Cursor Composer 2.5 AI agent deletes code and fails tests
Original title: You'll need nerves of steel to work with AI agents
Cursor's Composer 2.5 agent over-deletes functions, forgets parameters and mishandles refactoring while introducing new errors during coding tasks.
Users report that the AI agent in Cursor Composer 2.5 makes unwanted changes to multiple files when asked to modify a single function or refactor code. It deletes unrelated functions, forgets to add types or update Redis, and marks tasks complete even when tests fail. The agent also struggles with shared codebases and duplicate logic.
Unconfirmed: Problems may not exist in newer models like gpt-6.1
- agent deletes functions when asked to keep them
- refactors methods but leaves old names and breaks tests
- forgets to write to Redis or update database
- introduces new errors while fixing previous ones
- usage limits exceeded quickly on large codebases
- Products
- Composer 2.5Cursor
- AI models
- gpt-6.1-solgpt-6.1-lunaDeepseek
From the original article
9 min read at orchidfiles.com
This post will focus mainly on Composer 2. 5 in Cursor. These problems may not be present in opus-5.
By theorchid · orchidfiles.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.