Code 2.1 is here and the official change

log is packed with updates. In this

video, I'll focus only on what actually

matters, grouped in a way that's easy to

remember and apply. Let's start. First

of all, let's update Cloud Code. In

order to know the version you're

currently using, you can run cloud

version.

I already have version 2.1.4, so I don't

need to update. But if you need to

update, just run cloud update.

And that's it. Let's start with my

favorite category, sub agents. Let's

read the update first. Headed hook

support to agent front matter, allowing

agents to define pre-tool use, post tool

use, and stop hooks scoped to the agents

life cycle. What are hooks? Hooks are

automated commands that run at specific

moments during agent execution. The most

common hooks are pre-tool use and post

tool use that was mentioned in the

update. Before this update, hooks were

global. They applied to all the agents

you had and the main session equally.

Let's see two examples and it will be

clear. First of all, let's say we have

an agent for security scanner. We have

two pre-tool use and post tool use. The

post tool use is the relevant. What

we're going to do is we're going to see

if the tool name is addit. If we are

editing, we're going to check the diff

in the file and make sure we don't have

any mentions for password, secret or

key. And this operation happens every

time we are running this sub agent. The

second example is about performance. We

have a performance optimizer sub agent

and we configure two hooks. The pre-tool

use hook just checks the current time

and the post tool use just take the end

time minus the start time and gives us

the duration of the execution. These are

two simple examples but you need to keep

in mind that you have this option. By

the way, I forgot to introduce myself. I

help developers turn AI into real

workflows. So sub and like it really

helps me provide more value for you.

Let's continue with the general settings

category. Read the new update. Added

plan command shortcut to enable plan

mode directly from the prompt. So we

have a new plan command. If I'm not in

plan mode, I can just type plan and it

will change my plan to plan mode

directly as you can see here. But more

importantly, the plan mode is important

when you already have a plan and you

want to see it again. And in that case,

you can just type plan and you will see

the original plan code tries to execute.

Two additional updates in that category.

The first one is added respect get

ignore support in setting.json for per

project control over mention file picker

behavior. And the second one is added

language setting to configure cloud's

response language. For example, you can

talk with Claude in English and it will

respond in Chinese or any language you

prefer. Both of them can be configured

in the VS code level. And if you are not

using VS Code or any of its forks like

cursor, you can use the setting JSON in

the load folder. This is how it looks.

You can configure the language and you

can configure the respect git ignore.

Why this is useful? Let's see in VS

code. Let's say I want to mention a

specific file. When I write anything, I

can see a list here. The option to

exclude from g ignore means that you

will have a shorter list and it will be

easier for you to navigate through the

files. If you want to configure it in

your VS code, this is how it looks. I'm

going to the setting.json in VS Code

folder. And this is the respect git

ignore and the language config. If

you're still here, you're serious about

coding with AI, we launched a free

school community with extra videos and a

free tutorial called five ways to

maximize your token usage. The link is

in the description. The next category is

related to skills. We have two updates.

Let's start with the first one. Headed

support for running skills and/comands

in fork sub agents context using context

fork in skilled front matter. In my

opinion, this is the most exciting

update in this version and I'll tell you

why. As time goes on, it's clear that

context engineering is one of the most

important factors for getting good

results for any AI coding agent. Before

this update, if you wanted to keep a

task focused, you had to run a skill or

the slash command through a sub agent.

Just to keep the context clean and avoid

polluting the main conversation, but now

you can simply add this attribute and it

will run in isolation without the need

for a dedicated sub agent. For example,

this is how it looks in my design skill.

In the skill.md file, I added this

attribute. The next update is relevant

to how we load skills. If you ever

created or updated a skill, you know

that you had to restart your session

with cloud code. Now they added

automatic skill hot reload. It means

that skills that created or modified on

your computer or in the specific project

are now immediately available without

restarting the session. This is kind of

useful. Cloud code keeps improving, but

your results depend on how you use it.

Check out one of these workflow videos.