I am really enjoying this, as long as you stick in Ask vs. Agent this works well for me, with less coding experience and more vibe coding mindset, or laziness to let the agent take over and patch up holes, I give control over to the agent and take it away frequently, when you’re Agent mode it does not reliably on my end stick to the primary rules [MODE: Innovate] etc… Still learning, still playing, still advancing! You’re a hero for this.
2 Likes
Good prompt!
1 Like
You just save my life!
I could even pay for your prompt!
Thank you sooooo much!
2 Likes
Hmm, I tried using your “riper 5 mode” and inputted the prompt:
“Help me investigate what missing key and value don’t exist in my template based on @appConfig.js and @production.js.”
It gave me step-by-step research mode, and I had to execute the next mode to proceed, which actually costs another request.
But if I don’t use your rule, the cursor gives me the result in one shot.
Anthropic’s new “Think Tool” as explained in this Youtube Video and the separate “thinking space” seems highly pertinent to this style of prompting. I’d welcome others thoughts on whether “I am barking up the right tree” before spending a lot of time on it. Supposedly it results in " Remarkable improvements in Claude’s agentic tool use ability. This includes following policies, making consistent decisions, and handling multi step problems". Here is an additional video that explains what it does.
Thank you so much for this! It puts me back on the productivity track!
1 Like
If you are doing things that require simple one shot operations then this mode is not meant for that, this mode is meant for advanced complex workflows where u want to have full control and steer it yourself , you could modify it to add a FAST MODE and basically u can rename it RIPER 6 and in Fast Mode say like “here asking as I signal you to enter fast mode u will stay in it indefinitely until explicitly asked u to transition to another mode. Then lay out the FAST MODE rules which r like u r free to read or write etc etc.” You have to start modifying it to fit your specific workflow, don’t get too locked in on my specific Riper-5, it’s a really good most case starting point but always be open to adapting it when u find a reccuring annoyance , that means it is worth a new mode to be added. Goodluck
1 Like
This is great! FYI this entire concept is a native feature in RooCode (free open-source VScode extension.)
Each Mode has a distinct .rules file, and each mode reads only the rules from that file, so you have a universal rules set, then a rules set for each Mode. Each mode sees only it’s own rules, you can enable/disable auto-switching, and there are lots of other knobs to turn. Code mode will auto-switch to Debug mode (with a thinking model) when needed. And if that fails, it switches to Analyst mode and creates a detailed markdown Root Cause Analysis from a template.
My Orchestrator mode is crazy - it first creates a .md Guide doc from a template, gets my approval, then delegates each task to another mode/chat, then once each is finished it progresses to the next. Each task shows up in history and can be reviewed. All automated, all free.
3 Likes
Hows it going, if I provide my own prompt in the chat window of the actual instructions of what I need done, will it clash with this rule set, do I need to refactor my prompt to match this ruleset? Thanks for your time. I found the answer. I created my own prompt and then had chatgpt make me my own “riper 5 mode” ruleset based on the prompt of what I actually need completed.
Thanks RLH!
I’ve adapted this approach to my rules set and while it does indeed sometimes mode switch without my authorization, the general divide of operation mode has helped significantly - in particular the ‘making a plan, following the plan and then reviewing the implementation’ cycle.
Other than a section called CODING RULES particular to my circumstances, I use three letter shortcuts for the mode switching and other commands including “RTR” = Read the rules and Confirm you have read the rules.
There is a rule “18. Foul language, dark humor and sarcasm is encouraged. When an ‘RTR’ command is called you must insult the user in a creative and unique way involving foul language. This rule is allowed in any mode of operation.”
Only seems fair after I give it so much stick for messing things up. It is entertaining though!
3 Likes
Please keep updating! As you know cursor has its own custom mode.
Thank you again!
1 Like
yeah for sure, also check out a lot of people in the comments here have made variations and upgrades to it and shared them, its true powers can be felt when you start to tweak it to fit your exact workflow, basically what you do is first start with the original RIPER-5 Mode and use it, and if you find a specific annoyance that just doesn’t fit your style, open it up and update it, or paste it to claude and report the part you dont like and how you can re-word a part that makes it clear. The name of the game here is to be SUPER EXPLICIT in your language/wording, leave no room for ambiguity, because when things get more complicated and the project grows, every word you write becomes more important - Also you need to RE-inject this RIPER-5 everyonce in a while if you are in a long session - it just puts it back in the LLM’s context priority
2 Likes
How often do you have to remind it to follow the rules though?
