08:38Jay E | RoboNuggets · Demo
Someone Rebuilt Adobe for Free, and Claude Can Run All of It
ArtCraft cloned seven Adobe apps from scratch in Rust, open-sourced the code, and built them so Claude Code can drive every one.
October 8thThe argument in one line.
Claude Code's new mod system lets small, event-driven add-ons change what the tool looks like and does without writing a plugin from scratch, and nine early community and Anthropic-built mods already cover progress tracking, cost control, and safety guardrails.
Read if. Skip if.
READ IF YOU ARE…
- You already use Claude Code daily and want to customize its interface or add guardrails without learning its internals.
- You run multi-agent or long Claude Code sessions and want visibility into sub-agent cost and progress.
- You want a safety net against accidental destructive commands like rm -rf during an agentic coding session.
SKIP IF…
- You don't use Claude Code, or an equivalent agentic coding tool, at all. The whole video is Claude Code interface configuration.
- You're on the VS Code extension only. Mods weren't available there as of this recording.
The full version, fast.
Anthropic shipped 'mods' for Claude Code on October 1: small event-driven add-ons that can watch, change, or handle what happens during a session, changing the actual interface rather than just following instructions like a skill or plugging in an external tool like an MCP connector. Jay demos nine: Savvy Progress (sub-agent cost and progress panel), Claude Skins (reskins the tool-call UI), You Should Know (flags costs or risks Claude might bury in a wall of text), File Tree (live file-change panel), Cache Tax (shows the dollar cost of a cold prompt cache before you send), Blast Radius (confirms before destructive commands like rm -rf), Reflect (turns 'no, do it this way' feedback into a saved CLAUDE.md rule), Terminal Browser (embeds a real browser beside the terminal), and Replay Theater (steps through every edit Claude just made). Any mod can be built by describing it to Claude, which has a built-in plugin-authoring skill; save one you like as a permanent plugin, or it disappears when the session ends.
Free for members
Chat with this breakdown — free.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →Where the time goes.
00:00 – 00:39
01 · Intro
Cold open: Anthropic shipped Claude Code mods on October 1, and Jay previews the nine worth installing.
00:39 – 02:02
02 · What mods are
Defines a mod against a skill (instructions) and an MCP connector (external plug): a mod runs inside Claude Code and changes what's on screen.
02:02 – 03:02
03 · Mod 1: Savvy Progress
A progress bar and live sub-agent panel showing model, token count, and running cost for every helper Claude spawns, demoed on a cold-brew launch-kit task.
03:02 – 03:57
04 · How a mod works
Every Claude Code action fires an event; a mod watches, changes, or handles that event, the mechanism behind all nine mods.
03:57 – 04:34
05 · Mod 2: Claude Skins
Reskins tool-call icons and edit colors with any theme or palette; display-only, so it's the safest first mod to try.
04:34 – 06:05
06 · Mod 3: You Should Know
A second agent flags important details Claude might bury in a wall of text, like a Kling clip's real per-generation cost; mid-segment plug for Jay's RoboNuggets community.
06:05 – 07:04
07 · Mod 4: File Tree
A live file panel that shimmers while Claude edits a file and turns green on save, demoed on a codebase and on a folder of tax-receipt PDFs.
07:04 – 08:15
08 · Mod 5: Cache Tax
Shows the dollar cost of sending now on a cold prompt cache ($0.52) versus waiting for a warm turn ($0.03) before you hit send, plus a keep-warm ping option.
08:15 – 09:04
09 · Mod 6: Blast Radius
Anthropic's own guardrail mod: intercepts destructive commands like rm -rf, lists every file that would be deleted, and requires a manual Proceed click.
09:04 – 09:39
10 · Mod 7: Reflect
Detects correction language ('no', 'I wish we'd done it this way') and offers to save it as a permanent CLAUDE.md rule on the spot.
09:39 – 10:09
11 · Mod 8: Terminal Browser
Embeds a real web browser beside the terminal chat so built pages and GitHub pull requests can be reviewed without leaving Claude Code.
