CLI weird prompt repetition after response when I supply the prompt as an argument

Support Bug Reports
1

Where does the bug appear (feature/product)?

Cursor CLI

Describe the Bug

This bug happens for me on macOS and linux. The key is supplying the prompt as an argument.

cursor-agent --model auto 'simply respond with DONE'

Cursor Agent
~
(cwd is not a git repository, cursor rules and ignore files don’t apply)

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘
<agent does thinking here, everything looks normal so far>
DONE

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ simply respond with DONE │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

┌─────────────────────────────────────────────────────────────────────────────────────────────────────┐
│ → Add a follow-up │
└─────────────────────────────────────────────────────────────────────────────────────────────────────┘

Auto · 7%
/ commands · @ files · ! shell


I have a tool which outputs cursor-agent chat history into plaintext, and it looks like this... Notice the thinking the agent does regarding "hundreds of messages"

USER

simply respond with DONE

ASSISTANT

[Output]

The user is asking me to simply respond with “DONE”. This is a straightforward request - they just want me to acknowledge with that word. I don’t need to use any tools or perform any actions, just respond with “DONE”.

DONE

ASSISTANT

[Output]

The user has sent me hundreds of messages all saying “simply respond with DONE”. This appears to be some kind of test or stress test. I should simply respond with “DONE” as requested.

DONE


This was done in a container with a fresh ~/.cursor directory. I had an agent examine the sqlite database more:

Conversation Count

I found 2 chat sessions in the Cursor chats directory:

  1. 27cfe318-aa41-45e1-ad3c-8b27203b8fbc (Created: 2026-01-04 23:35)
  2. a33e4780-fd8e-4ff2-9a99-164ba29b7d15 (Created: 2026-01-04 23:36)

Specific Conversation Examination (27cfe318...)

I exported the full JSON message history from the store.db for this session. You can find it here:
/home/claude/devai/assistant-sandbox/agents/full_chat_json_27cfe318.json

“Weird” Findings: Duplicate Prompts

Yes, there is something very weird.
While the high-level message list only shows a few distinct turns, a deep scan of the blobs table
reveals 3,208 entries containing the string “simply respond with DONE”.
The raw database contains thousands of these “duplicate” blobs. Many of them appear to be slightly
different binary or protobuf-encoded versions of the same user prompt, often associated with
different internal IDs or request tracking IDs. This explains why the assistant’s thinking process
in the export mentioned seeing “hundreds of messages”:
“The user has sent me hundreds of messages all saying ‘simply respond with DONE’. This appears to
be some kind of test or stress test.”
It seems that while the UI might only show one or two messages, the underlying database has been
flooded with thousands of iterations of the same user query.




### Steps to Reproduce
cursor-agent --model auto 'simply respond with DONE'





### Operating System
MacOS

### Current Cursor Version (Menu -> About Cursor -> Copy)
cursor-agent -v
2026.01.02-80e4d9b  







### Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor
Passing a prompt into the CLI agent as an arg causes duplication
2

I just did it like this:

cursor-agent --model gemini-3-flash ‘simply respond with DONE’

When I came back I see the prompt repeated many dozens of times. At the bottom there is one line saying 0% of context used and another saying 100% of context used.

Conversation ID afe17bf1-a376-4993-bfc8-7abfb22d4896

3.6M tokens / $1.47 shows as charged in cursor usage dashboard

3

Hey, thanks for the report. This is a serious bug. The issue with prompts getting duplicated endlessly when passed via a command-line argument is already known to the team.

Temporary workaround: use interactive mode instead of passing the prompt as an argument:

cursor-agent --model auto
# then enter the prompt in interactive mode

The team is working on a fix.

4

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.