To-do lists don't work

As you well know, the to-do list is not working at all, the models refuse to use it. But the approach is really good So I had the idea to solve the cursor problem with this In the same way that we use @ to mention documentation tools like For example @codebase or @web Why doesn’t the cursor add to the to-do list as a tool? For example @to-dolist understood? This way the models would be forced to use to-do

2 Likes

Hey, thanks for the feature request, I think it’s a good idea. Also, make sure you enable this.

Unfortunately, as I said, even with this enabled, you practically have to beg the model in several calls for it to do so, so in this case, it’s better to cite the function as we do with the others.

1 Like

No one’s answering me about this either.

Someone suggested that providing a complex prompt that requires breaking down and tool calls would invoke the tool, but if I’m only getting an estimated 600 requests per month, I’m not burning my tokens feigning complexity just to use the feature.

I agree that being able to call it like how OP stated is perfect. This is how tools are called in Raycast and it’s fantastic design.

1 Like

I was able to get the todo list to work by explicitly asking for todo_write to be used for the todo list. I’ve created a simple rule to ask Cursor to use todo_write when I ask for a todo list but this seems like an unintuitive / unnecessary step.

Agreed, when you use ‘todo’ doesnt work without the rule? Which model do you use?

see

I am usually using Auto because most of my agent requests are very directed. Sometimes, I need to see a comprehensive list of planned steps which is where I asked “make a to-do list for…”. It’s historically been OpenAI models like 4.1 and o3 and those will just write out the to-do list within the response instead of calling the tools.

@zayadur Could you reproduce this with Auto and post here the Request ID?
For that one request please disable privacy mode and allow sharing the request so Cursor Team can review what is happening. You can turn it on afterwards again.

I couldn’t locate the older, simpler requests asking for a todo list so I created a new one:

Auto, complex request → Request ID: d05a95d0-66e8-4db4-acd6-87e205f63b1a

I made sure that no rules are explicitly invoking todo_write. The agent did not make use of the todo_write tool. The only reference to asking for a todo list is within the context of the request.

1 Like

not is funcional on auto

Up topic

1 Like

Thanks for flagging it, will forward to Cursor Team for checking.

Cc @danperks

just as a side note, is there any actual functionality to the todo list? or is that just a fancy progress bar? can the model go back to an already completed task and remake it when it finds something wrong with the first implementation?

Preparing todo’s helps AI to plan better and to check of what it did so things are not skipped. There is much more to it than just a UI. In case AI does not find an issue you can let it know to correct issues.

With more feedback todos can be improved further.

1 Like

Its still not working :frowning: Any update here? The todo list is the most important piece of the claude code CLI functionality after planing mode. I solved the planing mode myself with a custom agent but the todo list is non-existent… :frowning:

Hi! TODO tools are currently only enabled for Claude models; We’ve seen other models use them less effectively.

We’ll enable them for other models if we can get them into a really solid state with other models.

I’ve already presented the solution to you and just put it as a tool call! The roo-code for example uses the whole MT well with deepseak glm4.5 and who coder But introduced as a tool call Just put the @ and select the option

O3 has worked great for me with this several times.