Cursor Agent automatically injects false "previous response was interrupted" continuation after completed response

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

After an Agent response finishes normally, Cursor automatically triggers a new continuation/reasoning step a few seconds later even though I did not type anything, click Continue, or send any follow-up message.

The model’s thinking text says something like “the user mentioned the previous response was interrupted” or “continue from where it left off”, but that is not true. The previous response was already complete.

This causes the assistant to continue/repeat work unnecessarily, creates wrong conversation state, and may consume extra usage. I observed this in Cursor IDE on Windows, using Opus 4.7 Extra High, in a long Agent conversation with around 74–75% context.

Steps to Reproduce

  1. Open Cursor IDE on Windows.
  2. Open an Agent chat in a workspace.
  3. Use Opus 4.7 Extra High.
  4. Send a normal prompt and wait for the assistant response to finish completely.
  5. Do not type anything, do not press Enter, and do not click Continue.
  6. Wait about 5 seconds.
  7. Cursor starts another short thinking/continuation step by itself.
  8. The model says the previous response was interrupted, even though no interruption happened and no user follow-up was sent.

Expected Behavior

After the assistant finishes a complete response, Cursor should stay idle and wait for explicit user input.

Cursor should only send a continuation request when the response was actually interrupted or when the user explicitly clicks Continue / sends another message.

It should not automatically inject a hidden “previous response was interrupted” continuation after a completed response, and it should not consume extra usage for this false continuation.

Screenshots / Screen Recordings

DIIVY0CPPMSIH7JC_AI`XG9.png
DIIVY0CPPMSIH7JC_AI`XG9.png1286×1083 68 KB

Operating System

Windows 10/11

Version Information

Version: 3.5.17 (user setup)
VSCode Version: 1.105.1
Commit: d5b2fc092e16007956c9e5047f76097b9e626ca0
Date: 2026-05-20T02:43:31.559Z
Layout: glass
Build Type: Stable
Release Track: Default
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

For AI issues: which model did you use?

Opus 4.7 Extra High

For AI issues: add Request ID with privacy disabled

aba14128-4fd0-4be3-be5b-d0de8347f668

Does this stop you from using Cursor

No - Cursor works, but with this issue

4

Hey, thanks for the report with the screenshot and Request ID, I’ve got everything.

This is a known bug with the continuation injector. In some cases it falsely triggers on an already finished response and adds a hidden “previous response was interrupted”, which makes the model generate a new turn on its own. We’re tracking it, but I can’t share an ETA for a fix yet. I linked your case (Windows + Opus 4.7 Extra High + long context 74 to 75 percent) to the ticket, it’s a useful signal since most earlier reports were macOS or Sonnet.

Related thread with more details: Cursor automatically injects "Your previous response was interrupted" after every response

For now, workarounds:

  • If it starts looping, start a new chat. The bug is tied to the state of a specific conversation.
  • Try not to let the context grow too much. It triggers more often in long sessions.
  • If it reproduces reliably on one model, try switching to another one like Sonnet and see if it happens there too.

About usage, yes, each of these “continued” turns is a real request and counts toward your limit. If you hit your limit because of this bug, email [email protected] and we’ll see what we can do.

We’ll post here once we have an update on the fix.