Context leaking between chats

[3b400654ae527bf14030]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Context leaking between chats. I finish a task in chat A, then open chat B for new task. The LLM responses in chat B include information from chat A

Steps to Reproduce

Context leaking between chats. I finish a task in chat A, then open chat B for new task. The LLM responses in chat B include information from chat A

Operating System

MacOS

Version Information

Version: 2.4.23
VSCode Version: 1.105.1
Commit: 379934e04d2b3290cf7aefa14560f942e4212920
Date: 2026-01-29T21:24:23.350Z
Build Type: Stable
Release Track: Default
Electron: 39.2.7
Chromium: 142.0.7444.235
Node.js: 22.21.1
V8: 14.2.231.21-electron.0
OS: Darwin arm64 24.6.0

For AI issues: which model did you use?

Opus 4.5 thinking

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

[deanrie]

Hey, thanks for the report. This is a known issue and the team is aware.

To help with the investigation, can you share:

• A specific example of what “leaked” from chat A into chat B
• The Request ID from the affected response in chat B (three dots in the corner of the message > Copy Request ID)
• Are the chats in the same workspace/project, or different ones?
• Does this happen every time, or only sometimes?

This will help us understand the scope of the bug and prioritize a fix.

[3b400654ae527bf14030]

i think it may only happen when invoking custom workspace commands
.cursor/commands

for a specific example: in a previous chat i made some clarification (maybe with my /update-plan command), then when using the /update-plan command in a new chat, the LLM responds as if my new message contains information from the previous time I used the /update-plan command

the chats are in the same workspace

this happens frequently

i think i can reproduce it by just asking the LLM to tell me the content of the message. here is my update-plan command. the LLM response in the screenshot includes other junk from ealier conversations.
```

1. update your plan with this information

2. do not write any code

3. present the updated plan
```

ID: edfde79c-4fe6-4fb8-bd07-5b7adf36bb50

[3b400654ae527bf14030]

another reproduction
1595fdf6-7e70-4d2c-8103-255e420f4a22
it is 100% related to the workspace commands

[iamakulov (Ivan Akulov)]

Hey, I ran into the same issue. In my case, I launched a chat, duplicated it, and then sent different commands into both duplicates in parallel. The context from one of the chats leaked into the other chats.

Here’s me prompting the original chat (chat A) to do something in /tmp/framer-slideshow-ssr:

Here’s the duplicated chat (chat B), showing mentions of /tmp/framer-slideshow-ssr in its reasoning trace, without me mentioning it at all:

Here’s me prompting the model why it went to that directory. It directly quotes the message I sent into chat A:

Both chats are super short – just two messages from me. The chat was duplicated after the first message.

In my case, I was not using workspace commands.

Request ID from the second (duplicated) chat: 3be56a1f-5409-4364-b1cd-1b3710db9eaf

[iamakulov (Ivan Akulov)]

Curiously, chat B keeps knowing about /tmp even if you re-send that second message that started the new branch.

[ku_s]

I encountered a similar issue, but not across two separate conversations — it happened within the same conversation. After executing a plan build, or after multiple rounds of dialogue, if I restart from the middle of the conversation, the model still seems to remember the previous inputs or the earlier build results.(When restarting the conversation from the middle, the dialogue content after that node should be removed. This issue did not exist in previous versions.)