Agent Enters Infinite Reasoning Loop During Normal Usage

[lancechan00 (lance chan)]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

While using the agent under normal conditions, it unexpectedly enters a continuous reasoning loop (“repetitive thinking”) and fails to produce a final response. The agent appears to repeatedly process the same internal logic without progressing to an output.

This issue is not tied to a specific shutdown or hardware state. Instead, it occurs during standard interaction, where the agent should return a response but instead remains stuck in an ongoing internal processing cycle.

Steps to Reproduce

auto model , the same input, alway cycling,

Screenshots / Screen Recordings

Operating System

Windows 10/11

Version Information

Version: 2.6.20 (user setup)
VSCode Version: 1.105.1
Commit: b29eb4ee5f9f6d1cb2afbc09070198d3ea6ad760
Date: 2026-03-17T01:50:02.404Z
Build Type: Stable
Release Track: Early Access
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Windows_NT x64 10.0.26200

Does this stop you from using Cursor

No - Cursor works, but with this issue id:978cd1f6-abba-43c4-b91a-2f47af501c6b

[deanrie]

Hey, this is a known issue with Auto mode. Loop detection sometimes doesn’t catch a loop in the thinking block.

Similar cases were discussed here:

• Cursor stuck thinking: "CRITICAL INSTRUCTION 1: You may have access to a variety of tools at your disposal."
• Agent Gets Stuck on "Thinking" — Outputs Repeated begin_of_sentence | >%< Tokens

For now, as a workaround:

1. If the agent gets stuck, stop it and start a new chat
2. Try picking a specific model instead of Auto and see if the issue still happens

The team is aware of this. Can you share the Request ID of the chat where this happened? (top right of the chat > Copy Request ID) It’ll help us track what went wrong on the server side.