• Everyday AI
  • Posts
  • Ep 670: Claude Skills: How to build Custom Agentic Abilities for beginners

Ep 670: Claude Skills: How to build Custom Agentic Abilities for beginners

OpenAI hires Slack CEO, Adobe and ChatGPT team up, Major Microsoft Copilot AI study and more.

Author

Everyday AI
December 10, 2025

Sup y’all! 👋

Claude skills are low-key a cheat code, yet so few people use them.

We tackled that issue on today’s show, and are LITERALLY giving away some goodies that we could probably sell for a pretty penny.

Like this LIVE Claude Skills directory and a custom-built GPT that helps you build better Claude Skills.

Yes, seriously. For free.

✌️

Jordan

Outsmart The Future

Today in Everyday AI
8 minute read

🎙 Daily Podcast Episode: Claude has one of the most powerful agentic features available today, yet very few people are using it. On today’s show, we give the beginner’s guide to Claude Skills. Find out more in today’s show and give it a watch/listen.

🕵️‍♂️ Fresh Finds: ChatGPT 5.2 launch updates, OpenAI new photo model reportedly in LM arena, Nano Banana 2 Flash release and more Read on for Fresh Finds.

đź—ž Byte Sized Daily AI News: OpenAI hires Slack CEO, Adobe and ChatGPT team up, Major Microsoft Copilot AI study and more.  Read on for Byte Sized News.

🧠 Learn & Leveraging AI: Most teams still use LLMs like chatbots. Power users are building persistent agents. Today, we break down how Claude Skills actually work. Keep reading for that!

↩️ Don’t miss out: Claude hints at front end Agents, Trump Approves Nvidia GPU sales to China, Meta’s new model in 2026 and more. Check it here!

Ep 670: Claude Skills: How to build Custom Agentic Abilities for beginners

Capabilities? Through the roof.

Usage? Ground floor.

Claude Agent Skills might be one of the most useful features of any front-end LLM.

Yet....it's crickets in terms of chat around it.

For this 'AI at Work on Wednesday' episode, we're breaking it down for beginners and will have you spinning up your own Claude Agent Skills in no time.

Also on the pod today:

• Claude skills: hidden power? 🤫
• Building agentic abilities, step-by-step 🛠️
• Claude vs GPTs vs Gems ⚡

 It’ll be worth your 44 minutes:

Listen on our site:

Click to listen

Subscribe and listen on your favorite podcast platform

Listen on:

Here’s our favorite AI finds from across the web:

New AI Tool Spotlight – Incredible AI Creates AI agents that automate your daily repetitive tasks, Vybe Secures internal apps. Built by AI in seconds. Powered by your data, Yolk analyzes sales calls, spots skill gaps, and fixes them with personalized AI roleplays so teams close more deals

AI Cancer Tool — Microsoft’s new GigaTIME tool turns routine cancer slides into detailed tumor “neighborhood” maps in seconds. It could expose hidden patterns that change how we predict and treat cancer.

Nano Banana Flash — Google quietly lines up a cheaper twin to Gemini Nano Banana 2 Pro

OpenAI Image Model Testing — OpenAI’s LM Arena models quietly challenge Google’s visual AI lead

AI Coding — Google’s new AI agent quietly evolves faster algorithms for real-world optimization.

ChatGPT 5.2 Rumors — Notion leak hints GPT‑5.2 may launch sooner than expected. But also, kinda later.

1. Adobe plugs Photoshop and Acrobat directly into ChatGPT ✨

Adobe is rolling out new Photoshop, Acrobat and Adobe Express apps inside ChatGPT today, letting people create and edit images, PDFs and designs without hopping between programs. Users can call up an Adobe tool by name in the chat, then keep refining images or documents with plain-language prompts as the system offers tweakable options like sliders for brightness or contrast.

While these mini-apps do not match the full power of Adobe’s desktop software, they still cover popular tasks like selective photo edits, PDF conversion and compression, and on-the-fly design changes for things like social graphics.

2. War Department rolls out GenAI.mil with Gemini for Government 🇺🇸

The War Department has just launched its new GenAI.mil platform, headlined by Google Cloud's Gemini for Government, in a bid to rapidly push generative AI across the U.S. military workforce. Framed as a response to President Donald Trump’s July mandate for AI superiority, the rollout means powerful AI tools are now reaching desktops across the Pentagon and bases worldwide.

Officials say the move is about creating an "AI-first" culture that hardwires AI into daily military planning and operations, backed by free training and systems cleared for sensitive but unclassified work.

3. OpenAI Taps Former Slack CEO as New Revenue Chief 🤑

OpenAI has just named former Slack CEO Denise Dresser as its Chief Revenue Officer, a move that signals the company is gearing up for a more aggressive enterprise push as AI becomes standard in everyday work.

Dresser will oversee global revenue strategy and customer success, drawing on her experience scaling Slack and leading major sales organizations at Salesforce. The appointment comes as companies shift from dabbling in isolated AI pilots to relying on AI throughout their operations and core processes, with many workers reporting faster and higher quality output.

4. 37.5 Million AI Chats Reveal How People Really Use Copilot đź“–

A new Microsoft AI analysis of 37.5 million de-identified Copilot conversations in 2025 finds that health dominates mobile use, with people turning to the assistant around the clock for wellness-related help.

