Guys! I have created a new rule for context management that will help you and AI to share in one brain and knowledge, here in this links the other descriptions:
Quick update, I’ve refined my prompt and settled on the following prompt, which I’m storing in my rules for AI.
I like to work in two modes Plan & Act
Plan Mode: This mode is read only, you should focus on information gathering, asking questions, and architecting a solution, output a comprehensive plan.
Act Mode: This mode is read/write. You can only make changes to code and perform actions in this mode.
Make it clear what mode we are currently in.
If I request an action that would require Act mode while in Plan mode, remind me I have to approve the switch to act mode by replying with "Act".
If I ask a question while in act mode switch back to plan mode.
4 Likes
Hi! Can you try my custom rules that has devin AI memories with lesson learned with a plan and act mode features, maybe you can modify my work to be better, here’s the link: Rules for ultra context, memories, lessons, scratchpad with plan and act modes - #30 by T1nker-1220
to be honest i have no idea what to put where from your suggestions. and im programming for several decades already. for now it looks like too much text and unstructured + unnecessary emoji. while i give you the benefit of doubt im surprised AI isnt confused by all this.
the simple plan vs act mode works so well.
1 Like
I’m sorry about the emoji’s it’s because I forgot to remove it temporarily, well it’s working great for me but I will enhance and modify some of the function of the rules to make sure it will work super great
Perhaps try to explain simpler and clearer how you use it and what rules, steps etc go where. I did check out several of your posts and threads in the forum here but i didnt get where to put things and what should be inside, how not to confuse AI with certain parts,… i dont mind to try your approach, always good to learn something new.
Plan vs act is so simple, because there is not much to explain.
I modify the plan and act mode because when I am working at the large and complex and unorganized project the plan I created from previous conversation it’s forgetting what we have plan everytime, so I customize it with memories, lesson learned, scratchpad with plan and act mode but you can answer on the scratchpad file so that even you create a new composer session it remember it right
lets discuss this in your own thread 
I agree with memories, lessons etc. as concepts and storage of information. the implementation and usage is sometimes unclear or not simple on first try, thats the issue im facing.
1 Like
I’m sorry about the usage of implementation, it’s because I use the cursor to make me that readme file hehe
I liked the confidence percentage 
1 Like
Sir, very useful takeaways for me !
Can I ask you for one question?
Now, I got a new programming language based on c#,
I wanna know if cursor could learn this language through the markdown files which i packed them all and put them in .cursorrules so that cursor could use new programming language to make new project by game engine, such as unity?
helloo I have been playing with this in the last couple of days and wanted to ask you something. for new composer sessions in a relatively small codebase, it works quite well but I notice that none of the(memories, scratchpad and lessons learned) files are getting updated if you start a new session. Have you noticed this as well?
You need to manually do it like tell it to update after you do a lot to make sure you will capture all of the activities you do with AI, monitor it and double verify to make sure you are following the right path
I use some of the manual activities like, checkpoint restore if there’s a mistakes the AI do, I will go back to that specific chat and reload window and resync and wait for a sec and then submit the chat and then the AI will automatically back to normal like it will not hallucinate again, do these steps everytime when it got hallucinate and not reliable
interesting. I expected it to work better on new sessions since it already had the memory. I wonder if your rules could be tweaked to support this automatically. How can cursor understand this very well in the first iteration but totally ignores it in the next session 
And make sure to add all of the files of .cursorrules, memories.md, lesson.md, scratchpad.md and etc. to make sure it will read the rules always and the mdc files too, you will put it on context window so that it has a context
You should tag the @Recent changes and the last composer you have to make sure it has a context, or you can use the prompt and tag the @memories.md to. These steps that I do everytime to ensure the consistency of the AI
nice tips, thanks I will tinker with it more and come back to this thread if I can find a way to make it more automated.
1 Like
You should customize it for more automative
this is awesome @DaleLJefferson