the ~100ms debounce window
+    // before command-service starts the run. If validation later fails
+    // in the drain, command-service dispatches an `invalid` commandResult
+    // event and the button transitions to its invalid state — no risk of
+    // the spinner sticking.
+    if ((status === 'ready' || status === undefined) && this.willAutoExecute) {
+      return 'applying';
+    }
+    return status ?? 'ready';
+  }
+
+  private get willAutoExecute() {
+    let activeMode = this.args.roomResource.getActiveLLMModeForMessage(
+      this.args.messageCommand.eventId,
+    );
+    let isOwnedByCurrentAgent =
+      this.args.messageCommand.message.agentId === this.matrixService.agentId;
+    return isAutoExecutableCommand(
+      this.args.messageCommand,
+      activeMode,
+      isOwnedByCurrentAgent,
+    );
   }
 
   @use private commandResultCard = resource(() => {
@@ -233,7 +260,7 @@ export default class RoomMessageCommand extends Component<Signature> {
           @monacoSDK={{@monacoSDK}}
           @codeData={{hash code=this.previewCommandCode language='json'}}
           data-test-command-card-idle={{not
-            (eq @messageCommand.status 'applying')
+            (eq this.applyButtonState 'applying')
           }}
           as |codeBlock|
         >
@@ -253,7 +280,7 @@ export default class RoomMessageCommand extends Component<Signature> {
           @monacoSDK={{@monacoSDK}}
           @codeData={{hash code=this.previewCommandCode language='json'}}
           data-test-command-card-idle={{not
-            (eq @messageCommand.status 'applying')
+            (eq this.applyButtonState 'applying')
           }}
           as |codeBlock|
         >

packages/host/app/components/matrix/room.gts

@@ -55,6 +55,7 @@ import { DEFAULT_FALLBACK_MODELS } from '@cardstack/runtime-common/matrix-consta
 
 import UpdateRoomSkillsCommand from '@cardstack/host/commands/update-room-skills';
 import ENV from '@cardstack/host/config/environment';
+import { isAutoExecutableCommand } from '@cardstack/host/lib/command-auto-execute';
 import type { FileUploadState } from '@cardstack/host/lib/file-upload-state';
 import type { Message } from '@cardstack/host/lib/matrix-classes/message';
 import type { StackItem } from '@cardstack/host/lib/stack-item';
@@ -1861,13 +1862,25 @@ export default class Room extends Component<Signature> {
     if (!lastMessage || !lastMessage.commands) {
       return [];
     }
+    let roomResource = this.matrixService.roomResources.get(this.args.roomId);
+    let activeMode = roomResource?.getActiveLLMModeForMessage(
+      lastMessage.eventId,
+    );
+    let isOwnedByCurrentAgent =
+      lastMessage.agentId === this.matrixService.agentId;
     return lastMessage.commands.filter(
       (command) =>
         (command.status === 'ready' || command.status === undefined) &&
         !this.commandService.currentlyExecutingCommandRequestIds.has(
           command.id!,
         ) &&
-        !this.commandService.executedCommandRequestIds.has(command.id!),
+        !this.commandService.executedCommandRequestIds.has(command.id!) &&
+        // Commands destined for auto-execution must not surface the manual
+        // Accept All / Cancel bar, even during the ~100ms debounce before
+        // command-service flips `acceptingAllRoomIds`. Without this filter,
+        // the bar paints and then yanks itself once auto-execution starts,
+        // which is the CS-11647 glitch.
+        !isAutoExecutableCommand(command, activeMode, isOwnedByCurrentAgent),
     );
   }
 

