Recently I found this git: https://github.com/oraios/serena Basically its efficient code navigation. If you are making websites or writing in java I think this may help you.

Claude Fable 5 might just be one of those launches people may look back on later and say, “That was a shift.” Anthropic is calling it a Mythos-class model made safe for general use, and the message is pretty clear: this is not just a nicer chatbot. It is built for bigger, harder, longer work — coding, research, vision, knowledge work, and complex problem-solving. And apparently, the more complicated the task gets, the more Fable 5 pulls ahead. Also, Anthropic is launching it with conservative safeguards, and some higher-risk requests will be routed to Claude Opus 4.8 instead. Think: bio-weapons, cyber-security. They are also launching Claude Mythos 5 for trusted cyberdefenders and infrastructure teams, which tells you a lot about where AI is heading. What is this going to change for your projects?

Vibe coding is powerful, but shipping vibe-coded apps directly to production is dangerous. The problem is not AI writing code. The problem is people launching apps without proper testing, security review, and real validation. A web app can look finished, work in the browser, and still have serious issues: broken authentication, exposed APIs, unsafe user permissions, bad input validation, leaked secrets, or logic flaws. That is why every AI-built app should be treated like untrusted code until proven safe. Before going live, test it. Run manual QA. Check user roles and permissions. Test edge cases. Scan for security issues. Try to break your own app before real users or attackers do. Tools like Shannon by Keygraph are a good example of where this is going: AI-assisted pentesting that analyzes web apps/APIs and tries to prove real vulnerabilities before production. Vibe coding is not the problem. Blindly trusting the output is the problem. Here is the tool: https://github.com/KeygraphHQ/shannon

I found 3 interesting GitHub repos for Claude / Claude Code that might be useful if you’re building websites, UI, or frontend projects with AI. The idea is pretty simple: Instead of just telling Claude “make this better”, you give it a proper skill — basically a small set of rules, taste, and guidance for how to think about layout, spacing, typography, animations, and UI polish. Here are the 3 repos: 1. Impeccable https://github.com/pbakaus/impeccable This one feels like the most complete of the three. It helps Claude review and improve frontend work — things like layout, typography, spacing, colors, responsive issues, UX details, and overall polish. I think this is especially useful when you already have a page or component built, but it still feels a bit unfinished or “AI-generated”. 2. Taste Skill https://github.com/leonxlnx/taste-skill This one is more about helping AI stop making generic SaaS-looking designs. It pushes Claude toward better visual taste: better composition, better spacing, better motion, better style, and stronger design direction. I like this one because it’s not just technical. It tries to help the AI make better design decisions. 3. Emil Kowalski Skill https://github.com/emilkowalski/skill This one is more focused on design engineering and the small UI details that make a product feel polished. Think animations, micro-interactions, transitions, smooth UI behavior, and that “premium” feeling. In my opinion, this is best when the design is already decent, but you want to make it feel more refined. My takeaway: I wouldn’t install everything at once. I’d start with Taste Skill or Impeccable, test it on a real project, and only keep what actually improves your workflow. AI coding tools get much better when we don’t just give them prompts, but also give them structure, examples, and clear design rules.

Greettings everyone. Pepe here from Santiago, Chile. I've followed the basics of folder structure and it really makes sense. But i've got some issues. How the system records what i do? Could i do something in order to get memory? I've watched some advanced videos about making an OS System and i get that that woould be overkill on my current state: i'm just becoming an AI Freelancer that wants to make it. In order to do that, i would like to have a content lab and maybe some way in with the whole system records my actions. My first impression is that every folder should record his own, like in order to make content, the draft folder maybe resolves this issue. Is a memory layer to get what we do everyday overkill? Does anyone uses something like this? How can this be done respecting the folder structure? If you have an idea how to make this. I'd really apreciate some advice. Thanks a lot!