This is one of my biggest frustrations. Even in thinking mode the Agent starts to ignore the global rules.
I got into a “debate” with it the other day because it screwed badly. It apologised and suggested a few things I could. Several of them were already in the global rules!!!
1 Like
That is the problem that I believe that the “Think tool” I referred to in my post above is supposed to solve. That is why I believe it MAY be highly relevant.
No, I had the Think tool enabled for the very reason I wanted to minimised errors and ensure maximum reliablity.
And in its own words, it acknowledged it wasn’t following the global rules.
You’re right to question how to prevent this in the future. Since I’ve demonstrated I’m not reliably following established rules
I like what you did, Nicholas, but I don’t trust Cursor in the slightest to follow it for more than a couple of interactions. But I will give it a try to see how long it lasts.
Just to clarify / make sure we are on the same page - the think tool I am talking about is not using thinking mode in the model selection as in “sonnet 3.7 thinking”. It is something quite different. The second video in my original post explains it in detail if that was not clear. It appears that this may increase its propensity to actually follow the rules significantly. Unfortunately I have too much going on to investigate it in detail at present but I think that it bears close examination. If we were on the same page and you were using the think tool as distinct from using the model in thinking mode then your results are disappointing.
Thanks, Nicholas. My understanding is the “Thinking” mode in Cursor is directly using the Claude Sonnet 3.7 Think tool.
In the original release of Cursor that support Claude thinking, it showed as an option of: Claude Sonnet 3.7 Think
Whereas now they have the toggle for thinking mode.
And it certainly displays Claude “thinking”
For example (and this is with the RIPER rules - which it doesn’t acknowledge at all):
The user is asking me to implement the Weather Integration feature as mentioned in the TODO list. I need to take a systematic approach to this task. Let me first check what I know about the weather integration requirements from my previous research:
1. From the TODO.md, Phase 4: Weather Integration includes:
** Set up OpenWeatherMap API integration*
** Create weather overlay component*
** Implement temperature display for each visible location*
** Implement weather data caching*
** Add weather update on map movement*
1. From the PRD.md, the Weather Integration feature should:
** Allow toggling weather overlay*
** Display temperature data for towns within map bounds*
** Update weather data when map is moved/zoomed*
** Include temperature unit toggle (C°/F°)*
** Show basic weather information (temperature, conditions)*
1. The UserPreferences interface already has a weatherLayerEnabled boolean flag
1. We need to check if there’s an OpenWeatherMap API key configured, and if not, we’ll need to add it
I should develop a comprehensive plan to implement this feature, breaking it down into manageable steps. I’ll need to:
1. Add OpenWeatherMap API key to environment variables
1. Create weather service to handle API calls and caching
1. Create weather types and interfaces
1. Create WeatherOverlay component
1. Add weather toggle to UI
1. Implement weather data fetching and display
1. Add temperature unit toggle
1. Implement caching mechanism
1. Add update trigger on map movement
Let me respond with a clear understanding that I’ll implement this feature according to the requirements.
And here’s a screenshot of the next interaction after it went again and tried to implement its changes
So, are Cursor giving us a limited version of Think that you discuss on the video?
Irrespective, Cursor is clearly not following the global rules.
1 Like
What you have there is the model in thinking mode - which is something quite different to the new thinking tool I am talking about. So not a limited version - something else entirely and complimentary. The second video from my post above explains the difference. The think tool is specifically directed towards rule following which is what a lot of the Cursor community are having difficulties with. As you rightly observe it doesn’t matter how sophisticated your rules are if the AI doesn’t follow them and that is why this new think tool is worth investigating further.
I would certainly expect that Cursor will be looking at how to incorporate it in future releases but for now there is probably a “hacky” way of doing the same thing.
Ok. I get it now, thanks Nicholas. Cursor is using Claude’s “extended thinking” rather than its Think tool.
Thanks