packages/host/app/lib/command-auto-execute.ts

@@ -0,0 +1,35 @@
+import type { LLMMode } from '@cardstack/runtime-common/matrix-constants';
+
+import type MessageCommand from './matrix-classes/message-command';
+
+export const CHECK_CORRECTNESS_COMMAND_NAME = 'checkCorrectness';
+
+// Single source of truth for "this command runs without user approval".
+// Used by command-service (to decide whether to auto-run) and by the
+// room / room-message-command components (to decide whether to render
+// the Accept All bar and the per-command Apply button). Keeping all
+// call sites on the same predicate prevents them from drifting and
+// reintroducing the action-bar flash that prompted CS-11647.
+//
+// `isOwnedByCurrentAgent` mirrors the agentId gate in
+// command-service.drainCommandProcessingQueue: a command sent by
+// another agent is never auto-executed, even if it would otherwise
+// satisfy one of the three branches below. Callers that don't track
+// agents (e.g. unit tests) can pass `true` to focus on the other
+// conditions.
+export function isAutoExecutableCommand(
+  command: Pick<MessageCommand, 'name' | 'requiresApproval'>,
+  activeLLMMode: LLMMode | undefined,
+  isOwnedByCurrentAgent: boolean,
+): boolean {
+  if (!isOwnedByCurrentAgent) {
+    return false;
+  }
+  if (command.name === CHECK_CORRECTNESS_COMMAND_NAME) {
+    return true;
+  }
+  if (command.requiresApproval === false) {
+    return true;
+  }
+  return activeLLMMode === 'act';
+}

packages/host/app/services/command-service.ts

@@ -36,6 +36,10 @@ import type Realm from '@cardstack/host/services/realm';
 import type { CardDef } from 'https://cardstack.com/base/card-api';
 import type { CodePatchStatus } from 'https://cardstack.com/base/matrix-event';
 
+import {
+  CHECK_CORRECTNESS_COMMAND_NAME,
+  isAutoExecutableCommand,
+} from '../lib/command-auto-execute';
 import LimitedSet from '../lib/limited-set';
 
 import type LoaderService from './loader-service';
@@ -47,10 +51,15 @@ import type StoreService from './store';
 import type { CodeData } from '../lib/formatted-message/utils';
 import type MessageCodePatchResult from '../lib/matrix-classes/message-code-patch-result';
 import type MessageCommand from '../lib/matrix-classes/message-command';
+import type { RoomResource } from '../resources/room';
 import type { IEvent } from 'matrix-js-sdk';
 
 const DELAY_FOR_APPLYING_UI = isTesting() ? 50 : 500;
-const CHECK_CORRECTNESS_COMMAND_NAME = 'checkCorrectness';
+// How long drainCommandProcessingQueue waits for a room resource that's
+// still processing before giving up on the event. In tests we shorten this
+// so the stuck-timeout invalidation path can be exercised in a single test
+// without holding a real test open for a minute.
+const STUCK_PROCESSING_TIMEOUT_MS = isTesting() ? 1000 : 60_000;
 
 type GenericCommand = Command<
   typeof CardDef | undefined,
@@ -320,7 +329,7 @@ export default class CommandService extends Service {
             `Room resource not found for room id ${roomId}, this should not happen`,
           );
         }
-        let timeout = Date.now() + 60_000; // reset the timer to avoid a long wait if the room resource is processing
+        let timeout = Date.now() + STUCK_PROCESSING_TIMEOUT_MS; // reset the timer to avoid a long wait if the room resource is processing
         let currentRoomProcessingTimestamp =
           roomResource.processingLastStartedAt;
         while (
@@ -337,9 +346,23 @@ export default class CommandService extends Service {
           currentRoomProcessingTimestamp ===
             roomResource.processingLastStartedAt
         ) {
-          // room seems to be stuck processing, so we will log and skip this event
+          // Room processing is wedged. The synthetic 'applying' state in
+          // room-message-command.gts shows the spinner the moment an
+          // auto-executable command lands and only clears when we dispatch
+          // a terminal commandResult ('applied' or 'invalid'). If we just
+          // logged and continued, the spinner would hang indefinitely with
+          // no manual Run fallback. Mark each auto-executable command on
+          // this message invalid so the UI falls through to the
+          // invalidCommandState "Try Anyway" branch; manual-approval
+          // commands are left in 'ready' so the action bar's Run button
+          // remains the user's fallback.
           console.error(
-            `Room resource for room ${roomId} seems to be stuck processing, skipping event ${eventId}`,
+            `Room resource for room ${roomId} seems to be stuck processing, invalidating auto-executable commands on event ${eventId}`,
+          );
+          await this.invalidateAutoExecutableCommandsForStuckProcessing(
+            roomResource,
+            roomId!,
+            eventId!,
           );
           continue;
         }
@@ -379,26 +402,21 @@ export default class CommandService extends Service {
             continue;
           }
 