10:09 – 10:43
12 · Mod 9: Replay Theater
Anthropic's sample mod: a post-turn replay button that steps through every file edit Claude just made, for code or prose.
10:43 – 11:56
13 · Build your own mod
How to make a custom mod by describing it to Claude's built-in plugin-authoring skill, turn on hot reloading, and save it as a permanent plugin.
11:56 – 12:47
14 · Wrap up
Closing thoughts on software becoming more customizable, plus the free PDF guide and subscribe ask.
Lines worth screenshotting.
- Claude Code 'mods' shipped October 1 and change the actual interface Claude Code renders, not just what it does: a skill is a handbook and an MCP connector is a plug, but a mod runs inside Claude Code and redraws the screen.
- A mod works by listening for Claude Code's internal events and choosing one of three responses: watch it, change it, or handle it entirely itself.
- Savvy Progress adds a cost and progress panel for every sub-agent Claude spawns, including which model it used and roughly what it cost in API credits.
- You Should Know runs a second agent that reads Claude's own output and surfaces anything important buried in a wall of text, like a $0.45-per-clip cost Claude would otherwise mention only in passing.
- Cache Tax shows the dollar cost difference between sending a message that reloads a cold prompt cache ($0.52) versus waiting for a warm turn ($0.03) before you hit send.
- A Claude subscription's prompt cache typically goes cold after about an hour of inactivity, and the next message after that reloads the entire conversation at full price.
- Blast Radius, built by Anthropic, intercepts destructive commands like rm -rf build and lists every file that would be deleted before a human has to click Proceed.
- Reflect watches for correction language like 'no' or 'I wish we'd done it this way' and offers to save that correction as a permanent rule in CLAUDE.md so Claude doesn't repeat the mistake.
- Mods are session-only by default. A mod Claude builds for you disappears when the session ends unless you explicitly ask Claude to save it as an installed plugin.
- Claude Code already ships a built-in plugin-authoring skill, so building a custom mod is just describing what you want in plain language.
- As of this recording, mods work in the Claude Code terminal and the desktop app's Code tab, but not yet in the VS Code extension.
Nine small add-ons that change what Claude Code looks like and does
WHAT TO LEARN
Claude Code's new mod system lets you bolt on a cost tracker, a destructive-command guardrail, or a feedback-to-rule pipeline just by describing it, and none of the nine examples here require writing a traditional plugin from scratch.
02What mods are
- A mod is different from a skill or an MCP connector because it runs inside Claude Code and can change the actual interface, not just the instructions Claude follows or the data it can reach.
- Mods went live October 1 and work in the Claude Code terminal and the desktop app's Code tab, not yet in the VS Code extension as of this recording.
03Mod 1: Savvy Progress
- Savvy Progress adds a visible progress bar plus a live panel showing every sub-agent Claude spawns, including its model and running token cost.
- Treat sub-agent visibility as a cost-control habit: knowing which helper is burning tokens before a task finishes beats finding out from the final bill.
04How a mod works
- Every Claude Code action fires an internal event, and a mod is code that watches, changes, or fully handles that event, the entire mechanism behind every mod in this list.
- Knowing the watch/change/handle model means you can predict what any new mod actually does before you install it, just from its description.
05Mod 2: Claude Skins
- Claude Skins reskins tool-call icons and edit colors using any theme or color palette you give it, shipping with seven presets like Noir and Tokyo Night.
- It's a safe first mod to try because it only touches the display, never the behavior, so there's nothing to break.
06Mod 3: You Should Know
- You Should Know runs a second agent that reads Claude's own output and flags anything important that might get buried in a wall of text, like an unexpected per-clip generation cost.
- This mod costs extra tokens because it runs a second agent in the background, so factor that in before enabling it by default on every session.
07Mod 4: File Tree
- File Tree is free to run because it's deterministic code, not an LLM call, and it shows a live file panel that shimmers while Claude is touching a file and turns green once the change is saved.
- The same file-visibility trick works outside of code: the video's tax-filing example shows Claude working through PDF receipts with the same live file-touch indicator.
