BUG: Cannot reject with a message in the main agent when `continue_loop_on_deny`

[pedro757]

Description

Cannot reject with a message, as you can see in the video below I never have the chance to tell the model why I reject the proposed changes. The agent goes and tries to suggest new changes immediately.

I have "continue_loop_on_deny": true set in the config

The guy in the comment below suggested that this only happens on the main agent, apparently in subagents it's already implemented, you can reject with a message (I haven't had the chance to check this)

https://discord.com/channels/1391832426048651334/1391832428024430645/1471967672710201465

He said that he has a fork that solves this. Maybe he could file a PR.

Plugins

none

OpenCode version

1.2.6

Steps to reproduce

1. With edit set to "ask" tell the model to make a simple edit to a file
2. Reject

Screenshot and/or share link

screenrecording-2026-02-19_11-29-35.mp4

Operating System

Linux

Terminal

Ghostty

[github-actions]

This issue might be a duplicate of existing issues. Please check:

• Rejecting use of a tool aborts the response entirely #7730: Rejecting use of a tool aborts the response entirely — same core behavior where rejecting a tool use doesn't allow the user to provide feedback/context to the model
• [FEATURE]: Add "soft deny" permission action for self-correcting AI behavior #9651: [FEATURE] Add "soft deny" permission action for self-correcting AI behavior — related feature request for allowing the model to receive a message on denial (now closed)

[yurihbm]

I guess that this is very important since it would prevent additional prompts to give feedback. This could even save money/costs when using AI plans with prompt-based charging like GitHub Copilot.

[XXiaoA]

Any update on this PR? This feature is essential for AI-assisted coding workflows. Would love to see it merged soon!