Learn Rust, Script-kit and Vscode

Visual Studio Code is an integrated development environment (IDE) built and maintained by Microsoft. It is offered free of charge.

vscode

We look at the various features of the VSCode Extension for Playwright and how it can help you with your testing workflow.

Playwright VSCode Extension

How often do you avoid scripting something because it takes too much effort?\n\nScript Kit makes it easy to create and run scripts that solve your daily problems. Create a new script from the prompt then your script opens in the editor of your choice. Write a few lines of JavaScript. Then run the script from the prompt.\n\nSimply put, Script Kit helps you script away the friction of your day.

script-kit

Create, read, update, delete. Four actions that every developer becomes intimately familiar with over their career. Script Kit aims to make this process as pleasant as possible using a `db` helper and what better way to show it off than a Todos app?

You will also learn how to create and switch between tabs through Script Kit's `onTab` and `setTab` methods. 

[Install todos](https://scriptkit.com/api/new?name=todos&url=https://gist.githubusercontent.com/johnlindquist/3bdc967ffbd2f6d72491d07bf720f7e2/raw/a37b0058b502d6ba4bebd7a31a30be8789f5a146/todos.js")

```js
// Name: Todos

import "@johnlindquist/kit"

let { todos, write } = await db({ todos: [] })

onTab("Todos", async () => {
  while (true) {
    let todo = await arg("Toggle todo:", todos)
    let t = _.find(todos, todo)
    t.tag = t.tag === "open" ? "done" : "open"
    await write()
  }
})

onTab("New", async () => {
  let name = await arg("New todo:")
  todos.push({ name, id: uuid(), tag: "open" })
  await write()
  setTab("Todos")
})

onTab("Remove", async () => {
  let todo = await arg("Remove todo:", todos)
  _.remove(todos, todo)
  await write()
  setTab("Todos")
})
```


Build a Multi-Tab Todos App in Script Kit

When you want to collect news sources, airline ticket prices, or any events from sites that don't offer APIs, you can use scrapers to grab elements from off the page. 

Script Kit includes a `scrapeSelector()` helper that takes the URL you want to scrape and the selector you want from the page. Using the `// Schedule` metadata, you can also have this script run in the background on a Chron schedule and collect the data for you.


[Install scrape-tech-news](https://scriptkit.com/api/new?name=scrape-tech-news&url=https://gist.githubusercontent.com/johnlindquist/acd397a2ed487ef2a06c00d2951bed63/raw/3942107d2c92abc2c0317d15b0b3d025c5643ef0/scrape-tech-news.js")

```js
// Name: Scrape Tech News
// Schedule: 0 11 * * *

import "@johnlindquist/kit"

let h3s = await scrapeSelector(
  "https://news.google.com/topics/CAAqJggKIiBDQkFTRWdvSUwyMHZNRGRqTVhZU0FtVnVHZ0pWVXlnQVAB?hl=en-US&gl=US&ceid=US%3Aen",
  "h3"
)

let filePath = home("tech.md")
await ensureFile(filePath)
let contents =
  `

## ${new Date()}

` + h3s.map(h3 => `### ${h3}`).join("\n")
await appendFile(filePath, contents)
```

Scrape a Website on a Schedule with Script Kit

We put a lot of work into browsing the internet every day, it would be a pity to lose it all by closing all of our tabs. 

Script Kit can collect all of the tabs you currently have open in Chrome using its `getTabs` method and store them anywhere you want. 

This particular script converts the tabs to markdown links and stores them in a text file on your file system, but you could convert and store them using any approach you can think of.


[Install save-tabs](https://scriptkit.com/api/new?name=save-tabs&url=https://gist.githubusercontent.com/johnlindquist/4713a161a9ec48932b05436ba6a1a123/raw/33f753586d1a158dfb69d07f29850b6013d59f8b/save-tabs.js")

```js
// Name: Save Tabs

import "@johnlindquist/kit"

let tabs = await getTabs()

let tabsMd = tabs
  .map(tab => `* [${tab.title || tab.url}](${tab.url})`)
  .join("\n")

let notes = await editor(tabsMd)

let filePath = home("tab-notes.md")
await ensureFile(filePath)
await appendFile(filePath, notes)

edit(filePath)
```


Save Chrome Tabs in a Markdown File and Take Notes with Script Kit

Sometimes you just want to select and modify some text you have written somewhere on your system. To help you with this, Script Kit provides a `getSelectedText` and `setSelectedText`. Between these two helpers, your script and use and/or modify the text however you want. 

Applying an npm tool like `title-case` can help when you're not entirely sure when certain words need to be capitalized. In fact, I used the title case script on the title of this lesson!

[Install title-case](https://scriptkit.com/api/new?name=title-case&url=https://gist.githubusercontent.com/johnlindquist/0f8323d6aa599cfd62034351f9d38e8c/raw/625d0afd57ccbf8bf64c62bd135474cecf98c58b/title-case.js")

```js
// Name: Title Case

import "@johnlindquist/kit"

let { titleCase } = await npm("title-case")

let selectedText = await getSelectedText()
let titleText = titleCase(selectedText)

await setSelectedText(titleText)

// This Is the Title of My Next Blog Post
```


Apply Title Case to Text Anywhere Using Script Kit

Developers often keep their dev projects in a specific directory like "dev", "projects", or something similar. Script Kit allows you quickly read all the files of any directory then display them in a list so you can open them in your editor.


[Install open-project](https://scriptkit.com/api/new?name=open-project&url=https://gist.githubusercontent.com/johnlindquist/91cf994e450ecd5d45bce9aef9b345a1/raw/880be2d827b9602afe59d4e230e6753288f5bbb8/open-project.js")

```js
// Name: Open Project

import "@johnlindquist/kit"

let dirs = await readdir(home("dev"))

let selectedDir = await arg(
  "Open Project:",
  dirs.map(dir => ({
    name: dir,
    description: home("dev", dir),
    value: home("dev", dir),
  }))
)

edit(selectedDir)
```

List and Launch Dev Projects with Script Kit

We all have various tasks we need to accomplish with images. Whether it's resizing, optimizing, organizing, uploading, or anything else. Script Kit provides a `drop` component so you can simply drag and drop images into a script and process them in whatever way matches your workflow.


[Open drop-image in Script Kit](https://scriptkit.com/api/new?name=drop-image&url=https://gist.githubusercontent.com/johnlindquist/fe5b06f81cfea0d2dd94f743bc40ed15/raw/40a1b6ab271f7f664d7e54c4b779b056c6195b1c/drop-image.js")

```js
// Name: Drop Image

import "@johnlindquist/kit"

let files = await drop()
let imagePath = files[0].path

let Jimp = await npm("jimp")

let image = await Jimp.read(imagePath)
let newImagePath = path.resolve(
  path.dirname(imagePath),
  `new-${path.basename(imagePath)}`
)
await image
  .resize(parseInt(await arg("Width:"), 10), 100)
  .write(newImagePath)

```


Manipulate Images With Drag and Drop in Script Kit

Chrome allows developers to execute JavaScript using Apple Events. When you enable this feature, you can write scripts that can access and modify the page you're currently browsing. 

You can get the site's contents, update storage, or do any action you would typically do through an extension of the Chrome Dev Tools.

```jsx
let value = await applescript(`
tell application "Google Chrome" to tell window 1
	get execute active tab javascript "

${js}

"
end tell
`)
```

Control Chrome From Script Kit using Apple Events

Many developers use Chrome's Dev Tools as a playground for testing out data and JavaScript features. Script Kit enables you to launch Chrome Dev Tools instantly by simply using the `dev` method. Any data you pass into `dev` will show up in the console so you can inspect and play with the data in any way that you want.

[Install dev-tools-demo](https://scriptkit.com/api/new?name=dev-tools-demo&url=https://gist.githubusercontent.com/johnlindquist/6dcaf61e92819114c7478fefd24a71f8/raw/c2e69bfcff581f7929b881638627201b549d23f1/dev-tools-demo.js")

```js
// Name: Dev Tools Demo

import "@johnlindquist/kit"

let response = await get(
  `https://api.github.com/users/johnlindquist/repos`
)
dev(response.data)
```


Instantly Launch Chrome Dev Tools with Script Kit

This lesson walks through a script that can query a Hacker News API and display the information in Script Kit. 

Learn to do several things using Script Kit's global API methods, such as how to make and inspect data requests using `get` and `inspect`, how to display lists of data using `arg`, and how to display previews of the current selection.

[Install Hacker News Script](https://scriptkit.com/api/new?name=hackernews&url=https://gist.githubusercontent.com/johnlindquist/8728c8f56dd88951127810f0dea818d2/raw/8be4d28aa695f7842b610be258db7cd2e823a239/hackernews.js)

Write a Script with Script Kit to Browse Hacker News

In this course, Chris takes a README-driven approach to build out a Rust CLI that supports content authoring for a digital garden.

Specifically, you'll work through the first command for the CLI, writing files. You start by initializing a new Cargo binary crate through creating the `write` command that allows you to create files (and their contents) to the folder that you set. 

Step by step you'll learn about:
- Handling user input
- Error handling
- Rust Move semantics
- Rust Copy semantics
- Integration testing

...and more!

Rust is a systems programming language that runs blazingly fast, prevents segfaults, and guarantees thread safety.

rust

Creating a Digital Garden CLI with Rust

While we have `test_write_with_title`
which tests the garden CLI with a given
title flag, we can also set up additional
tests, such as for using a written title
in the document for the filename.

Copy/pasting the code from test one into
test two is a lot of duplication, and
setting up the fake editor isn't going
to change across tests, so let's abstract
the creation of the temporary directory,
the fake editor script path, and the
`Command` itself into a utility function.

The port is relatively straightforward,
with the only changes being that we cut
off the `cmd` modifications before the
`.arg` calls, and now have to return
the `cmd` and the `temp_dir` in a tuple.

