Codex Desktop: allow per-selection follow-up comments/questions for selected text
What variant of Codex are you using?
Codex Desktop App
What feature would you like to see?
Add a per-selection follow-up workflow for selected text in Codex Desktop.
Instead of only collecting multiple selected text snippets into one shared context bundle for the next chat message, allow the user to attach a short comment/question to each selected snippet individually.
A possible flow:
- Select a phrase or passage in a prior assistant/user message.
- A small contextual action appears, for example
Follow-uporAdd follow-up. - Clicking it opens a compact popover anchored to the selection.
- The user types a question/comment specific to that selection, for example
What does this mean?orCan you verify this step?. - The follow-up is saved as a small chip/card that keeps the selected quote and its comment together.
- The user can repeat this for multiple selections before sending.
- The composer shows separate follow-up chips/cards, each with its own quote label and attached comment.
Why this would be useful
The current selected-text workflow is useful for attaching context, but it becomes hard to use when the user has several independent comments or questions about different snippets.
Today, the user often has to write one large chat message and manually explain which comment belongs to which selected snippet. That becomes ambiguous once there are multiple selections, especially when the snippets are similar or far apart in the transcript.
In practice, this pushes users toward copying text from the assistant response, pasting it into the chat manually, and writing comments inline. That defeats the purpose of having selected-text context in the UI.
Example use case
A user reads a long assistant answer and wants to ask about two different parts:
- Selection 1:
Test Run - Follow-up:
What does this mean? - Selection 2:
Login and install Olden Era - Follow-up:
And what about this?
These should remain separate, visible, and auditable as two follow-up items, instead of being merged into one selected-text bundle with one free-form user message.
Expected behavior
Codex should preserve the association between each selected text snippet and the user's follow-up text:
[1] "Test Run" - What does this mean?
[2] "Login and install Olden Era" - And what about this?
This association should be visible before sending and ideally also visible in the sent message/history view.
Relationship to another issue
This complements #22670, which asks for selected text references to remain visible in sent user messages/history. This request is specifically about attaching a distinct user comment/question to each selected text snippet before sending.
Reference UX
A similar workflow exists in Craft Agents as a Follow-up action: selecting text shows a contextual follow-up action, then a small popover accepts the user's question/comment and attaches it directly to that quote. The resulting composer state shows numbered follow-up chips that each combine the selected quote and the user-entered follow-up text.
5 Comments
The use case that made me look for this was replying to a specific part of Codex’s response while continuing the thread normally.
I had a long Codex response, wanted to address two sentences in it, and ended up copying those sentences into my next prompt so the reference would be explicit.
In addition to the multiple comments/questions workflow described here, I think it is important that Codex knows, and the UI shows, when the next prompt is responding to an exact part of its prior response.
That may fit the same solution, but I think the “reply to this part of the response” framing matters for how this feels in normal iteration.
We definitely need this. Or like cursor does it, where quotes you select becomes an inline block/badge, that you can address individually.
For what it's worth when you highlight part of a response in codex now it gives you an option to "add to chat" or "ask in side chat", at least we have some progress
Oh yeah, this option has been there for a very long time, no? What we need is the ability to address quotes individually instead of as a group.
Okay yeah I see what you're saying, basically a step further. I wanna say it's been around a little over a month, I started noticing it not too long after I commented on here