Anton Gladkov · 2025-11-03 · source ↗ · whole document (4)
Вот в таком ключе я например фристайлом ему загнал и получил рабочие 11 промптов
Вот в таком ключе я например фристайлом ему загнал и получил рабочие 11 промптов:
Hey! We need to improve our analytics with the following: 1) What you see on screenshot: For reference - this is our daily stats (collected manually) for Airvoice on 31 October: • Active domains: 60 • Connected accounts: 200 • Warmup score: ✅ 100% • Sending limits: 25 • Emails sent: 📤 4,050 • Capacity usage: 🟢 85% • Reply rate: 🟡 0.9% I need you to read Instantly API documentation and find out how to get such data precisely, and add this to the dataflow that we're already getting for our analytics. 2) We don't track open rates and click rates as it's bad for email accounts health, so it would be great if we would disable showing it, while we can keep getting those stats as there is anyways always 0 values. 3) Currently Bounces, Unsubs and Quality score displays 0 values, which is indeed not correct. I need you to investigate how to call it right, because currently I think we do the wrong query type or something like that. 4) Copy_top & Copy_worst doesn't show anything with responses that you can see on screenshot. Which means that something is failing, because obviously we have sent super many emails, and running quite a few campaigns, with unique copywriting for each of them. So you need to find out why it responds like this and fix it. Also, we don't need to have two commands for that, let's change it to one command /copy that outputs best three texts and worst three texts with their statistical data explained in a convenient manner.
Please make sure that previous commands are gone from /help menu and the new /copy command will make it there. 5) We need to double check the precision of the data that we display by all the /status commands that we have. My concern is about dates. We need to make sure that day is today, week is this week (from monday to the current date), month is this month (from 1 day of this month and till the current date), and we need to output also "last day", "last week", and "last month" data in the same query response so user can compare on the go and analyze better. Also we need to add the ability to ask for stats for particular date frame. Like /status 1.10.2025-31.10.2025 will give stats for every day of October inclusive. Also we need to be able to ask /status October. It will dramatically increase the comfort of interaction. Please ultrathink through not only local but also global changes and optimizations that we need to perform to keep the efficiency of our code on a highest level, but keep in mind that we want to stay one-file python bot project, and we're not going to go enterprise level software, this is only for our usage, and will not be significantly integrated with anything. I need you to provide me a set of comprehensive CODEX prompts that will implement all the changes, because CODEX agent is coding for us on this project. What codex can't do (for instance changes in Supabase) - give me with your precise instructions.
For DB management I need just a prompt for SQL editor that will implement changes automatically with a single prompt or series of prompts:) If my current suggestions looks logical to you, and you will invent other enhancements for the userflow of the similar nature that I did not mention, please add CODEX prompts for their implementation as well, with your description of why they matters. К этому я добавил 2 файла:.py и schema из supabase
Он в итоге около 17 минут думал и выдал мне 11 промптов
Когда промпт отрабатывает (остается от 40 до 97% контекста), я ВСЕГДА даю ему
› commit and gitpush so it gets deployed to railway via already established connection through github
И когда он отчитывается мол задеплоено -
/new
/approvals - full access (double enter)
То бишь меня контекстное окно в принципе не ебет ибо я не понимаю зачем пытаться в рамках одного окна два промпта реализовывать
Если их все равно делает внешний GPT5-pro =)))
Я не увидел никаких бенефитов от того чтобы в одном контекстном окне более одного промпта жарить. Минусов при этом увидел миллиард)