We return the actual `Command` and `TempDir`,
yielding ownership to the calling function
because we have no further use for them here,
but the calling function will need to mutate
the `Command` more, and use `TempDir` to test.

```rust
fn setup_command() -> (Command, assert_fs::TempDir) {
    let temp_dir = assert_fs::TempDir::new().unwrap();

    let mut cmd = Command::cargo_bin("garden").unwrap();
    let fake_editor_path = std::env::current_dir()
        .expect("expect to be in a dir")
        .join("tests")
        .join("fake-editor.sh");
    if !fake_editor_path.exists() {
        panic!(
            "fake editor shell script could not be found"
        )
    }

    cmd.env("EDITOR", fake_editor_path.into_os_string())
        .env("GARDEN_PATH", temp_dir.path());
    (cmd, temp_dir)
}
```

The new write tests are considerably shorter
now and focus mostly on which subcommand to
call, and what the user interaction with that
subcommand is.

```rust

fn test_write_with_title() {
    let (mut cmd, temp_dir) = setup_command();
    let assert = cmd
        .arg("write")
        .arg("-t")
        .arg("atitle")
        .write_stdin("N\n".as_bytes())
        .assert();

    assert.success();

    temp_dir
        .child("atitle.md")
        .assert(predicate::path::exists());
}

fn test_write_with_written_title() {
    let (mut cmd, temp_dir) = setup_command();
    let assert = cmd
        .arg("write")
        .write_stdin("N\n".as_bytes())
        .assert();

    assert.success();

    temp_dir
        .child("testing.md")
        .assert(predicate::path::exists());
}
```

Abstracting test utilities inside integration test files

We can print lines to the terminal and
ask the user for various input. So let's
make the interface a bit more usable by
using colored output.

We can do this by bringing an extension
trait for formatting into scope:
`OwoColorize`.

Extension traits are one-off traits that
are used to add functions to a type that
our crate doesn't own. A crate can create
the new trait, then implement the
functions defined by that trait on a type
the crate doesn't own and if we bring that
trait into scope, the type will get the
extra functions. In this case, `OwoColorize`
is used for adding color functions to
formatters like Display and Debug.

We'll start with `ask_for_filename`

```rust
use color_eyre::owo_colors::OwoColorize;

fn ask_for_filename() -> Result<String> {
    rprompt::prompt_reply_stderr(&format!(
        "{}",
        "\
Enter filename
> "
        .blue()
        .bold(),
    ))
    .wrap_err("Failed to get filename")
    .map(|title| slug::slugify(title))
}
```

In `confirm_filename` we can use multiple
formatters to get different colors:

```rust
let result = rprompt::prompt_reply_stderr(&format!(
    "\
{} {}
Do you want a different title? (y/N): ",
    "current title:".green().bold(),
    raw_title,
))
.wrap_err("Failed to get input for y/n question")?;
```

Using owo_colors Extension Traits to colorize terminal output

`PathBuf` lets us check to see if a file
exists before writing to it, so we can
take advantage of that to make sure that
we don't overwrite a file that already
exists, even if the user has indicated
a title that would overwrite another file.

```
let mut i: usize = 0;
loop {
    let dest_filename = format!(
        "{}{}",
        filename,
        if i == 0 {
            "".to_string()
        } else {
            i.to_string()
        }
    );
    let mut dest = garden_path.join(dest_filename);
    dest.set_extension("md");
    if dest.exists() {
        i = i + 1;
    } else {
        fs::rename(garden_tmpfile, &dest)?;
        break;
    }
}
```

Checking for file existence with PathBuf to ensure unique filenames

We have been ignoring implementing tests for the write subcommand. Now is the time that we'll test the command we have been building.

We'll build out a fake editor script that our test will use to check if the command is functioning properly. We will then use `arg` to build the command as we would normally use it ourselves.

After writing the test, we'll see how to clean up after ourselves so that we aren't leaving testing artifacts leftover just from running a test.

---

How to integration test a cargo binary that calls
out to `EDITOR` for modifications and asks the
user for input.

First we'll create a fake editor script to use
instead of vim, etc.

`tests/fake_editor.sh` is a one-liner.

```shell
echo "testing" >> $1;
```

Then we'll write our test using `assert_cmd` to
build up a `Command` that points to the `garden`
binary in out `Cargo.toml`.

We grab the fake editor path script using
`std::env::current_dir()`, make sure the file
exists (which means we've grabbed the right
path), and spawn a new `Command` with the
`EDITOR` variable set to the fake editor
script. The equivalent call to our bin is

```shell
EDITOR=path/to/fake-editor.sh garden write -t atitle
```

And the rust code to execute.

```rust
fn test_write_with_title() {
    let mut cmd = Command::cargo_bin("garden").unwrap();
    let fake_editor_path = std::env::current_dir()
        .expect("expect to be in a dir")
        .join("tests")
        .join("fake-editor.sh");
    if !fake_editor_path.exists() {
        panic!(
            "fake editor shell script could not be found"
        )
    }

    let assert = cmd
        .env("EDITOR", fake_editor_path.into_os_string())
        .arg("write")
        .arg("-t")
        .arg("atitle")
        .write_stdin("N\n".as_bytes())
        .assert();

    assert.success();
}
```

Note that we're asserting that the command
exited sucessfully, and that we're inputing
`N\n` to the bin's stdin, which responds to
a request for user input after the editor
closes.

Note that this doesn't actually test for
the resulting file existence, and also that
we're writing our files out directly into
our default digital garden! We can solve
this with `assert_fs` and `predicates`.

We'll bring in both preludes (a prelude is
a bunch of stuff the library author thinks
will be useful to have around all the time)
and create a new temporary directory.

We'll use the `GARDEN_PATH` environment
variable to pass the temporary directory in
to the `garden` bin and that's it, we've
added a temporary directory to our test.
No more polluting the default garden!

```rust
use assert_fs::prelude::*;
use predicates::prelude::*;

fn test_write_with_title() {
    let temp_dir = assert_fs::TempDir::new().unwrap();

    let mut cmd = Command::cargo_bin("garden").unwrap();
    let fake_editor_path = std::env::current_dir()
        .expect("expect to be in a dir")
        .join("tests")
        .join("fake-editor.sh");
    if !fake_editor_path.exists() {
        panic!(
            "fake editor shell script could not be found"
        )
    }

    let assert = cmd
        .env("EDITOR", fake_editor_path.into_os_string())
        .env("GARDEN_PATH", temp_dir.path())
        .arg("write")
        .arg("-t")
        .arg("atitle")
        .write_stdin("N\n".as_bytes())
        .assert();

    assert.success();

    temp_dir
        .child("atitle.md")
        .assert(predicate::path::exists());
}
```

Now that's enough, but we can also check
to make sure the file we expect to exist
actually exists. `temp_dir.child` gives
us the possible file, and we can assert
on that file with a test from `predicates`
that tests to make sure the file exists.

```rust
temp_dir
    .child("atitle.md")
    .assert(predicate::path::exists());
```

Using assert_fs and predicates to integration test with a real temporary file system

What user experience do we want for the
act of writing?

We have a few options, including letting
users write directly in the console or giving
them a prepared file that they can then open
and write in.

Writing directly in the console doesn't give
a user access to their text editor of choice.
Which means lacking autocomplete, syntax
highlighting, snippets, and other functionality.

Creating a file and handling the path back to
the user doesn't quite work for us either, as
that reduces the utility of the garden write
command to the equivalent of running `touch
filename.md`.

So for our use case, we'd ideally like to open
the user's preferred editing environment,
whatever that happens to be, then wait for them
to be done with the file before continuing.

Since we're working with a CLI tool, we
can safely assume that our users either
know how to, or are willingly to learn how
to, set environment variables. This means we
can use the `EDITOR` variable to select an
editor to use. This is the same way `git commit`
works.

---

In Rust, there is a crate that handles not only
`EDITOR`, but also various fallbacks per-platform.
We'll take advantage of [edit](https://crates.io/crates/edit)
to call out to the user's choice of editor.

The quick usage of edit allows us to call the
`edit::edit` function, and get the user's data.

```rust
pub fn write(garden_path: PathBuf, title: Option<String>) -> Result<()> {
    dbg!(garden_path, title);
    let template = "# ";
    let content_from_user = edit::edit(template).wrap_err("unable to read writing")?;
    dbg!(content_from_user);
    todo!()
}
```

This results in a filename that looks like:
`.tmpBy0Yun` "somewhere else" on the filesystem,
in a location the user would never reasonably
find it. Ideally, if anything went wrong, the
user would be able to take a look at the
in-progress tempfile they were just working on,
which should be in an easily discoverable place
like the garden path.

*We don't want to lose a user's work.*

Additionally, the tempfile doesn't have a file
extension, which means that the user's editor
will be less likely to recognize it as a
markdown file, so we want to add `.md` to the
filepath.

```rust
use color_eyre::{eyre::WrapErr, Result};
use edit::{edit_file, Builder};
use std::io::{Read, Write};
use std::path::PathBuf;

const TEMPLATE: &[u8; 2] = b"# ";

pub fn write(garden_path: PathBuf, title: Option<String>) -> Result<()> {
    let (mut file, filepath) = Builder::new()
        .suffix(".md")
        .rand_bytes(5)
        .tempfile_in(&garden_path)
        .wrap_err("Failed to create wip file")?
        .keep()
        .wrap_err("Failed to keep tempfile")?;
    file.write_all(TEMPLATE)?;
    // let the user write whatever they want in their favorite editor
    // before returning to the cli and finishing up
    edit_file(filepath)?;
    // Read the user's changes back from the file into a string
    let mut contents = String::new();
    file.read_to_string(&mut contents)?;

    dbg!(contents);
    todo!()
}
```

We will use the `edit::Builder` API to generate
a random tempfile to let the user write content
into. The suffix is going to be `.md` and the
filename will be 5 random bytes. We also put
the tempfile in the garden path, which ensures
a user will be able to find it if necessary.

`wrap_err` (which requires the `eyre::WrapErr`
trait in scope) wraps the potential error
resulting from these calls with additional context,
making the original error the source, and we can
keep chaining after that. We `keep` the tempfile,
which would otherwise be deleted when all handles
closed, because if anything goes wrong after the
user inputs data, we want to make sure we don't
lose that data.

After the file is created, we write our template
out to the file before passing control to the
user. This requires the `std::io::Write` trait
in scope. We use a const for the file template
because it won't change. To make the `TEMPLATE`
a const, we also need to give it a type, which
is a two element byte array. Change the string
to see how the byte array length is checked at
compile time.

