🧠 New Blog Post

Google has dominated as a search engine for nearly two decades, but they are starting to lose their lead to AI chat tools. No one wants to dig through Google search results one at a time like a caveman anymore. We ask ChatGPT or Perplexity, who then asks Google, Reddit, Wikipedia, and other sources, then provides a comprehensive answer with links.

Instead of SEO, companies are starting to focus on AEO (Answer Engine Optimization) or GEO (Generative Engine Optimization). But how do you optimize your website and brand for this new AI SEO? And how would it be measured?

🤖 AI & Dev Tools

Anthropic just released a new Skills feature, allowing users to create and share custom skills that Claude can use across all tools (Desktop, web, CLI, API). They work like a mix between a system prompt and an MCP, giving Claude a programatic way to lookup context before performing specific tasks.

Skills are created by adding Markdown files with extra instructions, with a YML frontmatter description that can be quickly indexed instead of reading the whole document. This way the skill’s title and description can be scanned like tools in an MCP, without needing to perform retrieval on the full document unless it is needed. So it’s more than just a fancy system prompt. It actually reduces token usage, and lets Claude access extra instructions more like tools than a RAG source.

The next time you’re repeating a task or reusing a saved prompt, try creating a Claude Code Skill instead! Claude will look for mentions of the skill name in your prompt, and only reference it when needed.

💡Tips & Tricks

Speaking of creating Claude Code Skills: At first I was creating skills ‘as needed’ based on what I’m currently building. But then I realized Claude and ChatGPT can reference past threads, and should be able to give me a list of all the areas they have previously struggled on the types of projects I like to build.

𝐒𝐨 𝐈 𝐜𝐚𝐦𝐞 𝐮𝐩 𝐰𝐢𝐭𝐡 𝐭𝐡𝐢𝐬 𝐩𝐫𝐨𝐦𝐩𝐭:

𝘚𝘤𝘢𝘯 𝘰𝘶𝘳 𝘱𝘳𝘦𝘷𝘪𝘰𝘶𝘴 𝘤𝘰𝘥𝘪𝘯𝘨-𝘳𝘦𝘭𝘢𝘵𝘦𝘥 𝘤𝘩𝘢𝘵𝘴 𝘢𝘯𝘥 𝘱𝘳𝘰𝘥𝘶𝘤𝘦 𝘢 𝘭𝘪𝘴𝘵 𝘰𝘧 𝘱𝘳𝘰𝘣𝘭𝘦𝘮 𝘢𝘳𝘦𝘢𝘴 𝘸𝘩𝘦𝘳𝘦 𝘈𝘐 𝘳𝘦𝘱𝘦𝘢𝘵𝘦𝘥𝘭𝘺 𝘴𝘵𝘳𝘶𝘨𝘨𝘭𝘦𝘥 𝘵𝘰 𝘨𝘦𝘯𝘦𝘳𝘢𝘵𝘦 𝘤𝘰𝘳𝘳𝘦𝘤𝘵 𝘴𝘰𝘭𝘶𝘵𝘪𝘰𝘯𝘴 𝘰𝘯 𝘵𝘩𝘦 𝘧𝘪𝘳𝘴𝘵 𝘢𝘵𝘵𝘦𝘮𝘱𝘵. 𝘐𝘯𝘤𝘭𝘶𝘥𝘦 𝘰𝘯𝘭𝘺 𝘵𝘰𝘱𝘪𝘤𝘴 𝘵𝘩𝘢𝘵 𝘰𝘤𝘤𝘶𝘳𝘳𝘦𝘥 𝘪𝘯 𝘮𝘶𝘭𝘵𝘪𝘱𝘭𝘦 𝘤𝘰𝘯𝘷𝘦𝘳𝘴𝘢𝘵𝘪𝘰𝘯𝘴 𝘰𝘳 𝘸𝘩𝘦𝘳𝘦 𝘵𝘩𝘦 𝘮𝘰𝘥𝘦𝘭’𝘴 𝘳𝘦𝘴𝘱𝘰𝘯𝘴𝘦𝘴 𝘸𝘦𝘳𝘦 𝘰𝘶𝘵𝘥𝘢𝘵𝘦𝘥, 𝘪𝘯𝘤𝘰𝘮𝘱𝘭𝘦𝘵𝘦, 𝘰𝘳 𝘪𝘯𝘤𝘰𝘳𝘳𝘦𝘤𝘵 𝘥𝘶𝘦 𝘵𝘰 𝘧𝘳𝘢𝘮𝘦𝘸𝘰𝘳𝘬 𝘤𝘩𝘢𝘯𝘨𝘦𝘴, 𝘦𝘹𝘦𝘤𝘶𝘵𝘪𝘰𝘯-𝘦𝘯𝘷𝘪𝘳𝘰𝘯𝘮𝘦𝘯𝘵 𝘥𝘦𝘵𝘢𝘪𝘭𝘴, 𝘰𝘳 𝘪𝘯𝘵𝘦𝘨𝘳𝘢𝘵𝘪𝘰𝘯 𝘤𝘰𝘮𝘱𝘭𝘦𝘹𝘪𝘵𝘺.

