Misleading, Browser Debugging, Difficulty with Complex Backend tasks, Terminal Process Stuck And Disrupting Previously Corrected Code

[OLIVER_A_SIMON]

Where does the bug appear (feature/product)?

Somewhere else…

Describe the Bug

1. Misleading Confirmation of Completion:
   I frequently encounter situations where Cursor AI states that a task is complete and working perfectly, but upon manual testing, no changes have been made or the issue persists. To verify this, I sometimes use a different AI tool for testing, where I input the expected behavior versus Cursor AI’s output, and the results often confirm that the work was not done despite the system’s confirmation.

2. Browser Debugging Ineffectiveness:
   While the built-in browser is a good feature, I find it extremely difficult to use for debugging. When I instruct Cursor AI to use its browser to test and fix a problem, it reports the fix is complete, but the original issue often remains unchanged when I test it myself. This forces me to rely on external AI tools to effectively debug my code.

3. Difficulty with Complex Backend Tasks:
   I’ve been using Cursor AI for four months, and while it is fast and effective for simple code creation and finding, I find it struggles significantly with complex backend development. The AI consistently claims that the complex backend code is working perfectly, but testing reveals that nothing is happening as intended.

4. Terminal Process Stalling:
   Occasionally, when Cursor AI is executing a process in the terminal, the operation gets stuck. I have to manually cancel the task before the AI can proceed to other actions.

5. Disrupting Previously Corrected Code:
   When solving one problem, Cursor AI frequently makes unintended changes that break other parts of the code that were previously correct. This back-and-forth correction process makes complex development highly inefficient.

Steps to Reproduce

Try to make a different AI through Cursor AI and make a complex backend so you see all these problems here.

Because I make what I want to make, but I am facing the problem in the backend. My backend is complex, and Cursor AI has not fixed it for 20 days.

And it’s hard to tell you what I am making, but I am making an AI.

Screenshots / Screen Recordings

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.1.47 (system setup)
VSCode Version: 1.105.1
Commit: 2d3ce3499c15efd55b6b8538ea255eb7ba4266b0
Date: 2025-12-04T02:31:50.567Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Windows_NT x64 10.0.26200

Does this stop you from using Cursor

No - Cursor works, but with this issue

[deanrie]

Hey, thanks for the report. I see three related issues: browser debugging is misleading, the terminal hangs, and the agent breaks already fixed code.

This is a known set of issues. Please check in order:

1. Terminal hangs - most common root cause of the others:
   Settings > Agents > “Legacy Terminal Tool” - try toggling it, then Cmd+Shift+P > “Terminal: Kill All Terminals”, restart Cursor.
   If you use Oh My Zsh/Powerlevel10k/Starship - they can break the agent’s command completion detection. Details: 🛠️ Guide: Fix Cursor Agent Terminal Hangs Caused by .zshrc

2. Browser debugging for backend:
   The built-in agent browser is fine for UI tests, but not great for complex backend tasks. For full debugging, use MCP Browser Tools:

   • Install GitHub - AgentDeskAI/browser-tools-mcp: Monitor browser logs directly from Cursor and other MCP compatible IDEs.
   • Chrome extension + npx @agentdeskai/browser-tools-server@latest
   • Configure MCP in Settings > Tools & MCP
     This gives the agent access to console, network logs, and errors without manual copy-paste.

3. Agent breaks fixed code:

   • Use Checkpoints in Composer to roll back: Checkpoints | Cursor Docs
   • Start a new chat after 8-10 iterations - old context gets confusing
   • Use Review before applying changes: Review | Cursor Docs

4. Extensions can cause hangs:
   Run cursor --disable-extensions and see if it still reproduces.

To diagnose precisely, please share:

• Request IDs of problematic requests (Privacy Mode off): Getting a Request ID | Cursor Docs
• Which shell you use (zsh/bash/fish) and any prompt themes (oh-my-zsh, powerlevel10k, starship)
• A screenshot of the agent’s stuck terminal
• The specific command that hangs
• Your installed Cursor extensions
• Settings > Network > Run Diagnostics - the result

Share the details and I can help more precisely or escalate to the team.

[OLIVER_A_SIMON]

Hi Dean, thanks for the detailed reply and suggestions. I followed your instructions: I tried toggling the Legacy Terminal Tool, killed all terminals, restarted Cursor, and also installed and configured the MCP Browser Tools as described. However, I am still facing the same issues with terminal hangs, misleading browser debugging for my complex backend, and the agent breaking code that was already working. I will collect and share the request IDs, my shell and prompt theme, a screenshot of the stuck terminal, the exact command that hangs, my installed extensions, and the Network Diagnostics result so you can help more precisely or escalate this to the team.