Since we have access to the `file` already, we
can read the contents back into a string after
the user is done editing. This requires having
the `std::io::Read` trait in scope.

And now we've let the user write into a file
which will stick around as long as we need it
to, and importantly will stick around if any
errors happen in the execution of the program,
so we lose no user data and can remove the
temporary file at the end of the program
ourselves if all goes well.

Passing control to the user's favorite editor (such as VSCode) from the CLI

Given the potential `document_title` we now
want to ask the user if the title is still
what they want to use so that we can write
the file to disk. This will involve two
paths depending on if we have an existing
potential option or not.

We will need to use `ask_for_filename` in
`confirm_filename`, so we'll write that branch
first.

```rust
let filename = match document_title {
    Some(raw_title) => {
        // confirm_filename()?
        todo!()
    }
    None => ask_for_filename(),
};
```

`rprompt` allows us to ask for a small amount
of input from the user (compared to the  much
larger input possible by passing control to
the user's editor).

We'll use `rprompt::prompt_reply_stderr` to
get a response from the user, and `.wrap_err`
to add context to the error if anything goes
wrong.

```rust
fn ask_for_filename() -> Result<String> {
    rprompt::prompt_reply_stderr(
        "\
Enter filename
> ",
    )
    .wrap_err("Failed to get filename")
}
```

We'll use the same behavior for the first
part of `confirm_filename`. We'll also
need to use our first lifetime to account
for the shared reference argument.

`match` can match against multiple values
for the same branch, so we'll take advantage
of that to handle branches for Ns and Ys, as
well as the default case. If anything goes
wrong, such as someone inputting an "a",
we'll fall through using `loop` and ask again
until we get a usable answer

```rust
fn confirm_filename(raw_title: &str) -> Result<String> {
    loop {
        // prompt defaults to uppercase character in question
        // this is a convention, not a requirement enforced by
        // the code
        let result = rprompt::prompt_reply_stderr(&format!(
            "\
current title: `{}`
Do you want a different title? (y/N): ",
            raw_title,
        ))
        .wrap_err("Failed to get input for y/n question")?;

        match result.as_str() {
            "y" | "Y" => break ask_for_filename(),
            "n" | "N" | "" => {
                // the capital N in the prompt means "default",
                // so we handle "" as input here
                break Ok(slug::slugify(raw_title));
            }
            _ => {
                // ask again because something went wrong
            }
        };
    }
}
```

While filenames can technically have spaces
in them, we're going to slugify our filenames
like urls for a few reasons. One is that
when starting out, many programmers fail to
quote filename arguments in their bash scripts,
which results in filenames with spaces being
treated as separate arguments. Another is
that this is a digital garden CLI, and
digital gardens are often stored in git,
accessed from multiple file systems, as well
as from URLs. While strictly speaking we don't
need to slugify our filenames, we will here
so as to adhere to a safer set of characters.

```
cargo add slug
```

We'll map over the `Result` from `rprompt`
which allows us to operate on the internal
value if it's `Ok`.

```rust
fn ask_for_filename() -> Result<String> {
    rprompt::prompt_reply_stderr(
        "\
Enter filename
> ",
    )
    .wrap_err("Failed to get filename")
    .map(|title| slug::slugify(title))
}
```

We'll also use slugify in `confirm_filename`

```rust
match result.as_str() {
    "y" | "Y" => break ask_for_filename(),
    "n" | "N" | "" => {
        // the capital N in the prompt means "default",
        // so we handle "" as input here
        break Ok(slug::slugify(raw_title));
    }
    _ => {
        // ask again because something went wrong
    }
};
```

The fact that `confirm_filename` and
`ask_for_filename` have the same return type
is important because the branches of a `match`
need to return the same type.

Prompting the user for input with rprompt and loop

We want to build out functionality to let the user
write into a file in their garden. To do that, we
have to know the location of their garden.

We'll start by implementing a function to get the
default garden path.

```rust
fn get_default_garden_dir() -> Result<PathBuf> {
    let user_dirs = UserDirs::new().ok_or_else(|| eyre!("Could not find home directory"))?;
    Ok(user_dirs.home_dir().join(".garden"))
}
```

This uses the user's system directories via the
`directories` crate to find their home dir. We
then add `.garden` to the home dir for the default
garden.

Ideally, we would want the user to be able to
specify their garden path. Using structopt, we
can let them specify it as a short, long, or
env var.

In our README, we'll specify how this should work.

````

```shell
GARDEN_PATH=~/github/my-digital-garden garden write
garden -p ~/github/my-digital-garden write
garden --garden_path ~/github/my-digital-garden write
```
````

And in our `Opt` struct, we can tell structopt how to
handle it. Putting this in our `Opt` struct instead of
our `Command` enum make it a global flag that can be
specified for all subcommands.

```rust
struct Opt {
    #[structopt(parse(from_os_str), short = "p", long, env)]
    garden_path: Option<PathBuf>,

    #[structopt(subcommand)]
    cmd: Command,
}
```

structopt is only handling the user input, so in
`main()` we still need to fall back to the
default if the `garden_path` is not specified.

```rust
let garden_path = match opt.garden_path {
    Some(pathbuf) => Ok(pathbuf),
    None => get_default_garden_dir().wrap_err("`garden_path` was not supplied"),
}?;
```

if we force an `Err` in `get_default_garden_dir`
We can see how the wrapped errors end up stacking.

```
Error:
   0: `garden_path` was not supplied
   1: Could not find home directory
```

We'll also change the signature of `write` to
accept the pathbuf.

```
Command::Write { title } => write(garden_path, title),
```

Note that we need to import `PathBuf`. We can also
`dbg!` the garden path and title to see their
values.

```rust
use std::path::PathBuf;

pub fn write(garden_path: PathBuf, title: Option<String>) -> Result<()> {
    dbg!(garden_path, title);
    todo!()
}
```

Adding support for global flags that apply for all subcommands with structopt

In our subcommand match we'll add a new function called
`write` to handle our `Write` subcommand. We'll pass it
the argument title and we'll "import" it from a library
called `digital_garden`. 

```rust
use digital_garden::write;

fn main() -> Result<()> {
    color_eyre::install()?;

    let opt = Opt::from_args();
    dbg!(&opt);
    match opt.cmd {
        Command::Write { title } => write(title),
    }
}
```

To create this library, add a new section in `Cargo.toml`.
We can give our library a name and set which file to treat
as the entrypoint. `src/lib.rs` is the default file for
a library crate, but we can also be explicit about what we
want to happen.

```toml
[lib]
name = "digital_garden"
path = "src/lib.rs"
```

This leaves us with two crates, a binary crate, of which we
can have any amount, and a library crate, of which we can
only have 0 or 1.

We'll create the `src/lib.rs` file with the following
contents.

```rust
mod write;

pub use write::write;
```

Rust's module system is explicit, so when we write
`mod write`, we are saying that the library `digital_garden`
has a module `write`, such that the path to the module is
`digital_garden::write`.

In our case `digital_garden::write` corresponds to a file
at `src/write.rs`.

Secondly, the `pub use write::write` says that we have a
function called `write` in the module called `write` and
that the function will be exposed from lib publicly, so
people who have the `digital_garden` library installed
will be able to use the write function as
`digital_garden::write`.

In JavaScript the closest analogue is exporting a function
from another file that exports it.

```js
export { write } from './write.js';
```

By using use, we're shortening the public path to the write
function. A private module "write" with a function "write"
also exists at `digital_garden::write::write`. If you try
to use this path, Rust will tell you there is a private
module there.

Finally, in `src/write.rs`, we can write a public function
that matches the return type of our main function, since
we're using it as the return value from our match function.

We'll accept an `Option<String>` as an argument as well,
since that's what we expect from the cli arguments.

```rust
use color_eyre::Result;

pub fn write(_title: Option<String>) -> Result<()> {
    todo!()
}
```

The purpose of creating this module path is because we
know that we're going to set up more modules for each
subcommand in the future, and we can control both the
placement of our code on disk, as well as the path users
will interact with to use our module.

Creating a library crate to support a Rust binary crate in the same package

After getting the user's input, we need to
figure out a filename to write the file out
as. We can either use the title the user
passed in, or we can try to use heuristics
to find a title from the markdown content.

We can combine the user's title from the cli
arguments (an `Option`) with the logic we
use as a heuristic for the markdown (also
an `Option`).

```rust
// use `title` if the user passed it in,
// otherwise try to find a heading in the markdown
let document_title = title.or_else(|| {
    contents
        .lines()
        .find(|v| v.starts_with("# "))
        // markdown headings are required to have `# ` with
        // at least one space
        .map(|maybe_line| maybe_line.trim_start_matches("# ").to_string())
});
```

`option.or_else()` returns the original
option if it is a `Some` value, or executes
the function to return a different `Option`
if not. So `document_title` is an
`Option<String>` either way and we've
covered all of the possible scenarios:

1. the user has passed in a title
2. the user has written content with a heading
3. the user has done neither of these

Which leaves us to our markdown heuristic.

Markdown headings are required to have `# `
with at least one space, so we can turn
`contents` into an iterator and `find` the
first line that starts with `# `.

If we find one, we want to trim `# ` off of
the heading to get just the heading content,
so we can use `map` to operate on the value
inside of the `Option` returned by `find` if
it exists.

```rust
contents
    .lines()
    .find(|v| v.starts_with("# "))
    .map(|maybe_line| maybe_line.trim_start_matches("# ").to_string())
```

Now this code works fine, but it exposes a bug
in our file handling.

If we `write_all` to the `file`, that moves the
internal cursor to the end of that content. So
when we run our code now, `contents` is
skipping the first `# ` bytes, which means our
heuristic will only find the second heading in
the file.