08Mod 5: Cache Tax
- Cache Tax estimates the dollar cost of your next message before you send it, showing the gap between a cold-cache reload ($0.52) and a warm turn ($0.03) in the demo.
- A Claude prompt cache typically goes cold after about an hour idle; a keep-warm option in this mod pings the session to hold the cache open while you're away.
09Mod 6: Blast Radius
- Blast Radius, built by Anthropic itself, intercepts destructive commands and lists every file that would be deleted before you have to manually click Proceed.
- This exists because even frontier models occasionally issue a destructive command that wipes a project, so treat a guardrail mod as cheap insurance on any agentic coding setup.
10Mod 7: Reflect
- Reflect listens for correction language like 'no' or 'I wish we'd done it this way' and offers to save that correction as a permanent rule in CLAUDE.md on the spot.
- Saving corrections as rules the moment you give feedback beats manually editing CLAUDE.md later, since the rule gets captured in the exact words you used.
11Mod 8: Terminal Browser
- Terminal Browser opens a real web browser beside the chat inside the terminal, so you can review a built page or a GitHub pull request without leaving Claude Code.
- It targets people who work mainly in the terminal rather than an IDE, removing the context-switch of tabbing out to a separate browser window.
12Mod 9: Replay Theater
- Replay Theater adds a replay button after each turn that steps through every file edit Claude just made, one change at a time.
- It works for prose edits as well as code: an article draft gets the same step-through diff view as a source file.
13Build your own mod
- Building a mod is just describing what you want to Claude, which already has a built-in plugin-authoring skill for this, no separate framework to learn.
- A mod is temporary by default and disappears when the session ends; explicitly ask Claude to save it as an installed plugin to keep it permanently.
- Turning on hot reloading lets a mod you're actively building show up in the same session the moment Claude finishes it, instead of requiring a restart.
Terms worth knowing.
- Mod (Claude Code)
- A small add-on, introduced October 1, that runs inside Claude Code and changes what the interface looks like or does by listening to Claude Code's internal events.
- Skill (Claude Code)
- A set of instructions that tells Claude how to perform a specific task, like a handbook. It changes behavior, not the interface.
- MCP connector
- Model Context Protocol connector, a plug that connects Claude to an external application or data source.
- Prompt cache
- Claude's short-term memory of a conversation that avoids re-reading the full history on every message. It goes cold after roughly an hour of inactivity.
- Hot reloading (mods)
- A session setting that lets a newly built mod appear immediately in the current Claude Code session instead of requiring a restart.
Things they pointed at.
Lines you could clip.
00:00
“So Entropic just shipped one of the biggest Cloud Code updates we've ever had, and they're calling it mods.”
07:40
“What cash tax does is stop you before a message that would cost you to reload your entire conversation and show you an estimate of what that would cost in tokens and API credits so you know exactly how much you'll spend before you hit send.”
08:17
“I've seen so many edge cases already where even a smart frontier model, like even Opus or Fable, would wipe out someone's entire project.”
11:32
“A mod that Cloud builds for you is temporary by default. So when you find one that you like, ask Cloud to save it as a plugin and install it, and it will be there every time that you open Cloud Code.”
Word for word.
Read-along
Don't just watch it. Burn it in.
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
metaphorstory
So Entropic just shipped one of the biggest Cloud Code updates we've ever had, and they're calling it mods. The idea is simple. If you can describe a feature that you wish Cloud Code had, it can now build that feature into itself.
So Cloud Code can become almost any tool that you need. And in this video, I'll walk you through the nine mods that are genuinely worth installing and tweaking. so that by the end of this, you can master cloud mods and make full use of them to make your systems smarter and faster.
And if you're new, my name's Jay. I spent over a decade working with brands you probably know, have been in AI since my master's in data science, and now I'm leading our AI business in one of the largest AI communities globally. But first, what are mods?
So a mod is just a small add -on that basically changes how Cloud Code looks and how it works. It came out on the 1st of October, and if you've updated Cloud Code to the latest version, then it's already switched on for you and ready to use. It works in the Cloud Code terminal and in the Code tab of the desktop app.
