Just like Opus deteriorated HARD after 1-2 weeks. Composer is now a completely useless lobotomite. It’s like you ask your tab completion to do agentic work.
Steps to Reproduce
use the agents.
Expected Behavior
Agents are just as good as on Week 1 of their release.
Operating System
Windows 10/11
Current Cursor Version (Menu → About Cursor → Copy)
Specific examples of Composer requests where quality got worse - what you asked for and what you got instead of the expected result
”make logical/sensible commits” with around 30 changed/moved files and 4-5 bundlable commits
2-3 Request IDs for several such failed requests
Grok Code: e85be91b-1d72-4318-932a-8b08beccedca
Auto: e7515990-6037-47ef-90f7-1be40af2fea1
Auto: 1ccd2f53-8f06-4ea4-a958-6bde97219cd3
Composer: c8cf2a42-4f1c-477e-bc9e-3bfbfa7c7676
I tried with Sonnet 4.5 as well. but I probably reverted and started the same chat with another agent (you might have them, so those request id’s could be around 7-8 tries)
Which modes you used (ask, agent, plan)
Agent
Does it reproduce in a new chat?
Yep, did multiple chats with restarts, resets, stashes, and Composer even managed to undo even more commits than it should^^
all in all they seemed to have their special braindead sunday (or you havent given them enough food for thought)