To fix this, we can bring in the `std::io::Seek`
trait, and seek to the beginning of the file.

```rust
let mut contents = String::new();
file.seek(SeekFrom::Start(0))?;
file.read_to_string(&mut contents)?;
```

Using the Option type in Rust to pick from multiple values that might not exist

We can match on `opt.cmd` to determine which `Command`
variants was called. This lets us specify the variants
as match criteria and also destructure the values out
to use them.

```rust
    let opt = Opt::from_args();
    dbg!(&opt);
    match opt.cmd {
        Command::Write { title } => todo!(),
    }
```

As written, we run into two issues with this code.
One is that we aren't using `title` yet. In
destructuring, we can use the placeholder (`_`) in our
[pattern](https://doc.rust-lang.org/reference/patterns.html).

```
warning: unused variable: `title`
  --> src/main.rs:32:26
   |
32 |         Command::Write { title } => todo!(),
   |                          ^^^^^ help: try ignoring the field: `title: _`
   |
   = note: `#[warn(unused_variables)]` on by default
```

```rust
fn main() -> Result<()> {
    color_eyre::install()?;

    let opt = Opt::from_args();
    dbg!(opt);
    match opt.cmd {
        Command::Write { title: _ } => todo!(),
    }
}
```

The second issue is that the `opt` variable is moved
before we try to destructure it. If we look at the
compiler output before we fixed the `title` issue we
see the value use after move pointing at `title`
specifically.

```
error[E0382]: use of moved value
m  --> src/main.rs:32:26
   |
29 |     let opt = Opt::from_args();
   |         --- move occurs because `opt` has type `Opt`, which does not implement the `Copy` trait
30 |     dbg!(opt);
   |     ---------- value moved here
31 |     match opt.cmd {
32 |         Command::Write { title } => todo!(),
   |                          ^^^^^ value used here after move
```

If we look at it after fixing the unused `title`,
we see the compiler point at `opt.cmd`.

```
error[E0382]: use of moved value: `opt.cmd`
  --> src/main.rs:31:11
   |
29 |     let opt = Opt::from_args();
   |         --- move occurs because `opt` has type `Opt`, which does not implement the `Copy` trait
30 |     dbg!(opt);
   |     ---------- value moved here
31 |     match opt.cmd {
   |           ^^^^^^^ value used here after move
```

This is because opt is moved into the `dbg` call because
it doesn't implement the `Copy` trait. Now, we could
implement or derive `Copy` for `Opt` if `Command` can
implement `Copy`, but we can not implement `Copy` for
`Command`. This is because `String` does not, and can not,
implement `Copy`, and we have a `String` in our `title`.

Without diving into the depths of `Copy`, [`Clone`](https://doc.rust-lang.org/std/clone/trait.Clone.html),
and allocation, there is something we have that already
implements `Copy`, so we don't need to.

Shared references implement `Copy`.

So instead of passing `opt` into `dbg!`, and thus using
move semantics (which are the default in Rust), we can
pass a shared reference in: `dgb!(&opt)` which lets us
use copy semantics.

The difference between move semantics and copy semantics
in this case is that we can access `opt` after passing it
to `dbg!`.

Under the hood, both a copy and a move can result in bits
being copied in memory, although this is sometimes
optimized away. -- [Copy](https://doc.rust-lang.org/std/marker/trait.Copy.html)

```rust
fn main() -> Result<()> {
    let opt = Opt::from_args();
    dbg!(&opt);
    match opt.cmd {
        Command::Write { title: _ } => todo!(),
    }
}
```

Matching structopt commands and using Rust values after move

Writing clear and effective prompts or rules for AI (.mdc files in Cursor) can be tricky. Your instructions might start out clumsy or ambiguous, leading to less-than-ideal AI responses. This lesson demonstrates how to quickly refine them by referencing a dedicated "best practices" rule file directly within Cursor's inline editor (`Cmd+K`).

**Workflow demonstrated in this lesson:**

*   Select the text (a prompt or section of a rule) you want to improve.
*   Use `Cmd+K` to open the inline AI editor.
*   Reference your separate prompt best practices rule file using `@` (e.g., `@prompt-improve.mdc`).
*   Cursor's AI rewrites the selected text, applying the referenced best practices.
*   Accept the improved version (`Cmd+Enter`).
*   **(Bonus)** Use `Cmd+L` to send the edited section to the Agent chat and ask it to compare the changes with the previous version (using git history) and explain *why* the new version is better (e.g., improved clarity, stronger action verbs, explicit exclusions).

**Key benefits:**

*   Instantly improve prompt clarity and effectiveness based on your own defined standards stored in a rule file.
*   Reduces ambiguity in instructions, leading to better AI outcomes.
*   Leverages Cursor rules (`.mdc` files) as reusable templates for prompt refinement.
*   Provides insight into *how* and *why* prompts are improved via Agent comparison.

Stop wrestling with clunky prompts. Use this technique to systematically apply best practices and level up your AI instructions directly within Cursor.

```markdown
# Improve the User's Prompt Following the Patterns Below

> Practical prompt patterns to help anyone get clearer, more reliable answers from an AI agent.

| # | Pattern                       | Why It Matters                                                          | Template                                                                 | Example Prompt                                                                                             |
|---|-------------------------------|-------------------------------------------------------------------------|--------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------|
| 1 | Lead with the ask             | The model reads top-down; putting the goal first stops it wandering.    | `Do X. [context]`                                                        | "Summarize this PDF in 5 bullet points. The text is below: ..."                                            |
| 2 | Repeat the key ask at the end | Long contexts sometimes truncate; an end-cap protects you.              | `...[detail]... REMEMBER: Do X.`                                         | "List pros & cons, keep it balanced. REMEMBER: 5 pros, 5 cons."                                             |
| 3 | Specify output shape          | Dictating format cuts revision loops.                                   | `Return as: 1) short title, 2) table (CSV).`                             | "Give 3 holiday ideas. Format: destination, flight-time-hrs, avg cost."                                    |
| 4 | Use clear delimiters          | Backticks/headings/XML keep sections from blending.                     | `TEXT TO ANALYSE`                                                        | "Rate the style of the text between the fences."                                                            |
| 5 | Induce step-by-step thinking  | Planning first boosts multi-step accuracy.                              | `Think step-by-step then answer.`                                        | "Solve this puzzle. Think step-by-step before giving the final move."                                      |
| 6 | Ask it to plan its workflow   | For big jobs, AI outlines tasks before doing them.                      | `First draft a plan, wait, then execute.`                                | "We're writing an e-book. ❶ Outline chapters. ❷ Wait. ❸ When I say 'go', draft chapter 1."              |
| 7 | Limit or widen knowledge sources | Controls hallucination.                                               | `Use only the info below. / Combine basic knowledge + this context.`     | "Using only the product sheet below, write FAQs."                                                          |
| 8 | Guide information retrieval   | Helps AI pick the right docs before answering.                          | `List which docs look relevant, then answer.`                            | "30 sales memos attached. 1) Name the 5 most relevant. 2) Summarise their common points."                 |
| 9 | Show a style/example          | Anchors tone, length, vocabulary.                                       | `Match the style of: <example>`                                          | "Review this gadget in the style of the sample below: 'Short, witty, 3 key facts...'"                      |
| 10| Set correction handles        | One-line fixes let you steer quickly.                                   | `If length > 150 words, shorten.`                                        | "Describe blockchain to a child. If your answer is >150 words, cut it in half."                            |
| 11| Tell it when to stop or loop  | Prevents half-finished lists or runaway essays.                         | `Keep going until you list 20 ideas, then stop.`                         | "Brainstorm webinar titles. Give exactly 12, then finish."                                                  |
| 12| Request the hidden reasoning  | Good for audits; otherwise omit to keep it short.                       | `After the answer, include a brief reasoning section.`                   | "Which of these stocks looks over-valued? Answer first; add a 2-sentence rationale below a divider."        |
```

John Lindquist

Automatically Improve Cursor Rules Using Custom Prompts

Cursor's `Cmd+K` is fantastic for targeted code edits, but sometimes you need the AI to operate beyond the confines of your current selection—like extracting a piece of logic into a brand-new, top-level function. Standard `Cmd+K` often falls short, keeping changes localized. This lesson reveals how to overcome this limitation using Cursor's "Edit Full File" mode, turning complex refactoring of legacy code into a streamlined process.

**Workflow demonstrated in this lesson:**

*   Contrast the behavior of standard `Cmd+K` (Edit Selection) with the "Edit Full File" mode.
*   Use `Cmd+K`, then `Cmd+Shift+Enter` (or the dropdown) to invoke "Edit Full File" mode for broader AI refactoring.
*   Instruct the AI to extract selected code into new functions at the top level of the module.
*   Demonstrate breaking down a large, monolithic function into multiple smaller, exported, and testable functions.
*   Utilize the "Fix in Chat" feature (`Cmd+Shift+D`) to send error messages and surrounding code directly to an AI agent for quick resolution.

**Key benefits:**

*   **True Top-Level Extraction:** Move code out of its current scope and into new, independent functions anywhere in the file.
*   **Simplified Legacy Code Refactoring:** Easily break apart large, untestable functions into manageable, testable units.
*   **Improved Code Structure:** Quickly organize and modularize your codebase with AI assistance.
*   **Contextual Error Fixing:** Send errors and relevant code snippets to an AI agent via "Fix in Chat" for efficient debugging.

This lesson shows a powerful technique for serious refactoring in Cursor, especially when dealing with large or legacy codebases, making your functions more modular, testable, and maintainable.

Clean up Legacy Functions for Testability in Cursor (0.50+) with cmd+k

Hey Everyone 👋,  
  
John Lindquist here with the **fifth issue of AI Dev Essentials!** This week, I've been relying more on GPT-4.1 in Cursor (which seems vastly improved since I last tried it) when building CLI and node tools. GPT 4.1 seems more aggressive in using tools as a pseudo reasoning step, while Gemini 2.5 Pro often jumps straight to code changes based on its "real" reasoning. I'm finding that GPT 4.1 in Cursor is becoming legit, especially with a "please explain that back to me" workflow. The speed allows for almost instant alignment. This "time to alignment" is so critical. It's that moment you \*know\* the AI is on the same page after a brain dump. On another note, programming is getting weird! I had a moment working on an MCP process manager where all the code was correct, but my \*English descriptions\* of what it was doing were too technical and thus misleading. Wild times!  
  