But at the time I'm recording this, it doesn't seem to be available yet in the VS Code extension. So make sure you're using it inside Cloud Code. the terminal or the desktop app now you might be thinking if you already have skills and you already have mcp connectors then how are mods different well a skill basically just tells cloud how to do a job it's right and a connector just but a mod runs inside cloud code and actually changes it so it can change exactly what you see on screen so if you want cloud code to have extra buttons a side panel or even a mini game to play while cloud is working that's all entirely within the realm of possibilities.
And making a mod is really easy because all you need to do is to ask Claude. I'll show you a few examples so that by the end of this video, you'll have a clear idea of what mods to use or tweak for your setup. And every one of these, by the way, I put into this PDF guide, which you can just send that straight to Claude and you can have your agent grab these or customize them to your liking.
And it's just a matter of asking your agent to set them up for you. So the link for that is just in the description below if you need it. But for now, here are the ones I tried, I have liked, and think will be useful for you as well.
Number one is Savvy Progress. Now, I really like this one because it's the right mix of being useful while also being quite fun and delightful. Now what it does is it puts a progress bar above your prompt so you can see how close Claude is to finishing your task.
But for heavier, longer tasks where a team of sub -agents is doing the work, it also shows you a live panel of every sub -agent. If that's new to you, sub -agents are basically these helper Claudes that your main agent can spawn, and each one of them basically takes a piece of the larger job. And for each helper, you can see its model, how far along it is, and roughly what it's costing you in API credits.
Now that's already useful by itself, but the even better part is that you can actually customize how these little pixel cloud crabs look. So if you want one that looks like a designer, you can do that. If you want one that looks like a developer, you can do that too, just by asking cloud to customize their looks.
So apart from giving you more information about your work, especially long running tasks that are done better with helper agents, it just makes using sub -agents a much nicer experience with this view. Now, you might be curious, how does a mod actually change what's on your screen? Well, every time Cloud does something like running a command, for example, Cloud Code actually sends out a little signal called an event.
And a mod is just a small piece of code that listens for those events so that when one comes through, the mod can do one of three things. It can watch that event, it can change it, or it can handle it itself. Savvy Progress actually uses all three.
So it watches every time Claude starts a helper. And every time that helper asks the model to do something, it notes the model, the tokens, and roughly what it costs. And it changes your view so that it draws the area above your prompt.
It adds its progress bar and also shows you that side panel if you're using subagent. And when you open the agents panel, it draws that whole panel by itself, including all your custom Claude mascot crabs. And the good news is you don't need to know any of this to make one mod.
Claude Code actually comes with a built -in skill now for building them so you just describe what you want and cloud builds it out for you Number two is Cloud Skins.
Now, this one is really simple, but it's probably one of the very first mods that you can try without really changing your workflow. Because what it does is reskin how Cloud Code shows its work. So your tool calls gets their own icons.
Your edits can show up in a specific color if you want. Right now, this open source repo comes with seven themes like Noir and Tokyo Night. But because this is Cloud Code, you can give it pretty much any design system or color palette that you want, and you can just have it build it out.
So if you're looking for a great first mod that is quite safe to start with, then this is a good candidate because it only changes the display and you can customize it to fit your setup. Number three is You Should Know.
This mod is actually built by Antrovik themselves, and it runs a second agent on the side that reads what Claude writes and flags anything important that you might miss. Now, that's pretty crucial, especially when Claude gives you a wall of text as an output, which more often than we'd like to admit, we don't actually read in full.
In my testing, I asked Claude Co to generate a video with a model like Kling, and it flagged that I should probably know how much that would cost if I ran it through KeyAI, which is the provider that I use. If that was buried in a wall of text, I might not have seen it. But with this little pop -up, it lifted the important point right out.
And you can see what it looks out for, by the way. So if you're more worried about costs, customer issues, or anything your automations might miss, you can tweak this mod to surface that. Do note that this particular mod, it does run a second age, so it can increase your token usage.
