[Bug] Agent assumes user intent and executes unasked changes despite instructions

Resolved 💬 2 comments Opened May 22, 2026 by steinybot Closed May 25, 2026

Bug Description
This pisses me off. Every f***ing time I ask it a question it assumes some kind of intention and goes rogue making changes that I then have to try and revert. I will say "why did you do X?". It responds with "you're right I shouldn't have done X" and then undoes it. I even have my global CLAUDE.md telling it that it must stop and answer every question immediately, assume not intention, and take no action. You all need to stop trying to game these stupid agentic benchmarks and focus on real world uses.

Environment Info

  • Platform: darwin
  • Terminal: iTerm.app
  • Version: 2.1.141
  • Feedback ID: 65b9c4fa-30d6-47c8-8be3-9a4dcde38c26

Errors

[]

View original on GitHub ↗

This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