Big news: **Gemini 2.5 Pro upgrade** launched a couple hours ago. I'm super excited for this to land in Cursor, hopefully even later today! For now, you can try it out in AI Studio. I can't wait to see the improvements in Cursor's agent mode, especially since the tool calling is reportedly vastly improved. Beyond that, the major themes this week include impressive new **generative AI capabilities** for voice, a continued surge in **developer tooling and frameworks** (especially open source), and ongoing discussions around **prompting best practices and AI model quirks.**  
  
Let's dive in!

* * *

🎓 New egghead.io Lessons This Week
-----------------------------------

*   [When to Use Cursor Rules Instead of MCPs](https://egghead.io/when-to-use-cursor-rules-instead-of-mcps~zazdq)
    
    This lesson demonstrates creating a Cursor Rule for a CLI tool (\`ghx\` for scraping GitHub examples) to make it available to the agent. It highlights how agent requested rules can be a simpler alternative to MCPs for tool integration, offering more control over context.
    
*   [Brainstorming and Planning with Mermaid Diagrams in Cursor](https://egghead.io/brainstorm-and-planning-with-mermaid-diagrams-in-cursor~9iu69)
    
    Learn to use Cursor Rules to enable Mermaid diagram generation. The lesson shows how to create and iterate on diagrams (like basic to do app flows and sequence diagrams for optimistic updates) visually within Cursor, providing a way to plan before coding.
    
*   [Create Complex Cursor Shortcuts to Run Commands and Generate AI Commits](https://egghead.io/create-complex-cursor-shortcuts-to-run-commands-and-generate-ai-commits~7arof)
    
    This lesson walks through setting up a custom keyboard shortcut (Command+Shift+S) in Cursor's JSON settings to run a sequence of commands: stage all git changes, use Cursor to generate a commit message, commit, and then push/pull.
    

* * *

💰 AI Coding Assistant Valuations & Acquisitions Skyrocket
----------------------------------------------------------

_The AI coding assistant space is on fire with some massive financial moves happening._

*   **Anysphere (Cursor IDE) Hits $9 Billion Valuation**
    
    Cursor's parent company, Anysphere, reportedly secured a new $900 million funding round, pushing its valuation to an estimated $9 billion. Key investors include Thrive Capital, Andreessen Horowitz (a16z), and Accel. This is a huge jump from their $2.5 billion valuation in January 2025, showing incredible investor confidence. (Source details from PitchBook: "Maker of AI-coding software Cursor is doubling its valuation every 8 weeks", citing The Financial Times; and Crunchbase News: "AI-Powered Coding Tool Anysphere Lands $900M At $9B — Report".)
    
*   **OpenAI Reportedly Acquires Windsurf (formerly Codeium) for ~$3 Billion**
    
    In what would be its largest acquisition, OpenAI is said to be acquiring Windsurf. Windsurf, an AI-assisted coding tool, was previously valued around $1.25 billion. This move clearly signals OpenAI's intent to bolster its position in the AI developer tools market. (Source details from Mashable citing Bloomberg: "OpenAI is buying Windsurf for $3 billion. What does that mean for ChatGPT?"; and Investing.com citing Reuters/Bloomberg: "OpenAI agrees to buy Windsurf for about $3 billion, Bloomberg News reports".)
    
    My take: It seems like OpenAI, above all, needed the team behind Windsurf. The value of people talented in code generation is through the roof right now. But they also needed a product ASAP, so it's essentially like they're paying for same day shipping, even with this incredible amount of money. And it boggles my mind that Cursor's valuation is estimated at $9 billion. I know a lot of us developers look at both Windsurf and Cursor and kind of think to ourselves that we could have forked VS Code and built these things. So, seeing an incomprehensible amount of value being assigned to what looks like something that we could build if we just would have dedicated our time to it is always pretty shocking.
    

* * *

Here are the highlights that caught my eye this past week:

🚀 Model & Platform Updates
---------------------------

_This week saw some impressive new releases and updates in the model space, particularly around generative capabilities for voice, alongside ongoing platform refinements._

*   **Gemini 2.5 Pro Upgrade: Now with Better Coding and Function Calling.**
    
    Google announced a significant upgrade to Gemini 2.5 Pro. It boasts improved performance in coding (especially frontend web dev, editing, and transformation) and more reliable function calling. The new model (\`gemini-2.5-pro-preview-05-06\`) automatically replaces the previous version with the same pricing and rate limits. [Details](https://developers.googleblog.com/en/gemini-2-5-pro-io-improved-coding-performance)
    
    This is the update I've been waiting for! Reliable tool/function calling is a massive unlock for agentic workflows. Eager to put this through its paces in Cursor.
    
*   **Minimax's High Ranking Text to Speech Models on Replicate.**
    
    Minimax's speech-02 models (speech-02-hd and speech-02-turbo) and voice cloning capabilities are now available on Replicate. They are touted as top ranking TTS models. [Replicate Blog](https://replicate.com/blog/minimax-text-to-speech)
    
*   **Gemini 2.0 Flash Photo Editing.**
    
    Google's Gemini 2.0 Flash is showcased for its ability to edit photos using natural language, without needing traditional tools like layers or brushes. (Mentioned by Rishabh, [Source](https://x.com/Rixhabh__/status/1919641187696418896))
    
*   **Grok 3.5 Beta Coming Soon.**
    
    Elon Musk announced that Grok 3.5 will have an early beta release next week for SuperGrok subscribers. He claims it can accurately answer technical questions by reasoning from first principles. ([Source](https://x.com/elonmusk/status/1917099777327829386))
    

🛠️ Developer Tooling, Frameworks & Ecosystem
---------------------------------------------

_The developer ecosystem continues to expand with new open source tools for data extraction, updates to agent frameworks, and interesting discussions on platform choices and best practices._

*   **ContextGem for Structured Data Extraction.**
    
    A new open source LLM framework, ContextGem, aims to simplify extracting structured data and insights from documents with minimal Python code. ([GitHub ContextGem](https://github.com/shcherbak-ai/contextgem) via Sumanth)
    
*   **Cloudflare Agents SDK Update: Upcoming Auth and State Handling.**
    
    An upcoming PR for the Cloudflare Agents SDK will enhance Agent classes to handle OAuth, redirect URLs, and reconnections out of the box. This makes every Agent an MCP client with state management. (Mentioned by Sunil Pai, [PR Link](https://github.com/cloudflare/agents/pull/254))
    
    Authorization has been one of the biggest challenges for MCPs, specifically because the MCP sits behind the AI agent. Most things are just configured in a simple JSON file defining your MCP servers. Triggering OAuth workflows inside UIs, which are essentially streaming text and working on other tasks, has been weird. Any sort of standardization and convenience they can add to authentication is very welcome, in my personal opinion.
    
*   **Persistent File Storage for Cloudflare Durable Objects Demo.**
    
    A demonstration of a real filesystem and terminal for Cloudflare Durable Objects using a Rust based chunked SQLite FUSE strategy ported to \`ctx.storage.sql\` and an xterm.js frontend. ([Demo](https://dofs.verifried.workers.dev/), [Tweet](https://x.com/benallfree/status/1919224851409379786))
    
    I love that as more new developers come on board, we can give them paradigms they're familiar with, like file systems and basic terminals to manipulate files. I think one of the biggest holdups for developers with things like images on their websites is they're just so unfamiliar with the various concepts around routing and where things are hosted. Any effort that brings existing paradigms that beginners understand to complex subjects is welcome, in my opinion.
    
*   **Stripe MCP Monetization.**
    
    Stripe now allows MCP developers to monetize their MCPs with subscriptions or usage based billing. It's compatible with various clients including Cloudflare's Agents SDK. ([Stripe MCP Payments](http://stri.pe/go/mcp-payments) via Jeff Weinstein)
    
    Stripe is obviously the biggest player in allowing people to spend money on things, so it's no surprise they're introducing ways to make money with MCPs. I'm definitely curious to see what sort of services will pop into existence now that you could request them from AIs. I'm in no hurry to pay for any MCP service myself, and I don't have any to recommend yet, but I'll definitely keep my eyes out for what may manifest from this.
    
*   **Cloudflare Workers Monorepo Template.**
    
    Jacob Hands shared a template for creating Workers monorepos. It's inspired by the \`mcp-server-cloudflare\` repo which deploys multiple Workers from one repo with shared packages. ([GitHub Template](http://github.com/jahands/workers-monorepo-template) via Jacob Hands)
    
*   **A Story of Migrating from Vercel to Cloudflare.**
    
    A developer shared their experience migrating their entire infrastructure to Cloudflare. They cited reasons related to control, security, performance, and pricing. ([Source](https://x.com/Osintly/status/1917964184198823951))
    
    I've seen a lot of people try to do this, especially with the perception that Next.js kind of locks you into Vercel. So I just think it's valuable for people to realize that options are out there and that a lot of your projects are more portable than you think they are.
    
*   **New MCP Authorization Spec.**
    
    A new MCP authorization spec has been introduced. It defines the MCP server as an OAuth "resource server" and outlines the authentication flow. Den Delimarsky provided a blog post with more details. ([Den Delimarsky's Blog](https://den.dev/blog/new-mcp-authorization-spec) via Yenkel)
    
    It'll be interesting to watch any of the MCP specs evolve over time. I know there's been a lot of work done with streaming. Authorization is obviously going to play a huge part. It'll be interesting to see all these pieces of MCPs fitting together and how all their auth will sync up, where the walls are built, and what they can and cannot do to communicate with each other. I can't imagine a system of roles and identifications for the various MCPs and how they allow each other to speak to one another is going to be critical for MCP moving forward.
    

💡 Prompting, Workflows & Best Practices
----------------------------------------

_This week brought fresh perspectives on interacting with LLMs, managing context, and structuring UI generation prompts._

*   **"AI Rivalry" Prompting Technique.**
    
    Greg Isenberg shared a technique of pitting different AI models (ChatGPT, Claude, Grok) against each other. You feed one's output to another with critical feedback to elicit better results. ([Source](https://x.com/gregisenberg/status/1918661279641723158))
    
    There's so much potential left to unlock when multiple models work with and against each other. It's still a little cost prohibitive to use this for a lot of the tools you're probably building by yourself. But I love any sort of back and forth or ping pong or even parallel agents coming to conclusions and fighting over what the best answers will be. I can only imagine what these techniques unlock. Hopefully someone will send me a credit card to run a bunch of experiments.
    
