i genuinely can't understand why this isn't the first thing every claude user builds... anthropic quietly shipped a folder system where claude loads specialized instructions on its own based on what you ask. no prompt engineering after setup. almost nobody uses it here's how it actually works... you drop... a folder into your skills directory. inside: a SKILL.md that says "trigger this when the user asks about X". claude reads the description once at startup. next time you ask about X, it silently pulls in the whole skill - custom instructions, scripts, examples - like a specialist walking into the room 1. skill for your writing voice - loads only when you draft, never when you analyze 2. skill for your codebase conventions - loads only when you edit that repo, not others 3. skill for your accounting flow - loads only when you say invoice, receipt, or expense your context window stays empty until you need it. one claude behaves like 30 different specialists in one chat and you never manage the switching full breakdown in the article below i test things like this on myself and share the results. follow kai if you want to see what's nextshow more

This is an interesting analysis in this post Thank you

Majority have found cool words like “second brain, loops, harness” and thinks that it, just nobrainer download and that’s all. So thanks to this guide, hope majority will finally install everything and setup in proper way

this, most people collect skills like and never open half of them

most people would rather rewrite the same prompt than spend 10 minutes making a skill

The folder isn't the breakthrough - moving instruction-selection from the prompt into the runtime is. The agent decides what context to load instead of you hand-feeding it each turn. That's context engineering. #ContextEngineering @AnthropicAI

Because it’s limited use and dated approach to look at memory.

"Almost nobody uses it" is the real story here — the feature works exactly as you describe, the discovery problem is that most people never find out it exists:

this sounds like a game changer for claude users

having Claude dynamically load specialized instructions and custom scripts via a simple skill directory completely transforms how context is utilized

watched claude pick stale folder rules over current task, versioning matters more than setup

Benchmark scores rarely show what happens under real load. Tail latency, error rates, and throughput variance are where provider gaming or capacity issues actually surface. A setup like makes those swings visible across eval and production.

Skills are context routing with a maintenance cost. I would version the trigger description beside the skill and test near-miss prompts, because silent misfires are harder to spot than missing skills.

what happens when two skills could trigger at once?

AI 模型的 benchmark 越来越卷，但落地时发现最大的问题是场景理解。技术跑分和真实需求之间，隔着一条叫做'产品思维'的鸿沟。大多数 AI 创业公司，死在拿着锤子找钉子的路上。

This is crazy.

@grok summarise and make actionable

This is where AI starts feeling less like a chatbot and more like an actual operating system for your work.

how many skills are you running right now?

AI becomes much more useful when it knows which expertise to apply. 🚀

The description field is doing more work than the instructions. A skill with great instructions that never triggers is dead weight, and one with a loose description fires when it shouldn't and quietly poisons the context. I write the trigger conditions first now. The body is the easy part.

you’re always up to something cool 🐾 🤝 just gave you a follow :)

i know right, having context switch automatically based on the prompt saves so much time. folk still sleep on system prompt organization until they actually try it once

The reason almost nobody uses it isn't discoverability. It's that the setup cost lands before you know what your instructions should say. You have to have already noticed the pattern in your own work to write a folder for it. Most people are still at the stage where every task feels new.

which feature is this projects or something newer

the real unlock is skill for accounting flow — most people are still pasting the same "here's how I categorize expenses" prompt every single time and don't even realize it

It read the folder before you did

Skills feel like the missing layer between a general model and a truly personalized AI system.