𝑶𝒖𝒕𝒑𝒖𝒕 𝒇𝒐𝒓𝒎𝒂𝒕:
𝘛𝘰𝘱𝘪𝘤
𝘚𝘱𝘦𝘤𝘪𝘧𝘪𝘤 𝘧𝘢𝘪𝘭𝘶𝘳𝘦 𝘱𝘢𝘵𝘵𝘦𝘳𝘯𝘴 (𝘤𝘰𝘮𝘮𝘰𝘯 𝘪𝘯𝘤𝘰𝘳𝘳𝘦𝘤𝘵 𝘢𝘴𝘴𝘶𝘮𝘱𝘵𝘪𝘰𝘯𝘴)
𝘗𝘳𝘰𝘣𝘢𝘣𝘭𝘦 𝘤𝘢𝘶𝘴𝘦𝘴 (𝘦.𝘨., 𝘰𝘶𝘵𝘥𝘢𝘵𝘦𝘥 𝘵𝘳𝘢𝘪𝘯𝘪𝘯𝘨 𝘥𝘢𝘵𝘢, 𝘳𝘢𝘱𝘪𝘥 𝘦𝘤𝘰𝘴𝘺𝘴𝘵𝘦𝘮 𝘤𝘩𝘢𝘯𝘨𝘦𝘴, 𝘦𝘯𝘷𝘪𝘳𝘰𝘯𝘮𝘦𝘯𝘵 𝘮𝘪𝘴𝘮𝘢𝘵𝘤𝘩)

𝑬𝒙𝒄𝒍𝒖𝒅𝒆:
𝘖𝘯𝘦-𝘰𝘧𝘧 𝘮𝘪𝘴𝘵𝘢𝘬𝘦𝘴
𝘗𝘶𝘳𝘦𝘭𝘺 𝘶𝘴𝘦𝘳-𝘦𝘳𝘳𝘰𝘳 𝘪𝘴𝘴𝘶𝘦𝘴

𝑮𝒐𝒂𝒍:
𝘏𝘦𝘭𝘱 𝘮𝘦 𝘪𝘥𝘦𝘯𝘵𝘪𝘧𝘺 𝘢 𝘴𝘩𝘰𝘳𝘵𝘭𝘪𝘴𝘵 𝘰𝘧 𝘩𝘪𝘨𝘩-𝘷𝘢𝘭𝘶𝘦 𝘢𝘳𝘦𝘢𝘴 𝘵𝘰 𝘣𝘶𝘪𝘭𝘥 𝘤𝘶𝘴𝘵𝘰𝘮 𝘈𝘐 𝘴𝘬𝘪𝘭𝘭𝘴/𝘵𝘰𝘰𝘭𝘴 𝘢𝘳𝘰𝘶𝘯𝘥 𝘵𝘢𝘴𝘬𝘴 𝘸𝘩𝘦𝘳𝘦 𝘮𝘰𝘥𝘦𝘭𝘴 𝘩𝘢𝘷𝘦 𝘩𝘪𝘴𝘵𝘰𝘳𝘪𝘤𝘢𝘭𝘭𝘺 𝘶𝘯𝘥𝘦𝘳𝘱𝘦𝘳𝘧𝘰𝘳𝘮𝘦𝘥.

Try it in Claude, ChatGPT or any AI you use regularly that can access previous threads. You’ll get back a list of problem areas that can then be turned into new skills proactively, so they are ready the next time that same problem comes back up!

📺 Video Content

Retrieval-Augmented Generation is usually performed by fetching data from a vector storage, but it turns out graph databases can be even better for certain types of queries. Knowledge graphs enable LLMs to query records based on data about the relationships, in addition to the records themselves. This provides more accurate, repeatable and explainable LLM responses when working with structured data.

In this guide, I’ll show you how to run Neo4j and Appsmith locally, and build your own knowledge graph and GraphRAG pipeline. From there you can connect other datasources and build a custom UI with GraphRAG powered by Neo4j.

👥 Community Picks

About those Claude Code Skills though… How exactly do you create one? Did you know there’s a skill for that!? No, not the official skill-creator skill from Anthropic. This one is specifically made to crawl documentation websites!

I’m talking about Skill Seekers from Yusuf Karaaslan. It extracts, processes and summarizes documentation websites, and then creates the optimized *.md version for your new skill, along with the zip file to install it on web or Claude Desktop.

Shoutout to Yusuf for making this awesome Skill_Seekers tool open source!

📚 From the Archives

Whether you’re building a fullstack app from scratch, or using a no-code/low-code platform, at some point you’ll have to deal with users uploading massive images when high-resolution is not required for the use case. You can either accept the large files and process them server-side, and deal with the extra storage— or transform the images on the client-side and save some bandwidth and storage space.

In this guide, I show how to use native JS methods and the Dropzone library to build an image resizing and compression tool that works client-side, allowing users to adjust images before uploading. I’m using Appsmith for the platform and server, but the same frontend code could be used in a fullstack app to minimize storage requirements and reduce cost associated with your app.

Thanks for reading!

If you enjoy this newsletter, please share it with others in your network. Got a question or comment? Login at news.greenflux.us to share your thoughts.

Reply

or to participate

Keep Reading

No posts found