Yes Please, fire away. 
3 Likes
How do you use this? 
This feature is enabled, but there’s no description of how it works or how to create task lists.
The agent never uses it
2 Likes
Guessing, just include something like ‘to-do’ in prompts, plus you can add even more specifications…
It did it by itself. I described the task, and of course I have a rule to break down the task into smaller tasks. after the task list was defined, it made the queue all by itself.
I have a plan in a ‘.md’ file with everything laid out step by step, but the agent doesn’t use this new feature and continues to work the old way
1 Like
you mean a rule? .mdc type, right?
No, I only use ‘.md’ files where I describe the plan; I never use ‘.mdc’ rules
you can create rules which are applied as you wish.
try “/Generate Cursor Rules” in the chat and describe your rule, or attach the .md file as the basis for the rule, and choose Always, or let the AI decide. I have a rule which is always used and some other which AI will selectively use.
To clarify, the todo list is a set of tools the model has that it can use as long-running memory and task management!
Some models, like o3, have so far proven to be more reluctant to use it, but if you tell a model to use it, it should do so just fine!
2 Likes
I understand long-running memory as something that can be preserved across conversations. But the only tool AI seems to be aware of is todo_write. It cannot display todos at will, and cannot remember todos from previous conversation.
So, how exactly can we make it a long-running memory?
I would also like to hear more on this. The “previous conversation”, would be a game changer.
I have had really good luck with it asides from this one time it has issues (This has been going on for 30+ min. so its starting to hallucinate etc. but overall the To-dos is 100% amazing.
1 Like
Hello, could you please elaborate on this a bit more. Shall it be prompted to queue tasks?
Will that workflow also help prevent these kind of things, where Cursor basically fails to follow the rules I have defined in my settings and cursorrules files. (it’s a funny answer in a way but also completely unhelpful)
Ive been using GitHub - snarktank/ai-dev-tasks: A simple task management system for managing AI dev in Cursor to great effect so far, it keeps an up to date to-do list, marks things off when complete, sticks to the rules set for it, and has the added bonus of letting you stop during the list and coming back to it in a new chat window to reduce context size
1 Like
Previous session chat context is available to the LLM.
If a task runs out of context and I restart a neww session, I begin the prompt with "Please review the context of the previous chat session to become aware of the progress stage we are at with “xxx task”. also familiarize yourself with the guidelines created for the execution of this task identified as rule “xxxx.mdc”
The model has access to the previous chat and demonstates that.
Always before I get into a big refactor or deploying a complex new logic module, I generate a rule that is a guidleine for that development session, then with a built in checklist of phases, milestones, this to do list can carry me through multiple long context sessions with a very high accuracy. As well Each phase in the guidlines, get marked as done.
Rules are 20% of my codebase, and I spend time making sure they are up to date, relevant or I delete them.
You can also drag and drop a rule into chat.
Could you share the rule code please.
Also any other .cursor/rules that you have set up and are ok with sharing.
I’ve recently started adding them to our project and it’s really helpful
1 Like
try using o3 - o3-Pro, or opus … previous session recall works for me
you would have to have, continued a previous chat, rather than starting a new chat I think for this to work for you. It is telling you that session is not a continuation in its response.