Skip to content
This repository was archived by the owner on May 15, 2026. It is now read-only.
This repository was archived by the owner on May 15, 2026. It is now read-only.

Quality of generated code problem #2652

@uzhao

Description

@uzhao

App Version

3.11.17

API Provider

VS Code LM API

Model Used

Claude 3.7 Sonnet

Actual vs. Expected Behavior

Actually I don't think this is an real issue or bug. Anyway, I watched a video (https://www.bilibili.com/video/BV15VR3YKErt?t=252.0) compared LLMs, and tried to reproduce its result (Mars Mission in https://github.com/KCORES/kcores-llm-arena). Later I found that compare to ask LLM directly, Roo-Code generated result is much worse. The result come from LLM chat work well just like the video one, while Roo-Code generated do not move at all and no spaceship launched. I've tried 3.7 and 3.5, VS Code API and my own token. Roo-Code result never use FuncAnimation or animation from matplotlib.animation which is a key point. This is just a single example but time to time I feel this problem in some other cases.

Detailed Steps to Reproduce

1.Create a new task
2.Ask "Generate code for an animated 3d plot of a launch from earth landing on mars and then back to earth at the next launch window"

Relevant API Request Output

Additional Context

No response

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions