Claude Mods are a plug-in style way to modify Claude Code. According to a Chase AI video, you can have Claude automatically insert a check, a rewrite, or a backup step before, during, or after it does something. The rest of this article explains how that works through an illustrative scenario.

It is four o'clock on a Friday afternoon and Jie is rushing a small internal tool. She asks Claude Code to clear out the old files in a folder, then goes to pour a coffee. When she returns, the screen shows a line she never configured: "About to delete 41 files. Want to look at the list first?"

She pauses. She does not remember ever installing this.

(Jie is an illustrative fictional character created to explain the idea. She is not a real customer, and this scene is a scenario we wrote based on the video's content.)

What can Claude Mods change?

The night before, Jie had watched a YouTube video by Chase AI. Its title calls Claude Mods the biggest Claude Code upgrade since Skills, and it had roughly 210,000 views. The core claim of the video is that Claude Code can now be modded.

How? The video's description is that whenever Claude is about to do something, you can attach a mod that steps in before it happens, while it happens, or after it happens, and can even wrap the whole process. Taking the video's example of deleting a folder, there are roughly four things you could do. You can make Claude stop first and show you the list of files about to be deleted. You can change the action itself, so that instead of really deleting, the files are moved to the trash. You can have it produce a report afterward saying what was deleted and where it used to be. Or you can wrap the whole thing so that a backup is made before deletion and you can restore at any time.

The author stresses that this is only a demonstration of how one task can be altered, not the main point of his video, and the folder in the video was a project build folder. Still, the framework is clear: what Claude does is now something you have a chance to change.

How are Claude Mods different from Skills and hooks?

Jie had already used Skills, so her main question was how this differs from them.

According to the video, Skills need you to call them, while Mods behave more like hooks: once set up they run automatically until you remove them. You do not need to dig through a settings file to turn one off. You simply tell Claude to stop that mod. The video also mentions that Mods work through the plugin system, and that typing /plugins shows what is installed.

As for where it came from, the video says it is the official version of something that was called function hooks about a month earlier. It mentions that in early September someone named Boris posted on social media that they were working on ways to make Claude Code more customizable. The transcript passed through a translation chain, so only the first name can be made out, and the surname and affiliation cannot be confirmed; this article does not try to reconstruct them. At that time it was only a pull request on GitHub, and the video says it is now a feature open to everyone.

The video also includes a metaphor comparing this trend to something called "Deep Sea Harness," a direction where everything is a plug-in. That name comes from the translation chain, and we could not confirm what it refers to, so we make no claim about it here.

Why is a Claude Code cache countdown mod useful?

Jie assumed Mods were mostly for fancy interfaces. The video does show some fun examples. One person made a mod that, whenever Claude is busy with a task, connects to a Doom multiplayer server so you can play against others who are also waiting. Another person built a progress bar that shows project progress live.

What actually made her stop, though, was the cache countdown the video's author built himself.

His explanation goes like this. Every turn of a conversation, the whole context is sent again to Anthropic's servers. If the cache is still alive, he says you can save up to about 95 percent of the cost, but the cache only lasts one hour, and you often cannot tell how long it has been. So he built a mod that directly shows "cache has 43 minutes left," with a button next to it. The situation he described is when the cache is about to go cold and the context is nearly full and he has to step away for a while: pressing the button starts compacting the conversation. In the video his context window is 160K tokens.

These numbers are the video author's claims about his own environment, and we have not verified them independently. Please check Anthropic's official documentation for actual discounts and cache lifetimes. But Jie understood the point: she uses Claude Code every day, and nowhere does anything tell her whether the action she is about to take is expensive.

How do you build a mod that is only yours?

The method in the video is surprisingly simple. Instead of hunting for ready-made ones on GitHub, you first ask Claude to audit how you use it. The gist of the instruction is this: look at how I use Claude Code, read my last 30 sessions, find what I keep asking for, and then propose five mods that would solve those problems.

After the author ran this, he received five suggestions: a tool for checking mods, a cache timer, a restore module, a typo fixer, and something to help him track his projects. Then he only had to say "I like the first one, build it," and Claude created the plug-in itself, installed it, and asked him to reload.

Jie did the same last night. The top item on her list happened to be "confirm before deleting," because she really had slipped a few times and nearly wiped things she should not have.

When she got back to her desk that afternoon, Jie did not rush to confirm. She read through the list of 41 files from the top and noticed that a configuration file still in use was mixed in. She pressed cancel, and wrote one line in her notes: from now on, the question is not only what Claude can do, but where I want to take a look before it does it.

This video is sponsored, so know that before citing it

In the middle of the video, the author promotes his own paid course, Chase AI Plus, which includes advanced Claude Code courses, weekly updates, and his own AI operating system. This is a channel owner promoting his own product, not a neutral review. All the Mods information we cite comes from this author's demonstration and claims, so treat it as an introduction worth trying but needing your own verification, not as an official specification.

He also predicts at the end that within a few months an ecosystem will grow around Mods, just like the one around Skills today. That is a personal prediction, not something that has already happened.

FAQ

Q1: Are Claude Mods an official feature?

According to the video, they grew out of the earlier function hooks and are now open to everyone. For the actual name, available plans, and installation steps, please rely on Anthropic's official announcements.

Q2: How are Mods different from Skills?

By the video's account, Skills need you to call them actively, while Mods run automatically like hooks once set up. The video does not discuss whether the two can be used together.

Q3: Can I make one without programming experience?

The method shown in the video is to ask Claude to build and install it itself, and you only describe what you need. However, anything that runs automatically changes how Claude behaves, so we still recommend reading what it will do before installing it.

Q4: Are the savings figures in the video reliable?

They are the author's claims about his own environment, which this article cannot guarantee. Please verify against official pricing.

Source Statement

This article is a reorganized rewrite based on a YouTube video published by Chase AI on October 2, 2026. The video title is "Claude Mods Is The Biggest Claude Code Upgrade Since Skills," the channel is Chase AI, and the original link is https://www.youtube.com/watch?v=Rn4nmFRPe0s. The original transcript passed through multiple languages of translation, so some proper nouns and numbers may be distorted. We restored them where we could confirm them and marked what we could not. Jie in this article is an illustrative character, not a real customer case, and the scenario was written for explanation. This article is an information summary and does not constitute any purchasing or investment advice.

Further Reading

If this topic interests you, we have published related articles on this site: one on how to build reliable AI agent skills, one on what agent skills are and the open standard behind them in 2026, a Claude Code token-saving guide, and a piece on how to avoid hitting the Claude token limit.

Look Once Before You Act

What Jie took away was to look once before acting. If you also want to check your situation before calling a model, you can look at your balance and usage records on AI Token King. These records only cover model usage called through AI Token King. They are a separate matter from Claude Mods themselves, and they cannot represent Claude Code's internal billing. If you want to find out more, visit https://aitokenking.com.tw/home