This is great! Just need something about maintaining context. Something like always review all of the interactions to date before replying.
The latest update is god like compared to the ones before. Just letting you know
Find the link to the latest version in my original post.
I think the latest version is better at keeping context. I simplified the entire prompt a lot and removed unnecessary instructions to make it a lot more clear.
I also added āYOLO MODEā, to keep true to Cursor Agent settings. If this mode is enabled, the agent will rarely stop and ask you if it should proceed or not.
As a default, the agent will ask you if you want to enable this mode or not, right after youāve submitted the prompt.
Thank you
Thank you for sharing this. I notice that as humans, we can only work efficiently for so long without losing context our self. Having your prompt helps greatly when our efficiency drops and we start going off tasks our self. Your prompt helps the agent be a co-pilot when our human brain goes on auto-pilot.
Thank you very much for your kind words!
And I agree with what you say, the prompt was designed with helping with:
- If the agent looses focus of what to do, you can just point it to the task file youāre working on to refresh its memory.
- If you want to pick up a task later on, you just do the same thing.
- If you want to have a weekly review of what youāve done, or similar, you can just flip through the files in the .task directory.
I just added YOLO mode if you want to try. Although Iād rather be involved in the loop, some of you might just want full automation.
Iām sorry I havenāt responded to this post. Iāve been busy with a lot of things. I will take a look at it when I get the time. Thank you for sharing though!
Hi @maxfahl,
Thanks for the share, it looks like an amazing prompt template !
Since Iām new in the Cursor and AI game, I donāt find out how to use this prompt. Iāve saved this prompt in a .md file to refer to it and then when I copy/past the content into Cursorās composer (agent mode) itās marked as a reference and I canāt edit it inside composer.
Should I edit it before copy/pasting it in composer ?
For reference, Iāve found out how to paste without marking content as a reference :
Instead of pasting with Ctrl+V or Cmd+V, use Shift + Ctrl + V (Windows/Linux) or Shift + Cmd + V (Mac)
The idea is that you copy/paste the entire prompt to a temp file or something, then fill in the details at the top:
[TASK]: <DESCRIBE YOUR TASK>
[PROJECT OVERVIEW]: <ENTER PROJECT OVERVIEW, OR LINK TO FILE CONTAINING THE DETAILS>
[MAIN BRANCH]: <YOUR MAIN BRANCH>
[YOLO MODE]: <ask|on|off>
When that is done, you just copy/paste it into the prompt box in Cursor. Remember to paste with SHIFT+CMD+ALT+V (mac, on windows, maybe ctrl instead of cmd?), it will then psate it as text into the prompt box instead of as an attachment (or whatever its called). I my opinion, this should be the default for content that is not copied from a source file.
(I just realized you found this out by yourself)
You reminded me that my initial prompt was actually missing instructions on how to add your task and project overview etc. to the prompt before submitting it, so Iāve added it now thanks to you <3
I hope this can get you going.
It works like a charm !
By the way, Iāve made some custom tuning :
- removed the āOriginal Task Templateā section from the generated file as it causes some formatting issues and doesnāt look essential in the process.
- created an entry in Placeholder Definitions to populate the āOriginal Execution Protocolā section since the provided method also causes formatting issues :
- [EXECUTION_PROTOCOL]: >
**IMPORTANT**: The ENTIRE execution protocol (everything between "# Execution Protocol:" and the next "---")
must be copied verbatim and in full, including all steps, sub-steps, commands, and HALT points.
It should be wrapped in a markdown code block to preserve formatting.
NO SUMMARIZING OR ABBREVIATING IS ALLOWED.
Hi Maxfahl,
I just wanted to take a moment to thank you for the incredible effort youāve put into refining the Task Master Prompt. Your updatesāespecially the clearer placeholder definitions, streamlined initialization process, and the explicit step-by-step workflowāare truly impressive and make the entire approach much easier to follow and implement.
The thoughtfulness behind adding distinct checkpoints and improving the Git workflow really shows, and itās a huge help for structuring complex tasks efficiently. This level of detail and clarity is exactly what many of us needed.
Thank you again for sharing this and for being so open to feedback. Your work is incredibly valuable to the community!
Best regards,
Maybe I missed it, but it would be super helpful if you had an example prompt with example data.
@maxfahl would you agree to create a Github repo for this prompt so that everyone can contribute by opening issues or creating PRs ?
In the meantime, here is the link to my customized fork Original Execution Protocol
Thanks for chipping in. Itās kind of funny how fast it forgets about what itās doing sometimes. even when using this prompt, if the AI has to look through a lot of files before making the first analysis, it will sometimes fail to adhere to the steps in the prompt. Just feels a bit funny, and hopefully they will increase the context length limit soon.
Here you go. I got a feeling you were using an older version of the prompt by reading your comments. Please give this a try, and donāt hesitate to make any pull requests if you add or remove something to improve the performance. I would rather try and remove/streamline than to add. If you find something that works really well, please let me know.
Yes, I could add a small example for the placeholders. Thatās a good idea!
Okay, yeah, Iām not sure if I need to have this in the prompt. I still think itās valuable to have a copy of the execution protocol in there. I will experiment a bit with removing the Task File Template copy.
I took the liberty to implement your execution protocol instructions for the task file template
Great. Looking forward to seeing it.
Thank you very much, really appreciate that!