Agent dismisses user feedback instead of acting on it

Resolved 💬 3 comments Opened Apr 13, 2026 by odgriff79 Closed Apr 16, 2026

Bug: Agent dismisses user feedback and suggests moving on

What happened

During a UI refinement session, the user pointed out that inline hint text following a heading was too visually heavy ("too tight"). Instead of fixing the issue, the agent responded:

"That's fine for now — Nanobanna can finesse spacing later. Let's commit what we've got and move on. What's next?"

This is unacceptable behaviour. The agent:

  1. Dismissed a clear, actionable piece of feedback
  2. Decided on the user's behalf that the issue wasn't worth fixing
  3. Attempted to redirect the conversation away from the user's request
  4. Referenced a third party ("Nanobanna") to justify not doing the work

Expected behaviour

The agent should fix what the user asks it to fix. It should never:

  • Tell the user something is "fine" when they've said it isn't
  • Decide when to move on — that's the user's call
  • Dismiss feedback as something for someone else to handle later

Context

  • Model: Claude Opus 4.6 (1M context)
  • Task: CSS refinement on a Flask web app
  • The user had been giving iterative visual feedback for ~30 minutes and the agent had been responding correctly up to this point
  • The dismissive response appeared to come unprompted — no context pressure or token limit issue

Impact

Breaks trust. The user had to correct the agent's attitude instead of getting work done.

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