-          // Get the LLM mode that was active when this message was created
           let activeModeAtMessageTime = roomResource.getActiveLLMModeForMessage(
             message.eventId,
           );
 
-          // Auto-execute if LLM mode is 'act' AND the command came after the LLM mode was set to 'act',
-          // or if requiresApproval is false
-          let shouldAutoExecute = false;
-          let isCheckCorrectnessCommand =
-            messageCommand.name === CHECK_CORRECTNESS_COMMAND_NAME;
-
+          // The outer `message.agentId !== this.matrixService.agentId`
+          // gate above already short-circuited the not-our-agent case, so
+          // every command reaching this point is owned by the current
+          // agent.
           if (
-            isCheckCorrectnessCommand ||
-            messageCommand.requiresApproval === false ||
-            activeModeAtMessageTime === 'act'
+            isAutoExecutableCommand(
+              messageCommand,
+              activeModeAtMessageTime,
+              true,
+            )
           ) {
-            shouldAutoExecute = true;
-          }
-
-          if (shouldAutoExecute) {
             readyCommands.push(messageCommand);
           }
         }
@@ -422,6 +440,64 @@ export default class CommandService extends Service {
     }
   }
 
+  private async invalidateAutoExecutableCommandsForStuckProcessing(
+    roomResource: RoomResource,
+    roomId: string,
+    eventId: string,
+  ) {
+    let message = roomResource.messages.find((m) => m.eventId === eventId);
+    if (!message) {
+      return;
+    }
+    if (message.agentId !== this.matrixService.agentId) {
+      return;
+    }
+    let activeModeAtMessageTime = roomResource.getActiveLLMModeForMessage(
+      message.eventId,
+    );
+    for (let messageCommand of message.commands) {
+      if (this.currentlyExecutingCommandRequestIds.has(messageCommand.id!)) {
+        continue;
+      }
+      if (this.executedCommandRequestIds.has(messageCommand.id!)) {
+        continue;
+      }
+      if (
+        messageCommand.status === 'applied' ||
+        messageCommand.status === 'invalid'
+      ) {
+        continue;
+      }
+      if (!messageCommand.name) {
+        continue;
+      }
+      // The outer agentId gate already verified ownership, so this command
+      // is owned by the current agent.
+      if (
+        !isAutoExecutableCommand(messageCommand, activeModeAtMessageTime, true)
+      ) {
+        // Manual-approval commands stay 'ready' — the action bar's Run
+        // button is still the user's fallback for those.
+        continue;
+      }
+      let invokedToolFromEventId =
+        this.getCurrentEventIdForCommandRequest(
+          roomId,
+          messageCommand.commandRequest.id,
+        ) ?? messageCommand.eventId;
+      await this.matrixService.sendCommandResultEvent({
+        roomId,
+        invokedToolFromEventId,
+        toolCallId: messageCommand.commandRequest.id!,
+        status: 'invalid',
+        failureReason: `Room processing did not finish within ${Math.round(
+          STUCK_PROCESSING_TIMEOUT_MS / 1000,
+        )}s; command was not started`,
+        context: await this.operatorModeStateService.getSummaryForAIBot(),
+      });
+    }
+  }
+
   private async drainCodePatchProcessingQueue() {
     let waiterToken = commandProcessingWaiter.beginAsync();
     try {
@@ -444,7 +520,7 @@ export default class CommandService extends Service {
             `Room resource not found for room id ${roomId}, this should not happen`,
           );
         }
-        let timeout = Date.now() + 60_000; // reset the timer to avoid a long wait if the room resource is processing
+        let timeout = Date.now() + STUCK_PROCESSING_TIMEOUT_MS; // reset the timer to avoid a long wait if the room resource is processing
         let currentRoomProcessingTimestamp =
           roomResource.processingLastStartedAt;
         while (