*   **Theo's Take on Major AI Models' Quirks.**
    
    Theo (t3.gg) shared a list of issues and frustrations with various AI models and providers including Gemini, Claude, ChatGPT, and Mistral. He covers setup, cost, and availability. ([Source](https://x.com/theo/status/1918810601968513249))
    
    I really agree with Theo's breakdown here. I've had a very similar experience trying out each of the models. So, definitely worth checking out his list. It's just a well written, very concise list.
    
*   **Tips for LLM Tools Editing Files with Diffs.**
    
    Matt Pocock initiated a discussion seeking examples and best practices for implementing LLM tools that edit files using diffs in TypeScript. ([Source](https://x.com/mattpocockuk/status/1918758488777408528))
    
    I hopped in on this discussion with Matt about my personal experience working with diffs. It's just so challenging to get LLMs to produce precise diff worthy code based on line numbers and white space, and then trying to match that up with the original file. It was a fascinating thread to read about how some people are solving the problem by using a second LLM trained specifically on diffing two files and essentially rewriting the entire file. It's a problem space I spent a lot of time in last year, but I'm just going to leave it up to all of these IDEs and tooling who are so far beyond where I got to now that it's not worth spending time on for myself.
    
*   **Prompting Guide for Lovable.dev for UI Generation.**
    
    Felix Haas shared a step by step guide on how to prompt Lovable.dev for better UI generation. It covers high level descriptions, look and feel, brand guidelines, UI library specification, and component details. ([Source](https://x.com/felixhhaas/status/1918316680108736775))
    
    I have some non developer friends using Lovable, and they seem to be really enjoying this, so I'm sending this over to them. This newsletter is more focused on a bit more experienced developers and what they would expect to read about, but this is definitely something I'm going to shoot over to a few people I know.
    
*   **Survey of AI Agent Protocols.**
    
    Elvis (omarsar0) highlighted key takeaways from a survey paper on AI Agent Protocols. He discussed the agent internet ecosystem, timelines, protocol overviews (MCP, A2A), evaluation considerations, and use case analyses. ([Source](https://x.com/omarsar0/status/1918723145923453022))
    
    I know that MCP gets so much attention, and it's interesting how it's taken over so quickly, but there are many other concepts to pay attention to, such as the A2A protocol. So it's definitely worth reading through how some of these things, when and why you would use different protocols.
    
*   **Vercel AI SDK \`useAssistant\` Hook.**
    
    Sam Denty pointed to the \`useAssistant\` hook in the Vercel AI SDK for building AI assistants. ([AI SDK Docs](https://ai-sdk.dev/docs/reference/ai-sdk-ui/use-assistant))
    
    The \`useAssistant\` hook was something I wasn't aware of, and Sam shared it with me. For an awesome hook for building out web based AI tools, I didn't know it existed, and I thought I'd pass it along because it covers a lot of scenarios that I would be building for web cases.
    

👨‍💻 X Posts from the Cursor Team
----------------------------------

_Folks from the Cursor team continue to share valuable insights and tips for getting the most out of the AI-first IDE._

*   **Effective Context Management in Cursor.**
    
    Eric Zakariasson (from the Cursor team) provided tips on managing context in Cursor. He emphasized the combination of intent and existing codebase state, using surgical @ mentions, rules for patterns, and MCP for external systems. ([Cursor Docs](https://docs.cursor.com/guides/working-with-context) via Eric Zakariasson)
    
*   **New Cursor Guide for Web Development.**
    
    Eric Zakariasson (from the Cursor team) also published a new guide for web development workflows in Cursor. ([Cursor Docs](https://docs.cursor.com/guides/tutorials/web-development))
    
*   **Cursor Rules for Better Results.**
    
    Ian (shaoruu) (from the Cursor team) shared examples of Cursor rules they use. These include specifying libraries, listing UI components, defining commands, and using "Auto Attach" for specific file contexts. ([Cursor Docs](https://docs.cursor.com/context/rules) via Ian)
    

✨ Workshop Spotlight: Conquer the Complexity of Cursor
------------------------------------------------------

Master practical AI development workflows in Cursor. This hands on workshop covers Agents, multi file analysis, effective prompting, Cursor rules, and strategies for handling AI failures.

**When:** Thursday, May 8, 2025, 9:00 AM to 2:00 PM (PDT)

**Where:** Zoom (Live Q&A included)

**Investment:** $249

[Read More](https://egghead.io/workshop/cursor) | [Register Now](https://buy.stripe.com/3csdSh2GTdhucTedQW)

_(Team training also available)_

That’s the scoop for this issue! As a final note, I'm excited for Google I/O in a couple of weeks. Since they're already releasing a huge update to my favorite model, and there are rumors around other models that are even better, I just can't wait to see what they have in store for us.  
  
If you have any feedback or questions, hit reply! Always happy to chat about the latest in AI dev tools.  
  
John Lindquist  
[egghead.io](https://egghead.io/)

AI Dev Essentials #5: 🤯 Cursor Valued at $9B, Gemini 2.5 Pro & More AI Updates

Want Cursor's AI Agent to intelligently run your custom CLI tools without complex MCP setups or constant hand-holding? If only the AI could understand your tool's `--help` output, life would be much simpler. This lesson demonstrates exactly that, leveraging Cursor's "Rules" feature with the "Agent Requested" type. You'll learn to feed your CLI's help text directly into a rule, effectively teaching the AI to operate your custom commands through natural language.

**Workflow demonstrated in this lesson:**

*   Execute your custom CLI tool's `--help` command (e.g., `ghx --help`) and pipe its output to your clipboard (e.g., `| pbcopy`).
*   Create a "New Cursor Rule" (Cmd/Ctrl + Shift + P, then "New Cursor Rule").
*   Paste the copied help text into the main content area of the rule.
*   Set the "Rule Type" to "Agent Requested".
*   Provide a clear, concise "Description" (e.g., "Search GitHub") that will act as the trigger phrase for the AI.
*   In an Agent chat, prompt the AI using natural language that corresponds to your rule's description (e.g., "Search GitHub for examples of package.jsons using semantic-release").
*   Observe as the Agent uses the rule's help text to correctly formulate and run your custom CLI command.
*   Leverage the tool's output by further instructing the AI to generate or modify project files (e.g., create a `package.json`, `.releaserc.json`, or `biome.json` based on findings).

**Key benefits:**

*   Empowers Cursor's AI Agent to directly use your existing custom CLI tools without needing to modify the tools themselves.
*   Simple and intuitive setup: Uses your tool's own documentation (`--help` output) as the "API" for the AI.
*   Efficient context usage: The detailed help text for your tool is only loaded into the AI's context when your specific trigger phrase is detected.
*   Provides a straightforward and powerful alternative to MCPs for many CLI tool integration scenarios, especially when tools already have good help messages.
*   Enables complex, natural language-driven automation by combining custom tool execution with the AI's code generation and modification capabilities.

By using "Agent Requested" rules, you gain maximum control over how and when the AI interacts with your custom tools, making the Agent a more versatile and powerful assistant tailored to your specific command-line utilities.

**Resources**

[https://www.npmjs.com/package/@johnlindquist/ghx](https://www.npmjs.com/package/@johnlindquist/ghx)



When to Use Cursor Rules Instead of MCPs

Planning complex application logic or workflows before coding can be challenging. This lesson demonstrates how to leverage Cursor's AI (Chat and Agent modes) to visually brainstorm and iterate on features by generating Mermaid diagrams directly within Markdown files. You'll learn how to use prompts to create flowcharts, sequence diagrams, explore edge cases, and improve the viewing experience for detailed diagrams.

**Workflow/Benefits Demonstrated in this lesson:**

*   Use Cursor AI (Chat/Agent) to generate Mermaid diagrams from natural language prompts.
*   Employ Markdown preview, popping it into a separate window for better viewing and zooming independent of the editor font size.
*   Iteratively refine diagrams by asking the AI to add features or create new diagram types (e.g., sequence diagrams).
*   Generate diagrams to visualize complex scenarios like optimistic updates and edge cases (conflict resolution, offline sync).
*   Create persistent visual artifacts (diagrams in Markdown) for planning, documentation, and guiding further AI development tasks.