But for a lot of mods that I'll mention here, most of those are actually run using deterministic code, so they don't really require you to use an LLM in order to run them. And by the way, if you want to learn how to build and sell AI systems that businesses actually pay for, then that's pretty much all we do over at the Robonuggets community, where not only do you get access to the Cloud Living Masterclass, which we update every week and takes you from zero to mastery with the latest on AI, but you also get access to our agents as a service course, which walks you through how to actually get paid for all these AI skills that you are learning.
You also get to be part of a genuinely great community of AI builders. In fact, you can see just some of the recent wins our members are getting from the program right here. So if you want to start earning from AI, then check that just in the pinned comment below.
Now back to the video. Number four is File Tree. So this is one example where it doesn't really cost any tokens to run because what it does is just give Cloud Code a really nice file panel on the right that shows exactly which files Cloud is working on, which matters most, especially when you're working on real software.
Because when you're working in production code, it's important to know exactly what Cloud is reading, editing, or changing. With this mod, a file shimmers in the tree when Cloud is touching it, and when the change is committed, meaning it's saved, it turns green. So at a glance, you always know exactly which files Claude is working on.
And if you're not a developer, it's still useful whenever you're working across lots of files. To give one instance, let's say you're preparing a tax filing with multiple documents. With this mod, you can see exactly which ones Cloud has opened and which ones it has changed.
And it gives you a much nicer UI instead of relying on a chat reply that you always need to read every time in order to get the gist of what your agent just did. It's much more interactive and, in my view, much easier on the eye. Number five is cash tax.
Now, this is a mod that can actually save you money. Because if you don't know what the prompt cache is, it's basically this short -term memory that Claude keeps during your conversation. And it keeps that short -term memory so that it doesn't have to reread the whole conversation that you have with it every time that you send the prompt.
So it's there to stop your tokens from being burned. But after a while, usually about an hour on a normal Claude subscription, it goes cold. And once that prompt cache is called, the next prompt that you send actually makes Claude reread your entire conversation, which can really eat into your usage limits.
So what cash tax does is stop you before a message that would cost you to reload your entire conversation and show you an estimate of what that would cost in tokens and API credits so you know exactly how much you'll spend before you hit send. That really matters in long sessions because the bigger or longer your conversation is, the more that that one message would cost.
And if you're stepping away, you can actually type slash keep warm as part of this mod, or you can just have the mod turn it on for you by default. And what it does is it will ping the session with a tiny message right before the cache goes cold. So it's still warm when you get back.
Number six is blast radius. Now, this one, again, Anthropic made themselves, and it might just save your entire project. Because when Cloud is about to run something risky, like deleting a whole folder, it stops first.
It shows you exactly which files would get deleted. It shows you even a Proceed button and a Cancel button that you have to press yourself. Now, you might not think that your agent would ever delete your files, but I've seen so many edge cases already where even a smart frontier model, like even Opus or Fable, would wipe out someone's entire project.
There are some real horror stories out there, and if you don't want to become one of them, then you can try out this mod or something like it. So with Blast Rages, these kinds of commands get stopped automatically, and an actual human has to be in the loop to press Proceed before any of the files get deleted. And with Blast Rages, it actually shows you which of those files would be gone the moment that you do that.
Number seven is Reflect. Now, this mod is a bit like reinforcement learning from human feedback, but on a more micro scale and personalized to you. Because when you give Claude a piece of feedback with trigger words like no, or this doesn't look good, or I wish we'd done it this way.
What happens with this mod is that a small band would pop up above your prompt asking if you want to save that as a permanent rule in your Claude .md file or your skill file. And what you do is just click save and Claude writes that rule in for you. It's a more granular way to choose what Claude reflects on and to make sure the things you you've already corrected on are fixed or done better the next time that you use it.
Number eight is the terminal browser. Now, this one is for people who use Cloud Code in the terminal because what it does is it opens a real web browser right beside your chat. So if Cloud builds you a web page, you can look at it and review it without ever leaving the terminal.
Of course, it can also open GitHub pull requests the same way. So you save the time and the headspace of jumping between Cloud and a browser to check every change to your code or to your design. Everything now stays in one place.
So if that's a feature you've been missing because you work mainly in the terminal, then this mod may be for you. Number nine is Replay Theater. Now, this last one is another of Entropic's own sample mods, and it's called Replay Theater.
