WHATSSS
v1.5 notapads missing
Why hasnât a feature similar to âaugmentâ for âoptimizing prompts; enhancing promptsâ been launched yet? I think this feature is particularly friendly to programmers.
4 Likes
Has there been any consideration abut how the agent generates TODO lists and how it operates on them? Iâm starting to go out of my mind here with these todo lists.
The agent generates them automatically, but arbitrarily. When it generates a TODO list, it WILL NOT just progress through all the work Iâve prompted it to do, without stopping after EVERY SINGLE todo task. This is MADDENING, as I am often doing other things and I cannot seem to get the darn agent DO ALL the work Iâve requested it to do, without requiring me to stop, find the tab thatâs paused, and tell the darn thing to continue, repeatedly.
Even when you tell the agent to continue without stopping, it INSISTS on stopping after EVERY SINGLE todo task. I have 16 tasks here, and its gone through 5. Iâve had to tell it to continue 5 times, every few minutes, which really hampers how quickly I can get things like this done. I did not actually request that a todo list be created. In fact, my original prompt told it to iterate through each story and complete them automatically, and yet it still created a todo list.
At the very least, the agent should be able to be instructed to stop strictly following a start/task/stop paradigm if the user just wants all teh work to be completed without pauses or demands for user input. Each of these tasks, while they take a few minutes each, all follow a very strict and consistent pattern. I just need the work done, I donât want to have to babysit the whole process. I mean, whatâs the agent good for, if you canât have it just doing some mundane work for you in a background tab while you are solving other problems in other chat tabs?
After the chat goes in to summarizing, iâm getting sluck in a loop where the model just keeps eating tokens.
76313478-4667-497e-8029-cd36b97fe4b5
a5060608-71f1-4a87-b82f-c5388bb34846
Version: 1.5.4 (user setup)
VSCode Version: 1.99.3
Commit: f48f0974d52c41214775efcb96bbb5d7acd581c0
Date: 2025-08-22T20:11:36.742Z
Electron: 34.5.8
Chromium: 132.0.6834.210
Node.js: 20.19.1
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.19045
git bash as default terminal
I had it before this update as wel, but not with this frequency.
And just FYI: the first letter is still getting cut off of terminal commands. However I tell it in the .cursurrules now to start terminal commands with a blank space, which seems to help.
That was fixed in 1.5.4 - thank you.
1 Like
@alindsay55661 thats what I meant. You do not need to click keep all as its automatic with what I described above.
FWIW, I disabled TODOs entirely, and just had the agent start another TODO list on me. This is with GPT-5. So Iâm again stuck in this spurt, wait, spurt, wait, spurt cycle that I have to babysitâŚ
The GPT-5 agent integration really needs improvement. The thing doesnât seem to have a good general grasp of rudimentary concepts. As an example, I fairly frequently have had to request that files be moved or renamed lately, due to other small mishaps with this model in terms of naming or putting things in the right location. EVERY time Iâve asked it to MOVE or RENAME something, it deletes the original and recreates a new file, usually with something similar but not the same as the original file. I request things to be moved/renamed for two reasons:
-
AVOID issues with the probablistic nature of LLMs that produces non-deterministic results!!!
-
PRESERVE my git history for all files.
The model (and agent) just donât seem to grasp some of these rather rudimentary concepts. Even without the above two points, when a MOVE or RENAME is requested, the agent and model, should move and rename. Not delete and recreate. The latter is ENTIRELY different, destructive, and dangerous (especially if a file isnât actually version controlled.)
Rudimentary concepts. I can understand if a model, especially a simpler one, doesnât grasp the complex concepts we software engineers deal with each day. This is why prompt craft is so important. However, even WITH detailed prompts, models donât seem to grasp the rudimentary concepts a lot of the time. Even when you carefully specify that you want a file to be moved from one directory to another so that its history is preserved, the model will STILL delete the original and recreate some new (but not identical) version of it at the destination location. There are a lot of mundane, rudimentary things that we software developers do on a day to day basis, that are the perfect candidates to offload to the agent, so we can be freed up to put our minds to more complex problems.
Only, however, if the agent and model can handle it, though. Lately, to work around these problems, I have been creating a ridiculous library of rules that cover these really mundane, simple, rudimentary concepts, so that the model knows what to do every time I ask it to do something. Bummer is, all of these rules are âAlways Applyâ because there is no other way to guide the agent BEFORE it starts working with files (i.e. to start its âthoughtâ processes on the right track before it even gets to working with files, or running commands so files arenât actually involved, therefor globs arenât an option). This is really packing tokens into every request, because I have to have so many rules that always apply now.
Would be really nice to get some QOL improvements here, maybe in 1.5 or 1.6, where the Agent could take care of baselining some of these rudimentary, mundane things in the system prompt/API call so that we donât have to arbitrarily send a â â â â -ton of rules to the model every request.
1 Like
You need to activate in Beta options
I also agree with you that this has been around for some time, probably this was not the focus of the course for now.
The visual changes will come back - we just want to make sure hanging is resolved
It has become much better with the new terminal. After the return of the old one, there is a feeling that there are more freezes.
I also liked that in the new terminal, the output is simultaneously visible in both the user console and the Agent console.
And now the Agent often tries to use |cat at the end of the call, but this often breaks PowerShell scripts.
If itâs not a secret, have you found out the reason for this behavior?
There wasnât a problem until June, right?
The beta has already begunâby version 1.5, Notepad has simply vanished.
I get them back through Ctrl+Shift+P, Notepad
Hey all! We fixed the couple of reports of âno message.â No update required.
2 Likes
i am now on Version: 1.5.4 (user setup)
Commit: f48f0974d52c41214775efcb96bbb5d7acd581c0
Date: 2025-08-22T20:11:36.742Z
The issue with chats scrolling back up every time instead of staying in place is still not fixed.
Please, this is really ruining the work.
1 Like
With multiple tabs working on different things how does it manage a revert in one tab? How are you all managing that personally? Wouldnt a revert in one tab mess up the other tabs?
2 Likes