agent frequently hangs indefinitely (indicated by a persistent loading spinner) after a Sequential_Thinkingcall

[hgcode1130]

Describe the bug

When using Sequential_Thinking mode with either the augment agent or claude code, the agent frequently hangs indefinitely (indicated by a persistent loading spinner) after an mcp call. It fails to automatically proceed to the next thought or action, causing the entire task to stall.

To Reproduce

Steps to reproduce the behavior:

1. Start a task that requires multi-step reasoning using either augment agent or claude code.
2. Ensure the Sequential_Thinking mode is triggered.
3. Monitor the agent's execution flow until it makes an mcp call within its THOUGHT process.
4. Observe that after the current THOUGHT is complete, the agent gets stuck in a loading state and does not advance to the next step.

Workaround:
The user must manually stop the generation and then send a new message (e.g., "continue" or "go on") to prompt the agent to resume the task. However, the issue may recur in subsequent steps.

Expected behavior

Within a Sequential_Thinking flow, after completing a THOUGHT step that includes an mcp call, the agent should seamlessly and automatically proceed to the next THOUGHT until the task is complete, without requiring any manual intervention.

Logs

While detailed text logs are not available, the attached screenshot illustrates the point of failure. As shown, the agent has completed the 7th thought (THOUGHTNUMBER: 7) and recognizes that a next step is required (NEXTTHOUGHTNEEDED: true), but it fails to automatically advance to the 8th thought.

// State similar to the one in the screenshot
THOUGHT
... (mcp call and reasoning) ...
NEXTTHOUGHTNEEDED
true
THOUGHTNUMBER
7
TOTALTHOUGHTS
8
// <<< Agent hangs here and does not proceed to THOUGHT 8

Additional context

This issue occurs frequently and severely impacts the continuity of automated workflows for complex tasks. The manual workaround is inefficient and disrupts the user experience.