Replies: 1 comment
|
Completely agree. Coming from Claude Code, hitting Esc for rewind is a muscle memory at this point. Rn revert functionality is absolutely unusable with how it doesn't work outside of git repositories and has no shortcut. Also, revert is not be available when the execution was stopped before the agent finished answering (even though the agent still sees the prompt in the history). Claude Code has it right imo; we should probably just copy it's behaviour as is. |
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
Before submitting
Area
apps/server
Problem or use case
After sending a prompt, especially when working quickly, I often immediately realize that there's something I want to change in the prompt. I'm able to do this in Claude Code and I want to be able to do it in T3 Code, too.
The composer currently clears on send and the message moves into the conversation. Going back to the world in which you didn't send the message is clunky and requires clicking the stop button in the composer, copying the message back into the composer, reverting the conversation, and copying the originally-sent prompt back into the composer. There's no quick “undo send” functionality that allows returning the most recently sent message to the composer while it's still on the current message.
Proposed solution
When the most recent user message is eligible to be recalled, pressing Esc should move it back into the composer as an editable draft. It should work for both new threads and existing threads with messages in them.
The restored draft should preserve everything that can be re-sent, including:
It's a narrow, immediate undo for the message that the user just sent that's only available between the user sending the message and the assistant finishing thinking--just like in Claude Code. It should not be available once the assistant has fully started its turn.
Why this matters
Sending and immediately retracting a prompt for correction is a very common interaction when directing coding agents. Making that correction a single keystroke keeps the user in the keyboard flow and avoids needing to manually reconstruct the prompt. It also makes sending feel less irreversible/committed if the user can act quickly without worrying that a typo or premature submission will require a full rewind/reversion or manual copy/paste.
Smallest useful scope
The ability to recall the most recently sent message with Esc in the window between sending the message and when the assistant finishes thinking. This would only need to work for Claude and Codex messages only since I don't currently use any other providers.
Alternatives considered
Risks or tradeoffs
Examples or references
Working demo on my branch:
Screencast.from.2026-08-11.19-20-30.webm
Related-but-distinct upstream work:
Contribution
All reactions