fix: propagate executor errors instead of synthesizing failed tasks

[JakubWorek]

Description

Aligns JS SDK with Python SDK: when AgentExecutor.execute() throws, the framework now persists the task as FAILED and propagates the exception to the transport, instead of silently returning a synthesized FAILED-Task as a successful result.

Before: executor throws → framework catches → synthesizes Task{state=FAILED} → publishes on bus → returns as successful JSON-RPC result (HTTP 200).

After: executor throws → framework publishes internal AgentEvent.error(err) → ExecutionEventQueue.events() rethrows → drain loop catches → persists FAILED status to store → rethrows → transport maps to proper error envelope.

Fixes #370 🦕

[github-actions]

🧪 Code Coverage

⬇️ Download Full Report

	Base	PR	Delta
src/server/events/execution_event_bus.ts	92.12%	92.16%	🟢 +0.04%
src/server/events/execution_event_queue.ts	94.44%	94.82%	🟢 +0.38%
src/server/request_handler/default_request_handler.ts	87.96%	86.65%	🔴 -1.31%
src/server/result_manager.ts	82.19%	81.53%	🔴 -0.66%
Total	90.67%	90.57%	🔴 -0.10%

Generated by coverage-comment.yml

[gemini-code-assist]

Code Review

This pull request refactors agent execution error handling by introducing an internal 'error' event on the event bus. Instead of synthesizing FAILED events in the handler, the queue rethrows the wrapped error, allowing the drain loop to persist the FAILED status and propagate the original exception to the transport layer for proper error envelope generation. The review feedback correctly identifies a critical issue in both _runExecutor and _runStreamExecutor where the hadError parameter of _settleBus is hardcoded to false in the .finally block, which prevents proper cleanup and can leave the event bus dangling on executor rejection.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

[gemini-code-assist]

The hadError parameter of _settleBus is hardcoded to false here. If the executor rejects with an error, this is a terminal failure and the event bus must be cleaned up immediately, even if the last observed state was an interrupted state (like INPUT_REQUIRED or AUTH_REQUIRED). Hardcoding false prevents the bus from being cleaned up in those cases, leaving it dangling.

We should track whether an error occurred and pass that state to _settleBus.

    let hadError = false;
    this.agentExecutor
      .execute(requestContext, eventBus)
      .catch((err: unknown) => {
        hadError = true;
        console.error('Agent execution failed for message ' + finalMessageForAgent.messageId + ':', err);
        // Publish an internal error event so the drain loop's
        // try/catch observes the exception via eventQueue.events()
        // rethrowing it. The drain loop then persists a FAILED status
        // update and rethrows so the transport can build a proper
        // error envelope.
        eventBus.publish(AgentEvent.error(err));
      })
      .finally(() => {
        // Force-close on error regardless of last observed state:
        // errors are terminal and must not leave the bus dangling for
        // an INPUT_REQUIRED / AUTH_REQUIRED holdover.
        this._settleBus(taskId, eventBus, stateTracker(), hadError);
      });

[gemini-code-assist]

Similar to _runExecutor, the hadError parameter of _settleBus is hardcoded to false here. If the streaming executor rejects, the event bus must be cleaned up immediately to avoid leaving it dangling.

We should track whether an error occurred and pass that state to _settleBus.

    let hadError = false;
    this.agentExecutor
      .execute(requestContext, eventBus)
      .catch((err: unknown) => {
        hadError = true;
        console.error(
          'Agent execution failed for stream message ' + finalMessageForAgent.messageId + ':',
          err
        );
        eventBus.publish(AgentEvent.error(err));
      })
      .finally(() => {
        this._settleBus(taskId, eventBus, snapshotTracker().state, hadError);
      });