SuperMiraclesTreasure

The new era of local dogs is coming. All decentralized products originally created by humans, and AI that allows people to have control, should all participate once.

Because the official Claude released a task, I created a cross-conversation backlog to manage the tasks, simple and straightforward.

clawdbot using Claude code for subscriptions also has a chance of getting banned. It's not directly calling the Claude code client.

Those who make good use of AI are actually more like good educators.
Even after finishing a lesson, you might not score 100, and open-book exams can still have mistakes.
If you know which question you got wrong, you'll check more carefully;
if you're told which formula to use, you'll naturally make fewer mistakes.
When you make mistakes in elementary school math, it's not because you don't know addition or subtraction, but because you're careless;
when you write wrong characters, it's not because you can't recognize the characters.
AI that helps with homework is like a college student,
b

My recent insight from using superpowers is to choose B, it's never wrong. The issues can be divided into a few types: if you want all of the above, prioritize all of the above. If you need to choose one among several changes, A and C are two extremes, and B is the answer you want.

I originally intended to seriously develop products, but now I realize that the most valuable takeaway from the product development process is understanding AI development, the collaborative paradigm of AI developed by multiple people, and the experience of choosing the right people to hire or partner with. Additionally, it's important to distinguish between AI trend-following KOLs and bloggers who genuinely have experience using AI tools. Anyway, having money is great; it accelerates my learning and experience far more than working alone on projects.

I originally wanted to seriously develop products, but now I realize that the more valuable gain during the product development process is understanding AI development, the AI collaboration paradigm with multiple people, and how to choose the right people to work with. Anyway, it's great to have money; this experience gained from working with others is much more valuable than doing things alone.

Today, I worked with Dev to troubleshoot a web crawler issue.
He spent quite a while checking whether the AI-generated code had problems.
I just organized the phenomena I observed:
Manually, you can scroll the page with the mouse, but the script can't scroll.
Taking this phenomenon directly to ask AI, the problem quickly became clear.
During the review, I realized that the real difference isn't about who is better at coding,
but about what the first step is when a problem occurs. The AI-first mindset is truly rare. I've had him use AI to write code for half a year, yet his first re

A small project every day. Created a simple task management and push rules.

Interviewed many interns, and a clear feeling is:
99% of people will say "AI can't write perfect code."
But the actual situation is often:
The capability of the Agent depends on the quality of the person's prompt / problem modeling.
If you don't believe from the start that AI can get things done,
then you are unlikely to write instructions that enable AI to accomplish the task.

Spent two days figuring out how to manage vibecoding development, which is to let AI manage everything within a markdown file. Avoid using any frameworks to eliminate the learning curve. When your markdown becomes too long and affects the context, let AI organize it into modular documents. When you modify a specific module's document, let AI review the corresponding module documentation.

Claude code skill now supports hot reloading. My personal method is to update the documentation using skill after each code modification, and check the related workflows. For example, if I write a crawler, the crawler, quantitative strategy, and factors are all under development. Skills will naturally update the documentation.

My super middleware has finally completed monitoring Telegram, detecting alpha at the millisecond level.

Why doesn't the Telegram bot have a feature to monitor other Telegram bots?

I wrote a script that allows opencode to send content via claudecode. If a banned account is due to user habits, then at least you can use this relay station during usage.