The data shows a weekly rhythm where programming surges on weekdays while gaming wins on weekends, a sharp February spike tied to Valentine’s Day and relationships, and a late-night rise in religion and philosophy compared with daytime travel planning. Researchers also report a notable shift from pure information search toward more personal, advice-driven conversations, suggesting users increasingly treat Copilot like a trusted guide in daily life.

5. Big Tech Unites To Standardize AI Agents 👨‍💻

The group is centering its early work on AGENTS.md, a simple Markdown file format that gives AI agents consistent project instructions so they can behave more predictably across different codebases and tools.

6. Nvidia Shrugs Off DeepSeek Smuggling Allegations 🤷‍♂️

Nvidia is pushing back hard today against a report that Chinese AI startup DeepSeek has been secretly using smuggled Blackwell chips, a high‑stakes claim given Washington’s export ban on Nvidia’s most advanced AI hardware to China. The company says it has seen no proof or tips of so‑called “phantom datacenters” used to sneak its chips into the country, while insisting it will investigate any credible lead.

The dispute lands as Nvidia sits at the center of the global AI race and rising U.S.–China tech tensions, with its GPUs viewed as critical infrastructure for training and running powerful AI models. The broader backdrop is getting even hotter after President Donald Trump recently allowed Nvidia’s H200 chips to be sold to “approved” Chinese customers in exchange for a cut of sales.

The smartest way to build enterprise-grade Claude agents might just involve forcing its biggest rivals to write the code.

Lolz. 

We exposed this cross-platform strategy on today's Everyday AI, and the implications for your Monday morning workflows are HUGE. 

Claude’s Agent Skills are a legit cheat code, yet so few people have actually used them. 

Why? 

Because most leaders are still treating front end LLMs like the chatbots they once were. 

They type a prompt, wait for a response, and repeat.

That’s manual labor.

While your team is clicking buttons in ChatGPT, smart operators are building "passive" agents in Claude that execute complex code without being asked.

We broke down exactly how not just the best ways to build Claude Agent Skills for beginners, but also unveiled how to use ChatGPT and Gemini for even BETTER Claude Skills. 

So on today’s edition of AI at Work on Wednesday, we simplify the Claude Skill ecosystem and are literally giving away the playbook. 

(More on that below.) 

Ready to dive in?

1. Stop Pasting Prompts Manually 🚀

You are likely wasting hours every week manually selecting custom GPTs or pasting instructions for recurring tasks.

Unlike a GPT or a Project where you have to manually navigate to a specific interface, Skills are "persistent procedural knowledge." They sit in the background and scan your conversation. When Claude detects a relevant trigger—like a request to analyze a spreadsheet or format a newsletter—it automatically invokes the specific training data required to do the job perfectly.

This isn't just automation. It is governance.

It allows you to enforce strict brand guidelines or financial reporting standards without relying on an employee to remember the right prompt. The architecture relies on a simple folder structure containing a skill.md file. Once uploaded, you stop "using" AI and start managing an agent that knows exactly what to do when work hits its desk.

Try This

Audit your team's workflows this week for "high-frequency, high-precision" tasks. These are the processes where you have a strict SOP but are currently pasting prompt libraries into a chat window manually.

Select one pilot workflow—like "Quarterly Financial Review" or "Client Email Triage."

Map out the exact steps, required data inputs, and the "trigger" language that should activate this mode. Stop treating AI as a tool you have to pick up and put down. Start building persistent skills that activate the moment work arrives.

2. Force Rivals To Code ⚡

Here is the competitive secret we uncovered that makes most AI developers uncomfortable.

The most efficient way to write high-performance Claude Skills is to ask ChatGPT to do it.

If you are a power user, ChatGPT or Gemini already holds your entire context history, memory, and communication preferences. We demonstrated a workflow where we leveraged ChatGPT’s deep knowledge of our formatting style to write the complex Markdown code required for a Claude Skill.

You could use Claude's internal "Skill Creator," but it starts from zero. By using a rival LLM that knows your business context, you generate a skill.md file that is 90% optimized before you even test it. This allows you to bypass the technical learning curve of writing YAML and Markdown from scratch while ensuring the output sounds exactly like you.

Try This

Open your primary LLM (ChatGPT or Gemini) where you have the most chat history.

Use this prompt: "Based on our history and my communication style, write a strictly formatted skill.md file for Claude that automates [insert task], ensuring it uses YAML front matter and clear trigger definitions."

Take the code block it generates, paste it into a simple text editor, and save it as skill.md inside a dedicated folder. You just used your competitor's intelligence to build your proprietary advantage.

3. The Testing Matrix Trap 🔥

But that is also their biggest liability.

Because you don't manually select them, a poorly defined skill will jump into conversations where it doesn't belong. We found that if your trigger definitions are too broad, a "Financial Analyst" skill might try to format your casual lunch order like a P&L statement.

This creates friction rather than speed.

You cannot just upload a zip file and pray. You need a testing matrix that defines exactly when the agent should wake up and, more importantly, when it should stay asleep. Precision in your skill.md file is the difference between a helpful agent and a broken workflow that frustrates your team.

Try This

Create a "Skill Test Matrix" before deploying this to your team.

Set up a grid with three columns: "Intended Trigger," "Similar but Unrelated Request," and "Completely Out of Scope." Run the skill against all three scenarios.

If your skill triggers during the "Similar but Unrelated" test, your definitions are too loose. Go back to the Markdown file and refine the description until the skill fires only when strategically necessary. That is the difference between a smart agent and a nuisance.

Reply

or to participate.