Repository navigation
Chained prompts executing together instead of one after the other #3098
Description
Exactly as the title is,
huge fan of OpenCode but really miss chaining prompts,
it seems like when the agent reads files or does an action it ignores any other consecutive task and takes the next lined up prompt instead. I've seen this happen more frequently when the next lined up prompt is a command.
Metadata
Metadata
Assignees
Labels
No labels
Activity
github-actions commentedon Oct 11, 2025
This issue might be a duplicate of existing issues. Please check:
Feel free to ignore if none of these address your specific case.
rekram1-node commentedon Oct 11, 2025
does this relate to queuing? I just wanna make sure I understand what you mean by chaining prompts
bma-d commentedon Oct 11, 2025
Yes it's queuing, my bad.
goniz commentedon Oct 11, 2025
I'm thinking about this as a feature.
Some times I have multiple comments to steer the agent and I want it to see them all for the next turn without having to super engineer the prompt
rekram1-node commentedon Oct 11, 2025
Wdym by next turn? Are you saying when you queue they queue too quickly?
rekram1-node commentedon Oct 11, 2025
Right now we do the same thing that claude code does
goniz commentedon Oct 11, 2025
I don't think anything needs changing.
I like it that I'm able to send multiple queued messages while the agent is still working.
When it's done it will receive all queued messages in a single request instead of going through them one by one.
Attaching an "example" conversation: https://opencode.ai/docs/s/xvgdyOQZ/
rekram1-node commentedon Oct 11, 2025
Ah sorry I think I misread your comment:
my bad
bma-d commentedon Oct 11, 2025
I think Codex handles this the right way, as in you have the option to edit the previous message using alt + ↑ but also queue.
rekram1-node commentedon Oct 12, 2025
Okay I played with codex, opencode intentionally has a different queueing strategy
But I did see the alt + arrow, which allows you to edit queued message, we could add that
bma-d commentedon Oct 21, 2025
That would be fantastic, I would be primarily using OpenCode if this was added 👍
rekram1-node commentedon Oct 21, 2025
Good to know!
We are in the midst of a rewrite of the tui but after that is done (this week) we can readdress the ux around some of these things better
sai-sy commentedon Nov 24, 2025
@rekram1-node I'll "chain" prompts and it will forget some of them. even if I just sent one then after a little bit send another, it might fully abandon the first task when it get's a break from it. gpt-5-codex haven't tested on gpt-5.1-codex
98caa9deab129460e181aae781d377 commentedon Dec 1, 2025
I've been running into this too — queued messages sometimes get lost or conflated. I know models can struggle to keep track of things in longer contexts anyway, but consecutive user messages currently serialize with nothing between them, which might not be helping.
Also feels a bit off that hitting enter to send separate messages doesn't preserve any separation.
Put up a one-liner PR (#4939) that just adds a newline between them. Perhaps a middle ground — keeps the batching but with clearer separation. Small tweak, not sure how much it helps.