## Mermaid Cursor Rules
````markdown
---
description: Whenever diagrams or mermaid is mentioned.
globs: 
alwaysApply: false
---
Here are the essential rules for creating Mermaid diagrams that render well in GitHub Markdown:

1.  **Use the Correct Fenced Code Block:**
    *   Always start the block with ```mermaid and end it with ```.
    *   ```markdown
        ```mermaid
          graph TD
            A[Start] --> B{Decision};
            B -- Yes --> C[Action 1];
            B -- No --> D[Action 2];
        ```
        ```

2.  **Stick to Well-Supported Diagram Types:**
    *   GitHub generally has good support for the most common types:
        *   `graph` (Flowcharts - `TD` is often best for readability)
        *   `sequenceDiagram`
        *   `classDiagram`
        *   `stateDiagram-v2` (Prefer v2 for better features/rendering)
        *   `erDiagram` (Entity Relationship)
        *   `pie` (Pie charts)
        *   `gantt` (Gantt charts)
        *   `mindmap` (Mind Maps - Basic structure only. *See syntax notes below*)
    *   Avoid extremely new or less common diagram types, as GitHub's Mermaid version might lag slightly behind the latest release.

3.  **Keep Syntax Simple and Standard:**
    *   **Node IDs:** Use simple alphanumeric IDs (e.g., `node1`, `processA`, `userAuth`). Avoid spaces or special characters in IDs.
    *   **Node/Actor/State Labels:** **Use quotes (`"..."`)** for labels, especially if they contain spaces, punctuation, special characters, or Mermaid keywords. This is the most common source of rendering errors.
        *   *Good:* `A["User Input"] --> B["Validate Data"];`
        *   *Bad (potential error):* `A[User Input] --> B[Validate Data];`
    *   **Arrows:** Use standard arrow types (`-->`, `---`, `==>`, `->>`, etc.).
    *   **Comments:** Use `%%` for comments if needed.

4.  **Mindmap Specifics for GitHub:**
    *   GitHub **supports the basic `mindmap` structure**, using indentation to define hierarchy (parent/child relationships).
    *   Ensure each node/item is on its **own line** with correct indentation relative to its parent.
    *   GitHub **DOES NOT support** experimental or advanced `mindmap` features like `::icon()` syntax. Using icons **will cause rendering errors**.
    *   Stick to plain text nodes for mind maps intended for GitHub Markdown.
    *   *Correct Example (GitHub compatible):*
        ```mermaid
        mindmap
          Root
            Parent
              Child 1
              Child 2
        ```
    *   *Incorrect Example (GitHub incompatible due to icons):*
        ```mermaid
        mindmap
          Root
            Parent
              ::icon(fa fa-one) Child 1
              ::icon(fa fa-two) Child 2
        ```

5.  **Prefer Vertical Layouts (`graph TD`):**
    *   For flowcharts (`graph`), `TD` (Top Down) or `TB` (Top Bottom) usually renders more readably within the flow of a Markdown document than `LR` (Left Right). GitHub often gives diagrams ample width, but vertical flows are easier to follow on typical screens.

6.  **Let GitHub Handle the Styling:**
    *   **DO NOT** try to set themes (e.g., `%%{init: {'theme': 'dark'}}%%`) or apply custom styling using `classDef` or `style` *within the Mermaid code*.
    *   GitHub **ignores** these theme directives and applies its own styling based on whether the user is viewing GitHub in light, dark, or dimmed mode. Your diagram will automatically adapt. Trying to force colors or themes will likely just be ignored or look out of place.

7.  **Keep Diagrams Focused:**
    *   Avoid overly complex diagrams with excessive nodes, edges, or nesting in a single block. While GitHub *can* render complex ones, they might become hard to read or hit rendering limits. Break down very complex ideas into multiple simpler diagrams if possible.

8.  **Test in GitHub Preview:**
    *   Always use GitHub's preview tab when editing your Markdown file to see exactly how the diagram will render. This is the best way to catch syntax errors or unexpected layout issues.

9.  **Tooling & Edit Verification Note:**
    *   Automated code editing tools may sometimes struggle to apply changes correctly within Mermaid blocks, especially with complex syntax involving indentation or specific formatting (like the mindmap example above).
    *   **Always carefully review edits** made to Mermaid blocks by automated tools. If errors persist or the diagram doesn't render as expected, manual correction might be necessary.

**In short:** Write standard, clearly quoted Mermaid syntax using common diagram types. For mindmaps in GitHub, use only the basic indentation structure (no icons). Prefer `graph TD`. Avoid trying to control the theme/colors yourself. Use the GitHub preview, and double-check automated edits to Mermaid blocks.
````

Brainstorm and Planning with Mermaid Diagrams in Cursor

Manually staging files, writing commit messages, committing, and syncing to Git involves multiple repetitive steps. This lesson demonstrates how to significantly streamline this common Git workflow in Cursor (or VS Code) by creating a single, powerful keyboard shortcut using `keybindings.json` and the `runCommands` feature.

**Workflow demonstrated in this lesson:**

*   Open `keybindings.json` to customize keyboard shortcuts.
*   Unbind default shortcuts (like `Shift+Cmd+S`) that you might not use.
*   Utilize the `runCommands` command to create a sequence of actions.
*   Chain together `git.stageAll`, `cursor.generateGitCommitMessage`, `git.commitAll`, and `git.sync` commands.
*   Bind this entire sequence to a single custom keyboard shortcut.
*   Execute the full stage -> AI commit message -> commit -> sync workflow with one keystroke.

**Key benefits:**

*   Combines multiple Git actions into one efficient step.
*   Automates staging, committing, and syncing changes.
*   Leverages Cursor's AI to generate commit messages automatically.
*   Frees up valuable keyboard shortcuts for powerful custom workflows.
*   Significantly speeds up your development cycle by reducing manual Git operations.

```json
  {
    "key": "shift+cmd+s",
    "command": "runCommands",
    "args": {
      "commands": [
        {
          "command": "git.stageAll"
        },
        {
          "command": "cursor.generateGitCommitMessage"
        },
        {
          "command": "git.commitAll"
        },
        {
          "command": "git.sync"
        }
      ]
    }
  },
  {
    "key": "shift+cmd+s",
    "command": "-workbench.action.files.saveAs"
  },
  {
    "key": "shift+cmd+s",
    "command": "-workbench.action.files.saveLocalFile",
    "when": "remoteFileDialogVisible"
  }
```

Create Complex Cursor Shortcuts to Run Commands and Generate AI Commits

In development we often want to test out features we are building but can only do so with deployed or staging applications or setting up Ngrok or Cloudflare Tunnels. 

UntunJS allows us to get the benefits of Cloudflare tunnels without having to go into the Cloudflare dashboard and set it up.

In this quick lesson we utilize UntunJS to run our local Nextjs application in a Cloudflare Tunnel

Ben Patton

Run local Nextjs App in a Cloudflare Tunnel with UntunJS

UntunJS is a great package to get the quickest start with Cloudflare tunnels and development and testing


Hey everyone 👋,  
  
John Lindquist here with the **fourth issue of AI Dev Essentials!** After months of teaching Cursor workshops, I've really enjoyed iterating on the AI assisted feedback loops: using AIs for spec work, diagramming, task management, and integrating them every step of the development process. I've even built a local MCP server that can monitor dev servers and test runners, checking on their progress to remove my biggest friction points when letting AI agents work autonomously. Expect an egghead.io lesson on it any day now. 🔧✨

🎓 New egghead.io Lesson
------------------------

[AI Driven Design Workflow: Playwright MCP, Screenshots, Visual Diffs & Cursor Rules](https://egghead.io/ai-driven-design-workflow-playwright-mcp-screenshots-visual-diffs-and-cursor-rules~aulxx)(egghead.io)

Walk through an automated loop where Cursor uses Playwright screenshots + Pixelmatch visual diffs to iteratively nudge your UI toward a target design.

🤖 Model Mania & Platform Shifts
--------------------------------

_The model landscape continues to shift rapidly, with major players releasing new contenders and tweaking existing ones based on user feedback and performance goals._

*   **OpenAI Rolls Back “Sycophantic” GPT 4o Update**
    
    OpenAI reverted last week’s GPT 4o tweak after users complained the model became an over eager yes man. The post explains how they detect and curb flattery bias. [Details](https://openai.com/index/sycophancy-in-gpt-4o/)(openai.com)
    
    It's fascinating when AI models get pushed so far by providers that public backlash forces a rollback to a previous state. This serves as a strong reminder that AI is software; it can be reverted, and users grow to expect predictability not just in information, but also in behavior and tone.
    
*   **Alibaba Unveils Open Weight _Qwen3_ Family**
    
    Eight new dense/MoE models (0.6B 235B) land with strong coding & math scores. The 235B flagship rivals closed source giants while staying fully open weight. [Benchmark blog](https://qwenlm.github.io/blog/qwen3/)(qwenlm.github.io)
    
    It's great to see more open source models being released. Personally, I tend to prefer the smartest models available for planning and coding tasks, and often lack the home hardware to run larger open models effectively. Still, these releases fuel hope for powerful on device models eventually rivaling giants like Gemini 2.5 Pro and O3.
    
*   **Gemini API Adds “Reasoning Effort” Knob**
    
    Google’s OpenAI compat endpoint now lets you pick _low → high_ reasoning, making it trivial to swap between Flash speed and deep thought. [Docs](https://ai.google.dev/gemini-api/docs/openai)(ai.google.dev)
    
    The distinction between questions needing quick answers versus deep thought is interesting. It's notable how AI models sometimes seem to first assess if a question requires deep thought before actually processing it, much like humans do. We instantly recall basic trivia but need to search our experiences for more complex answers.
    
*   **Rumor Mill: o3 pro Imminent?**
    
    Sam Altman mentioned an "o3 pro" tier a couple of weeks ago, fueling anticipation that a paid model between o3 and GPT 4o could drop any day now. Keep an eye on the pricing page. [Tweet](https://x.com/koltregaskes/status/1917143793649828320)(x.com)
    
*   **DeepSeek R2 Leak Hints at 1.2T Param Bargain LLM**
    
    Slides suggest a Huawei trained hybrid MoE vision model that’s 97% cheaper per token than GPT 4o, if the specs hold water, brace for price wars. [Leak thread](https://x.com/OpulentByte/status/1916195106580828344)(x.com)
    
    This lands squarely in the rumor mill and hasn't been confirmed. However, DeepSeek R1 was super impressive. If R2 lives up to these leaked specs, it has the potential to significantly disrupt the larger model landscape.
    

🛠️ Dev Tooling, Education & Ecosystem
--------------------------------------

_Beyond the models themselves, the surrounding ecosystem of tools, educational resources, and key personnel moves continues to evolve at pace._

*   **React Creator Jordan Walke Joins Replit**
    
    Walke says Replit’s AI first IDE feels like the early Facebook days and could onboard “the next billion programmers.” [Announcement](https://x.com/jordwalke/status/1915103101205594211)(x.com)
    
    I personally don't use full app generators like Replit, Lovable, or Bolt much, but I always keep a close eye on their announcements to see how far they push the tech. As an experienced developer, I haven't had much luck getting them to build apps where I don't know the final shape, but I can only imagine the feeling for a non programmer launching something from these tools. Definitely curious what the team will build next.
    
*   **Microsoft Drops “Generative AI for Beginners” Series**
    
    Free 18 episode curriculum covers LLM basics, responsible AI, and prompt craft, perfect onboarding material. [Watch](https://learn.microsoft.com/en-us/shows/generative-ai-for-beginners/)(learn.microsoft.com)
    
*   **OpenAI Launches Free “Academy” Courses**
    
    A new hub of AI courses, from intro to deep learning, built with universities like Georgia Tech, all 100% free. [Explore](http://academy.openai.com/)(academy.openai.com)
    
*   **GitHub Lists for Repo Curation**
    
    Use Lists to group related repos into smaller, AI friendly contexts, handy for prompting or onboarding. [Demo](https://x.com/janwilmake/status/1915338073061167487)(x.com) | [Example](https://github.com/janwilmake/forgithub.lists)(github.com)
    
    I didn't even know GitHub Lists was a feature! I'm definitely exploring ways to use AI to bring multiple repos together, since so many products are composed of different repositories. This looks like something fun to play with.
    
