Multi-agent mode leaks plans between models when updating a single agent’s plan

[bastienlimbour (Bastien Limbour Pruvost)]

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

In multi-agent mode, when using multiple models simultaneously, the conversation scopes between agents are not properly separated. After asking two different models to generate plans for the same task, I requested additional clarifications from Model 2 and asked it to revise its plan accordingly. Instead of responding only with its updated plan, Model 2 first displayed the plan generated earlier by Model 1, followed by its own modified plan.
This cross-contamination of context is not expected. Each agent should operate within its own isolated scope, and one agent’s plan should never appear in another agent’s conversation.

Steps to Reproduce

1. Enable Multi-agent mode with at least two models (Model 1 and Model 2).
2. Ask both models to generate their own plan for the same task.
3. Provide clarifications to Model 2 only.
4. Ask Model 2 to update or refine its plan based on the clarifications.
5. Observe that Model 2 outputs Model 1’s plan before its own updated plan.

In my case the models are:

• Model 1: Claude Sonnet 4.5 Thinking
• Model 2: Composer 1

Expected Behavior

In multi-agent mode:

• Each model should maintain an isolated conversation scope.
• Model 2 should only reference its own plan when updating it.
• No content from Model 1 should appear in Model 2’s response.

Operating System

MacOS

Current Cursor Version (Menu → About Cursor → Copy)

Version: 2.0.77
VSCode Version: 1.99.3
Commit: ba90f2f88e4911312761abab9492c42442117cf0
Date: 2025-11-13T23:10:43.113Z
Electron: 37.7.0
Chromium: 138.0.7204.251
Node.js: 22.20.0
V8: 13.8.258.32-electron.0
OS: Darwin arm64 24.6.0

For AI issues: which model did you use?

• Model 1: Claude Sonnet 4.5 Thinking
• Model 2: Composer 1

For AI issues: add Request ID with privacy disabled

Impossible to get the request id of another model than the first in Multi Agent Mode (probably another bug…)

Does this stop you from using Cursor

Sometimes - I can sometimes use Cursor

Screenshots

[deanrie]

Hey, thanks for the report. This definitely shouldn’t be happening, each agent should have its own isolated conversation scope.

I’ll pass this on to the team. The behavior you’re describing (Model 2 showing Model 1’s plan) suggests there is a conversation context isolation issue in multi-agent mode.

For now, you can work around this by running agents one at a time in separate chat sessions instead of using multi-agent mode.