And what it does is after Claude finishes a turn, a replay button actually shows up. And when you press it, it steps through every file edit that Claude just made. So you get a rundown of what it did and can understand the changes to your workspace.
And again, even though this was originally created by Entropic for code, it's actually not limited to just that use case. If you're writing an article, for example, it can let you replay exactly what cloud changed in your draft. So if it made major changes, you can pinpoint exactly where they are in that article.
And the great thing about this is that mods are entirely modifiable to your liking. And the way you do that is really easy. First, you just need to make sure that cloud code is up to date because mods need this version 2 .1 point.
287 or newer. In the terminal, you can just type cloud update or in the desktop app, there will be a pop -up that shows you if you're on the latest version. Then just open Cloud Code, describe the modification that you want.
And Cloud Code comes with a built -in skill for this. It's called plug -in authoring. So Cloud already knows how mods are put together.
It'll start building it and then it'll ask if you want to turn on hot reloading for a session. What that means is it will just allow that session to sort of reload or restart itself so that when you turn that on, your mod would show up in that session. as soon as Claude finishes building it.
And if you don't like something or you want to tweak a feature, just tell it what to change the way you talk to a normal person. One thing to know is that a mod that Cloud builds for you is temporary by default. So when you find one that you like, ask Cloud to save it as a plugin and install it, and it will be there every time that you open Cloud Code.
Now, by default, mods don't really cost anything extra, but you, of course, do need a paid plan because that includes Cloud Code. And also, any mod that needs to call on a model in the background uses your Cloud usage like everything else. So there you go.
That's all nine of them. As usual, I hope that was useful. I think mods are a really good move by Entropic, and I hope OpenAI follows suit.
Because I think the era of more malleable software is definitely coming soon. And the fact that Entropic, one of the biggest and most influential companies right now, is opening up their own application to modification is a good step towards software that can be hyper -customized to our needs. I hope that gave you an idea of the mods you could build for your own workspace.
And again, the PDF guide is just down below. If you need it, send that to your cloud code to pick out the ones that are useful for your setup or to customize. As usual, thanks for watching till the end.
And if it was useful, then consider tapping subscribe down below because it helps me a lot to put out more educational stuff like this. I'll see you all next time. Cheers.
The Hook
The bait, then the rug-pull.
Nine days after Anthropic shipped mods for Claude Code, RoboNuggets creator Jay E had tested enough of them to rank the ones worth installing, from a sub-agent cost tracker to a one-click guardrail against an accidental rm -rf.
Named ideas worth stealing.
03:02model
Watch / Change / Handle
- Watch
- Change
- Handle
The three things a mod can do when it receives a Claude Code event: watch it, change it, or handle it entirely itself.
Steal forUnderstanding any Claude Code mod or plugin architecture before building one.
How they asked for the click.
VERBAL ASK
05:47product
“if you want to learn how to build and sell AI systems that businesses actually pay for, then that's pretty much all we do over at the Robonuggets community... check that just in the pinned comment below”
Mid-roll pitch for his own paid RoboNuggets community and course, delivered as a value-add aside rather than a hard sell, with an explicit return cue ('Now back to the video') after.
MENTIONED ON CAMERA
FROM THE DESCRIPTION
PRIMARY CTAWhere the creator wants you to go next.
AFFILIATECommission earned if you click.
OTHER LINKSAlso linked in the description.
- Get RUBRIC - The Command Centre for AI Agents ↗
- Savvy Progress ↗
- Claude Skins ↗
- Filetree ↗
- Cache Tax ↗
- Blast Radius (Anthropic's sample) ↗
- Reflect ↗
- Terminal Browser ↗
- Replay Theater (Anthropic's sample) ↗
- 🥚 Free trial of n8n ↗
- 🥚 Free trial of Make ↗
- If you learned something new and want to see more, support the channel by subscribing ↗
- ➗ Instagram ↗
- ➖ Tiktok ↗
- ✖️ Twitter/X ↗
- ➕ LinkedIn ↗
- For business, reach out at ↗
Visual structure at a glance.
