I had an idea and I pitched it to Claude.ai first and then we decided to try out Bernie's capabilities of vibe coding. Now, I haven't used Claude Code with Bernie, I still had the $200 OpenAI sub so I told Bernie to use Codex for vibe coding. The idea was something simple and whimsical: let's make payday fun. My pain point was gathering all my subscription expenses and make something more fun out of it because well, we all hate paying bills. So we created a silly PRD of gamifying Bills and then when we get paid, a hero comes out and slays them. We called it Payday Kingdom lol. Here is the app so far. I don't think I plan to do anything big with it yet but sure as hell gives me content to share with likeminded people 🤝 Let me know what you think! https://payday-kingdom.vercel.app/ It doesn't need any auth or any type of data sharing. It's just a boardgame that you put numbers in it and trigger math. So no data is being collected at all.

I spent six months fighting the same problem every Claude Code session. Context bloat. You know the pattern. You have 50+ rules you want Claude to follow. System prompts, coding standards, project conventions, workflow patterns. So you dump everything into CLAUDE.md and hope for the best. Except now every single message burns tokens on rules that aren't relevant. Asking Claude to fix a typo? It still gets injected with your entire MCP server development guide. I built something to fix this. Let me introduce you to CARL, (the Context Augmentation Reinforcement Layer), designed to dynamically inject rules based on what you're actually doing. Say "fix this bug" and you get coding rules. Say "update project status" and you get project management rules. Say "just chatting" and you get either a command to discuss injected (if you set it up that way) or just the global rules (only if you have them active - they can be disabled). The architecture is simple. A manifest file defines domains with keyword triggers. A Python hook runs on every prompt, matches keywords, and injects only the relevant rules. An MCP server lets you create new domains or toggle them on the fly. That's it. Text files and a matching algorithm. The results surprised me. Token overhead dropped 50 to 80 percent depending on the session type. But more importantly, Claude started following the right rules more consistently because they weren't buried in noise. I had an earlier version that worked well enough for my own use. But after refining a keyword exclusion system, adding domain scoping, and building the MCP management layer, I realized this could help others dealing with the same context management problem. The whole system is file-based. No database. Git-friendly. Plain text you can read and edit. I'm thinking about packaging this for others who use Claude Code heavily. If context management is something you've struggled with, I'm curious what your approach has been.

So, every time I make a new project, do I need to create a New Directory Folder, or just /init everytime i want a new one in the same directory? And how will I know it doesnt messed with the existing one I created?

Just tested Claude Cowork on an inbox with 3,600 unread emails and it completely cleaned the whole thing in about 5 minutes 🤯 Not only did it clear the inbox, it also built a full label management system so every new email gets organized automatically. Most people are still manually sorting emails or paying for tools that barely help. Meanwhile AI can literally act like a personal inbox assistant if you connect the right tools. If you want to see exactly how I set this up step by step, watch the video attached!

What's up gang, I don't know how you guys handle newsletters, but the way I do it is: I use Claude Code > invoke this skill I made > Claude Code writes the newsletter with my voice and then gives me a HTML file > upload to mailchimp > mailchimp sends I like to make newsletters look more nerdy by using a HTML template I make instead of the ones provided to you that look generic. If you're interested in the newsletter maker skill you can have Claude Code examine this one and then tell it to replace the brand DNA with yours. The brand DNA I got it from Google Pomelli. Let me know if you have suggestions or a better way to do this. Cheers!