Claude Skills: how you can stop re-explaining everything to AI.
You'll walk away with everything you need: what Skills are, how they work, when to use them — plus a Meeting Notes Analyzer you can build yourself in 30 minutes or download ready-to-use
You’ve surely heard about MCP over the last year and about Skills for the past 2 weeks, both concepts by Anthropic, both can be used to increase your AI capabilities. MCP was still a pretty “dev required” thing—you need an API key at least. Turned out that Anthropic realized that MCP doesn’t really fulfill user’s needs and came up with the Skills concept. And like MCP, it can be used with any AI; I tested it in Cursor. So you don’t really need to lock into Claude—think about it as an instructions template that is highly consumable for AIs.
In this issue I want to show you what Skills are at their core, best practices and limitations to consider, and how to find, build, customize and manage skills.
What’s Inside
The Matrix moment — Why Skills changes everything you know about AI automation
Token economics — How to cut costs 80% while fixing hallucinations
Skills vs MCP — Why Anthropic quietly abandoned their own protocol
30-minute build — Your first Skill that actually works (Meeting Notes → Action Items)
Security blindspots — What malicious Skills can do + protection protocol
The Cursor edge — Why some teams get 3x better results
Early adopter notes — Current limitations & workarounds that work
What are skills from the first principles
Remember how Trinity in The Matrix downloaded a helicopter piloting Skill? That’s pretty much it. If your Agent (can be a general agent like ChatGPT or specialized like Claude Code) has a skill, whenever you set a task that can be solved with that skill it will find, download and use it on demand.
If you have repetitive tasks or prompts that you store somewhere and paste every time you need to perform a specific task, Skills is the solution. Instead of explaining your process every single time you need something done, you create a Skill once – and Claude remembers it forever. Need a weekly status report? There’s a Skill for that. Want consistent brand colors in every presentation? Create a Skill. Have a specific way you analyze customer feedback? Package it as a Skill.
A Skill is a folder containing:
A SKILL.md file - This is your instruction manual. It tells Claude what this Skill does, when to use it, and exactly how to do it. You write this in plain text using simple formatting.
Optional extras - You can add template files, reference documents, example outputs, or even small scripts if you’re comfortable with that.
At its core, it has the idea of progressive disclosure (like in The Matrix) meaning that at every single moment it uploads only the necessary amount of information (=tokens), which saves you from two main problems:
→ Hallucinations and mistakes because it doesn’t contain huge unrelated context
→ High costs because it doesn’t upload all context, only what’s necessary
Here’s how it works in practice. When you start a conversation with Claude, it sees a short description of each available Skill (like “Creates weekly team reports with wins, blockers, and priorities”). These descriptions are tiny – they barely use any of Claude’s “memory” for the conversation.
When you ask for something that matches a Skill description, Claude thinks “Oh, this person needs a weekly team report!” and then – and only then – it opens up that Skill folder and reads the full instructions. It’s like having a filing cabinet where Claude can instantly grab the right procedure when needed.
What makes Skills different from other features:
Custom instructions - Those brief, always-active settings apply to everything. Skills are detailed procedures that activate only when relevant.
Projects - These are workspace-specific and accumulate context over time. Skills are portable recipes you can use anywhere.
MCP (Model Context Protocol) - This connects Claude to external data sources like your CRM or Google Drive. The main problem with MCP (Anthropic just admited it in this article is token consuptions – it was the main reason to shift paradigm from MCP or projects (also called Agents in Claude Code) to Skills.
Use Cases
You read this far — here’s a small thank-you. If you want to keep reading (this post + every past post), your first month is $5.20, that’s 60% off. Feel free to cancel in the first 30 days and you pay nothing more.
The $5 link is open until June 1st. After that, it goes back to full price.