Six months ago, "how to write good prompts" was the hottest topic in group chats. Now this question is clearly outdated. What has replaced prompts is Skills.
The most obvious tipping point for this shift was, of course, the emergence of OpenClaw.
Even if you could call it plagiarism, and it wasn't the original creator of the agent concept, it truly brought the concept of an agent into the mainstream view, closer to the AI you've seen in movies: possessing personality, able to remember things, capable of planning, and truly able to complete tasks for you, rather than just answering your questions.
In the past, when people used AI, they were essentially using a very smart search engine—you ask, it answers, and the next round starts fresh. The agent lengthens this thread. It actively pushes tasks forward, finds ways around obstacles, and after completing one step, it moves on to the next. The first time you see it actually handle a complete task, you get a strange feeling: this thing is really working for me.
Then people started thinking: how to make it more capable.
This is the real reason Skills have become popular. It's not because Skills themselves are particularly novel, but because the agent made people seriously consider this question for the first time. What Skills do is equip the agent with specialized capabilities.
Why are Skills so important now?
An agent without Skills is like a smart but completely uneducated newcomer. If you ask it to do financial analysis, it can think, but it's slow, prone to errors, and requires you to guide it step-by-step through many stages. Skills are equivalent to it having pre-learned the complete workflow of that field—it can get started immediately without you having to repeatedly correct it.
The most widely shared Skills in the community currently focus on a few areas: workflow automation, stringing together operations that originally required jumping between multiple tools into a chain the agent can run through on its own; injection of rules for professional fields, ensuring the agent doesn't improvise freely when performing tasks requiring high precision, like in law, medicine, or finance; personalized adaptation, tuning the agent to your most efficient way of working, remembering your preferences, language style, and judgment criteria; and, of course, a category of Skills related to money, such as trading.
Arbitrage opportunities on Polymarket are something the average person can't decipher from the order book, nor do they have the time to watch the trends and calculate price differences. But an agent equipped with specialized Skills can: monitor in real-time, identify discrepancies, judge whether to enter the market, running the entire suite without you needing any background knowledge in predicting markets.
The same goes for quantitative trading. In the past, this was the domain of investment banks and hedge funds, requiring writing strategy code, connecting APIs, and watching backtesting data. Now, people have packaged the entire process into Skills; an agent can install them and start executing strategies on exchanges. The barrier to entry has shifted from "knowing how to program and understanding finance" to "knowing how to install Skills".
This change isn't about making people lazier; it's about pushing the boundaries of capability outward.
Behind these needs lies a common logic: people are starting to seriously treat agents as long-term collaborators, not just tools to be closed after use.
So, what novel ideas do you have that you want to turn into a skill for your agent?
Before, you had an idea, you spotted a market gap, but couldn't implement it. You didn't know how to code, didn't have time to learn, hiring外包 was expensive and slow, and eventually that idea just rotted in your notes. Now it's different. Using vibe coding, you can directly shape your idea into a Skill—no need to make a webpage, no need to make an app, no server required, no maintenance team needed.
The underlying logic of this is: agents will be a necessity for everyone. The Skills you make don't need to find their own users; they naturally run on the agent that everyone is already using. The market is there, the channel is there, you just need to build the thing that nobody else has made yet.
Before, there was a technical team standing between "I have a good idea" and "I have a working product". Now that distance has been compressed to a weekend.
