Claude Code and Codex Never Have To Build From Scratch Again
🔗 Link do vídeo: https://www.youtube.com/watch?v=b9k-AE4v5yI
🆔 ID do vídeo: b9k-AE4v5yI
📅 Publicado em: 2026-08-10T14:20:28Z
📺 Canal: AI LABS
⏱️ Duração (ISO): PT12M
⏱️ Duração formatada: 00:12:00
📊 Estatísticas:
– Views: 6.361
– Likes: 257
– Comentários: 8
🏷️ Tags:
Faster builds using Claude Code, with 2 skills that stop your agent rebuilding what already exists.
Start building with Lovable at https://lovable.link/hJppXFl and turn your next idea into a live app.
Building using Claude Code is the easy part now. These 2 Claude Code skills fix the hard part: your agent never checks what already exists. Real Claude Code tips, the Claude Code setup we run, and how to use Claude Code skills to build faster.
Community with All Resources: http://ailabspro.io
The Roundup, our daily newsletter covering the AI stories that matter. Join now: https://www.theroundup.so/
Skill 1: https://github.com/AaravKashyap12/advise-project-approach
Skill 2: https://github.com/UditAkhourii/neuroarxiv
Agents build fine now. The problem is they start from scratch every time and never check whether what you're asking for already exists, as a paid tool or an open source project on GitHub.
How to start using Claude Code skills
Paste the GitHub repo link into your agent and tell it to install the skill. It lands in your .claude/skills folder. We run Claude Code here, but both work in Codex.
Skill 1: advise-project-approach
– It works out where you are first. A brand new idea and a new feature on an existing product are different problems, and it handles them differently.
– If your idea is vague it asks questions before researching anything: who this is for, what it needs to do for them, and whether you can pay for tools or want to stay free. Whatever you don't answer it writes down as an assumption instead of quietly guessing.
– Then it finds the paid tools and the open source projects already doing your thing. If docs and GitHub aren't enough, it asks whether to go into X, Reddit and YouTube too.
– The GitHub part that changed how we download things: agents reach for whatever's most popular, and popular on GitHub means stars, a bookmark count that only ever goes up. It tells you people liked it once, not that it fits you. Big projects are big because a team and a lot of users are behind them, so the skill separates what applies to you from what's only there because of their scale.
– Then the money. Want DMs in your app? Stream and Sendbird built it. Sendbird's free plan is about 100 users, Stream starts around $500 a month for 10,000, and past that both charge per extra person who opens a chat. So it checks cost at three points: while you're building, the day real users turn up, and once it grows.
What it told us
We asked for a web app where we drop our skills so anyone who downloaded them gets updates automatically. If you came here for building a website using Claude Code, the honest answer this time was that we didn't need one: it pointed us at a tool that already existed, the skills CLI, plus GitHub for storage. It also caught that Claude Code's plugins self-update, but only for Claude Code, and we use Codex.
Skill 2: neuroarxiv
– The first skill finds what somebody already built. When the thing is experimental and nobody has shipped it, it stops. Ours was pushing files onto other people's laptops automatically, with no idea what security problems that causes.
– This one searches arXiv, where researchers publish their work free and public and others pull it apart. The paper behind the technology that made ChatGPT possible went up there years before there was a product.
– If you're looking at how to create agentic AI using Claude Code, this is a working example. Every paper gets its own sub-agent with a fresh context window, started by the main agent, and none can see what the others are reading. Then it commits to one answer.
– It pulled twelve papers, argued over which one applied to us, landed on one about package systems, implemented that technique and tested it.
What we ended up with
A CLI tool, which just means it runs in the terminal, where these agents already live, so they can use it directly. On our side it's two commands: point it at a skill, then publish it, and you can tell Claude Code to do that for you. Everyone else's agent pulls it down in the background.
We've also built a third skill combining the two, because the second is expensive to run, so the one time you actually need it is the time you don't bother. It runs the first on everything and only pulls in the second when it hits something genuinely new. That one's in AI Labs Pro, with the other resources from our videos.
Less a Claude Code tutorial than the two workflows we run before we build anything, so if you've been hunting for a Claude Code course on how to code using Claude, start here.
Hashtags:
#ai #claude #claudeCode #claudeAi #vibeCoding #claudeCodeTutorial #howToUseClaude #usingClaudeCode