I try not to create dead codes.
AI is always driven by detailed specifications, rules, and roadmaps, so there is very little waste.
I think that I have produced so much code because I often produce large scale OS, browsers, JavascriptEngine, and so on.
I maintain server configs for linux, unix, sql and a lot of json’s to deal with api calls to providers as part of my codebase. Cursor is really helpful with my server management issues.
Having limitd experience with Rust myself, I can see totally your high usage in a Rust environment, cargo etc. Rust pulls in a lot. Much respect to you as a Rust Developer … 
Not much compared to those in the millions but I spend some decent time on cursor, mcp locally and other tools, alot of it isnt being done on the cursor side to save me tokens.
Since I use multiple copilot’s and for some reason my dashboard isn’t showing my full stats. I will just show my project stats for the month, doubt many if any will believe my actual stats but:
Now who will be the first to call lies? lulz
What models are you using Aqua ?
Who are you working for ???
I’m a solo dev, not working for anyone at the moment.
Dr. Evil
Wow you guys are doing a lot of work! How long does it take you to fix all those errors? When Cursor writes lines of code for me they rarely work, not to mention the LLM can rarely fix it. It just generates error prone faulty code, page after page that never actually does anything!
Tell it to make working code.
Tell it to not make any errors.
Write these in the rules file. You can tell it to add rules by itself for itself, in the general lines of these two requests. Maybe also include ULTRATHINK keyword for this set of rules.
I’m building with Rust, it’s hard to generate errors with Claude if the project is started by Claude (familiar patterning). But the errors that appear, it catches itself by running a check before finishing the task and checking the dry run output and logging, and it will not stop until the feature or change is working, all fixed by itself. There are rules that it first checks syntax, then compilation, then runtime, I just ask it for new things with simple messages.
I am jealous of the guys above doing million lines of code, I never get above 10k a day.
■■■■, I didnt know it was that simple! All I have to do is tell cursor not to make any errors, That’s genius level stuff right there! Thank for all you help, I didnt know there were rules either… Guess this is all my fault after all, Cursor is the best software out there now that I can just tell it, Don’t code in any errors - But really, Why should I even have to tell it that? It’s designed to produce error prone code to begin with? I’m confused, Why would Cursor design it so you have to explicitly tell it not to make errors in every prompt? I’m wondering from your experience, does it listen to you and actually design code with no errors? Are you sure your using CURSOR? What version is this that listens when you tell it not to make errors? Really doesn’t sound like Cursor to me
Also use emojis, but realistically this is what a markdown rule looks like for “don’t fake data”.
This is just for a small daily projects for a quick experiment, the big projects have 500+ lines of new rules, above those from Cursor, and you also have to load FEATURES and CHANGELOG and maybe even a well detailed plan for it to listen. It also logs changes to an unreleased version, and checks output after compiling/running, explaining in the documentation how to run it, what it does and sample output expected, after which it will push a version number and commit all the changes automatically before the final report.
This is not a Cursor issue.
You can see here a regular task using 900k tokens and outputting useless trash in Auto mode. It failed to self-report runtime errors and output failure but declared party mode and kept insisting it was successful.
This is a model issue.
Is it worth it? I would say so.
A weirdly excessive set of rules. The last one is almost impossible.
Man, I’m getting it today! Emoji’s make it better? Like this, Don’t make errors in my code! 
Does it reply with 
I have to tell “don’t fake data” also, Like this? Don’t fake data 
It’s not a Cursor issue means all LLM’s need theses rules?
Don’t fake Data
Don’t write errors in my code
Don’t do the same thing over and over again. 
We need to program this so it actually works! 
Stop trying to find the .env files that are hidden, They are hidden for a reason. Do not look at them, do not create new ones, do not edit them, do not copy them, do not move them, do not hide what your doing to my .env’s, do not lie to me, What else am I missing because it still does all that!
What are you working on, that you encounter more languages than in TIOBE? 
I have to write all this in 70+ Languages now? With Emojis?
About that…
First rule is about not using .env at all. Otherwise you won’t sleep well tonight if you keep fighting it about that.
JSON-based approach provides better structure and validation
Pydantic models ensure type safety and validation
Private directory approach keeps all sensitive data in one git-ignored location
Default fallbacks allow the application to run without configuration files
No hardcoded secrets anywhere in codebase
External credential management required
Private directory structure for sensitive data
Comprehensive input validation via Pydantic
Environment-based configuration with sensible defaults
Development environment setup via scripts/setup_dev.sh
Virtual environment support (standard Python practice)
Configurable logging levels and output formats
