I am having a hard time knowing when the agent is done working on a prompt now. I am not sure when this started, but the last couple of days I cannot really tell when the agent is done, and I will often wait a while to see if it proceeds. There used to be some markers in the chat ui that made it pretty clear when the agent was done working. Those seem to be gone. They were a bit subtle (dark gray text) before.
I am not sure if the removal of these markers after the agent was done doing work was intentional? In any case, it would be nice to have a much clearer indicator that work on your last prompt is well and truly completed, to remove ambiguity about it and reduce wasted wait time.
Steps to Reproduce
Execute a prompt (especially one that does not change any code files, like planning efforts).
When agent/llm complete their work, there is no clear indicator that the work IS DONE.
Expected Behavior
Clear indicator when agent/llm work is in fact well and truly done.
Operating System
MacOS
Current Cursor Version (Menu → About Cursor → Copy)
So those occur at the TOP of the chats, or before a tool is called, etc. The issue I am having is, when the agent is done working…that often is not evident. I’ve had numerous cases where there was a fairly significant lag time between the last tool call completing, and the next appearance of one of those indicators that more work is being done. Especially, the bouncing elipses doesn’t appear immediately in all cases. Which is exactly why I’ve gotten into the habit of waiting, to see if it appears, before assuming the work is done. Sometimes the last message from the llm does not make it clear that its done.
Thinking more, maybe this is due to my current usage of GPT-5. GPT-5 is, and in some ways thankfully, more direct and to the point in getting work done, than some of the otehr models. Maybe it is that I became used to Sonnet’s approach, and perhaps it was finalizing its output at the end of work in away, that made it more apparent the work was really indeed done?
I am not sure if that is a system prompt sort of thing or not. In any caae, due to that potential lag between the last tool call completing (or last text output from the model completing), and the next appearance of bounding elipses, Generating Command or Planning next moves, without some other thing to indicate that the work is done and awaiting the next user prompt, sometimes I find myself wondering if the work is actually done. Some kind of indication that the agent is ready for the next prompt might be good. Doesn’t have to be overt, either…an icon that is added to the agent chat after the completion of the last output (and when no further work for the last prompt is pending) would just clear the issue up entirely.
I just discovered the option to play a sound when the agent is done. This seems to work some of the time, but not other times. At least, I don’t seem to get a chime every time the agent is done. It helps! Would be nice if it chimed every time the agent is done regardless (I wonder if, say, my cursor is over teh agent, or maybe in the prompt input textbox, if that stops the sound from being played?)
So, I am seeing a LOT of errors in the developer tools console. A whole heck of a lot. Some seem to be from extensions, a lot I am not sure. Do you want all of this?
For the diagnostic, that always seems to work. Even when I know there are issues occurring with the agent or models, the diagnostic has never caught any of them.
You can clear the console log before you start your Agent process. Then copy them after the issue occurs. While not very likely we have seen extensions interfering with processes and it would help exclude that issue.
Ok. It is definitely not playing on completion most of the time for me. I am not sure why. It seems that if the agent tab I’m expecting a chime from is in the foreground, it will not chime, except when I’m focused on another window, or potentially doing other things in cursor (i.e. if I’m on another agent tab.) It is either very flaky, or there seem to be some rules that limit when the chime occurs.
Agent rules do not influence Cursor app tool behavior, It is supposed to play on every completion. That’s why I was hoping something would show in Dev Tools Console.
I don’t mean agent rules. I mean, conditions implemented in your code base that prevent the chime from playing. As I don’t see any errors related to the lack of a chime…there is just lack of a chime. It primarily seems to be that the foreground agent tab, especially if the cursor is over it or the text cursor is in teh prompt input panel, then teh chime will not sound. If the tab is not foreground, then I’ll get a chime, or if I am off in an entirely different app (although, even getting a chime in that case is inconsistent.)
I don’t seem to get a chime every time an agent completes.
I am starting to think there may be a deeper issue occurring. I am noticing that there are a lot of….hiccups in the progress of any given agent chat. Not only do I not get a chime every time after the agent completes work on a prompt… I am finding over the last few days here, more and more so, that the agent is hanging on trivial tasks. Most often terminals…it’ll open a terminal, the command will execute, and then the agent will just hang. It will not continue or progress, no …, no Preparing next steps, no Generating. It just sits there and never reads the output of the terminal. I open the terminal instance, and the terminals are fine, the commands completed successfully, however the agent is not reading the results. This happens on commands as simple as pwd or as complex as rg or advanced git operations.
It is not just terminal commands, though. I’ve had it hang on executing MCP commands, and sometimes trivial operations like just grepping or reading a file, although those happen less often.
I don’t know if any of these are connected, but, its been troubling the last few days here, with the…seeming instability of the agent chat. If you would like me to open a different bug report for this, I can. I’ll see if I can identify if somehow an extension is causing the issues as well, however, I’ve encountered this kind of instability before withotu any extensions running.
Here is an example of a chat, that has been sitting at this point, for a very long time. It just stopped here, has not progressed, shows no indicator that work is being done, so I am honestly not sure what’s happening right now. I suspect it will eventually continue, but, this is behavior I experience quite frequently these days. Usually, this occurs when the entire process is just extremely slow…SLOOOW. Every aspect of it, not just communication with the LLM API, but every single aspect of the chat is just so slow, that I often don’t know if its done, hung, thinking, working, or what. There is just no indication of what is going on.
This has happened numerous times throughout the process I am working through, and often it happens right at or near the end of the work for the given phase. Hence the reason, I am frequently wondering if the work on the last prompt is in fact complete, or not.
If I wait long enough, often the agent will eventually continue. Sometimes that’s five minutes or so, though!
Current cursor details (this is the latest I can upgrade to on the path I’m on). I’ve been on this version for a while now (since it became available):
Sounds interesting! Thanks! Hopefully that will resolve the issue. This will work across multiple agent chat tabs? I usually have many open and several working at once on different things.