*   **Music AI Sandbox Upgrades (Lyria 2)**
    
    DeepMind’s tool now lets musicians generate, extend, and edit tracks with higher fidelity audio, all in browser. [Announcement](https://x.com/GoogleDeepMind/status/1915421048171856047)(x.com)
    
    Google getting into generative music tools for musicians is a welcome surprise. Their Veo 2 video generation is top tier, so you can only imagine the potential if they combine their music and video generation for ads and other media formats.
    
*   **Cloudflare Workers + Vercel AI SDK Starter**
    
    Kristian Freeman shared a template showing Gemini web search via the Vercel AI SDK in a Cloudflare Worker. [GitHub Repo](https://github.com/kristianfreeman/workers-vercel-ai-starter)(github.com)
    
    Adding web search to AI tooling significantly improves answers for many questions, especially in the fast moving dev space. I use deep research queries constantly in Gemini and OpenAI, so building easily launchable custom tools with web search enabled is something I'll definitely explore more soon.
    

🔌 Advancements in Agents & Protocols (MCP/A2A)
-----------------------------------------------

_The push towards more capable AI agents continues, with significant developments in the protocols and tools designed to help them interact with the world and each other._

*   **MCP Demo Day (Tomorrow!)**
    
    Cloudflare hosts demos from Atlassian, Intercom, Linear, PayPal, Sentry, Stripe, Webflow & more this Thursday (May 1st). [Details](https://demo-day.mcp.cloudflare.com/)(demo-day.mcp.cloudflare.com)
    
*   **Cloudflare Agents SDK Update**
    
    v0.0.66 includes a default MCP client, letting agents (and even MCP servers) call other MCP servers. [Details](https://x.com/threepointone/status/1915735083278880950)(x.com)
    
*   **\`mcp-memory\` Tool Released**
    
    An open source MCP server memory layer built on Cloudflare tech to share context between AI clients. [GitHub Repo](https://github.com/Puliczek/mcp-memory)(github.com)
    
*   **Zapier MCP Session (May 7th)**
    
    Learn best practices for leveraging Zapier's 7,000+ integrations via its MCP server in this free session. [Register](https://bit.ly/3XRrqfp)(bit.ly)
    
*   **Google: A2A + MCP Needed**
    
    Google reiterates that complex agentic apps likely need both A2A (agent to agent chat) and MCP (tool/data access). [Blog Post](https://goo.gle/4jSta0l)(goo.gle)
    
*   **Goose Agent Primer**
    
    A quick explainer on Block's Goose AI agent and how it uses MCP for task automation. [Project Site](https://block.github.io/goose)(block.github.io)
    
*   **Motia AI Agent Framework**
    
    Check out this open source framework with examples for deep research using OpenAI & Firecrawl. [Tweet](https://x.com/donvito/status/1915451174670749869)(x.com)
    

🚀 Level Up Your AI Dev Skills: Join My Cursor Workshop!
--------------------------------------------------------

Ready to master practical AI development workflows in Cursor? Join me, John Lindquist, for this hands on workshop! I've been teaching these sessions for months, refining the content, and I'm excited to share my latest insights on Agents, Ask mode, Custom Modes, multi file analysis, effective prompting, Cursor rules, and handling AI failures. Let's conquer the complexity together!

**When:** Thursday, May 8, 2025, 9:00 AM - 2:00 PM (PDT)

**Where:** Zoom (Live Q&A included)

**Investment:** $249

Act Fast! Get the Early Bird price of $200 if you register by midnight tonight (Wednesday, April 30th)!

[Read More](https://egghead.io/workshop/cursor)(egghead.io) | [Register Now ($200 Early Bird)](https://buy.stripe.com/3csdSh2GTdhucTedQW?prefilled_promo_code=SLV2PKOR)(buy.stripe.com)

_(Team training also available)_

⚡ Quick Links / Community Buzz
------------------------------

*   **Cursor Tips for Large Codebases:** Enable project structure, @ mention files, break down tasks, start new chats often, use Ask mode for planning. [Thread](https://x.com/ericzakariasson/status/1915072216523497628)(x.com)
    
    These tips come directly from the Cursor team themselves. Definitely worth reading through the thread if you work with large projects.
    
*   **Building Internal Tools w/ Cursor:** How Eric Zakariasson uses Cursor + Rules + MCP to build tools at Cursor. [Thread](https://x.com/ericzakariasson/status/1914696699274252405)(x.com)
    
    Another great thread from a Cursor employee detailing how they set up their internal configurations. Check this one out too!
    
*   **Gemma 3 OCR Demo:** A local OCR app powered by Google's Gemma 3. [GitHub Repo](https://github.com/patchy631/ai-engineering-hub/tree/main/gemma3-ocr)(github.com)
    
    I love seeing local models being used! Google is building AI directly into Chrome, and it's fascinating to watch how far they can push this. I often use OCR on screenshots for context in queries; the future of productivity might involve AIs constantly reading our screens. Faster OCR is better for all of us!
    
*   **Color Matcher Model:** Fix white balance or match colors for image model outputs. [Replicate](https://replicate.com/fofr/color-matcher)(replicate.com)
*   **o3 "Write in My Style" Prompt:** A two prompt technique for better voice cloning. [ShumerPrompt](https://shumerprompt.com/prompts/o3-write-in-my-style-prompt-017ae28e-c135-4f48-8bc5-8b78785a51d7)(shumerprompt.com)
*   **Open Source Prompt Library:** Conversational prompt templates for "vibe coding." [GitHub Repo](https://github.com/TechNomadCode/Open-Source-Prompt-Library/)(github.com)
*   **Kaggle 5 Day GenAI Course:** A quick intro to GenAI concepts. [Kaggle Learn](https://www.kaggle.com/learn-guide/5-day-genai)(kaggle.com)
*   **Underrated AI Coding Tools?:** Peter Yang asks the community for hidden gems. [Join the discussion](https://x.com/petergyang/status/1915934134507344095)(x.com)
*   **Value at the Application Layer:** Logan Kilpatrick on why now is a great time for builders (falling costs, rising intelligence & willingness to pay). [Clip](https://x.com/tbpn/status/1915110608149889385)(x.com)

That’s the scoop for this issue! The pace continues with new models, platform refinements, and a growing ecosystem of tools and educational resources aimed squarely at developers building with AI.  
  
If you have any feedback or questions, hit reply! Always happy to chat about the latest in AI dev tools.  
  
John Lindquist  
[egghead.io](https://egghead.io/)(egghead.io)

AI Dev Essentials #4: GPT-4o Rollback, Qwen3 Models, New AI Workflow Lesson & MCP Updates

Hey Everyone 👋,

John Lindquist here with the third issue of AI Dev Essentials! Another week, another whirlwind of updates in the AI development space. We saw Google drop Gemini 2.5 Flash, OpenAI counter with o3/o4-mini and a new Codex CLI, Anthropic beef up Claude with research tools, and continued evolution in the MCP and agent framework landscape. Personally, I've been digging into Gemini 2.5 Flash – it's impressive to see Google pushing the performance frontier for smaller models and comparing it directly against the competition like O4-mini. Let's dive in!

___

### New egghead.io Lesson This Week:

* **[Generate and Debug MCP Servers from Scratch using GitHub Examples and AI](https://egghead.io/generate-and-debug-mcp-servers-from-scratch~k98ad):** Building complex components like Model Context Protocol (MCP) servers from scratch can be challenging, especially without good reference implementations. This lesson demonstrates a powerful workflow combining the `ghx` CLI tool (a wrapper I built for the GitHub CLI - find it [here](https://www.npmjs.com/package/@johnlindquist/ghx)) with AI assistants (like Google's AI Studio and Cursor's agent) to overcome this hurdle. Learn how to scrape real-world examples from GitHub using `ghx` ([see scraped examples here](https://gist.github.com/johnlindquist/b19673bbf6a396406dff869beecace1d)), feed them to an AI to generate a new server, and then integrate and debug it within Cursor. It's a practical look at bootstrapping complex tools using AI and targeted code searching.

___

Here are the highlights that caught my eye this past week:

## Model & Major Platform Updates

* **Gemini 2.5 Flash Arrives:** Google released Gemini 2.5 Flash ([Google Developers Blog](https://developers.googleblog.com/en/start-building-with-gemini-25-flash/)), their first unified reasoning model with thinking budgets. It's positioned as a high-performance, cost-effective option. Notably, Google's benchmarks included comparisons against OpenAI's very recent O4-mini, drawing praise for transparency ([Nathan Lambert](https://x.com/natolambert/status/1912958672881303943)). It's available via the Gemini API, AI Studio, Vertex, and the Gemini app. The [Live API docs](https://ai.google.dev/gemini-api/docs/live) are also updated.
* **OpenAI Counters with o3/o4-mini & Codex CLI:** OpenAI launched [o3 and o4-mini](https://openai.com/index/introducing-o3-and-o4-mini/), highlighting their agentic capabilities and tool use during Chain-of-Thought. Alongside these, they released the open-source [Codex CLI](https://github.com/openai/codex), a coding agent designed to leverage these new models. Interestingly, the OpenAI dev behind Codex CLI was spotted [using Cursor](https://x.com/chuhang1122/status/1912786904812294312).
* **Claude Gets Research Superpowers:** Anthropic rolled out a major [Claude upgrade](https://x.com/jowettbrendan/status/1912446262756151732) introducing a "Research" agent capability. It can search the web and internal docs, chain queries, and explore topics agentically, building citations along the way. They also added Google Workspace integration (Gmail, Calendar, Docs) for paid users. (Official blog post link pending confirmation).
* **Cursor 0.49 Update:** The latest Cursor update ([Changelog 0.49](https://www.cursor.com/changelog/0-49)) brings several enhancements: improved Cursor Rules generation and attachment, better chat history access, built-in diff reviews, image support in MCP context, agent command editing/skipping, global ignore files, added models (Gemini 2.5 Pro, Grok 3, GPT-4.1), and project structure context for the agent. [AGIdevice](https://x.com/agidevice/status/1913575514670776520) shows how to enable the beta features.

## Tooling & Protocol Developments

* **MCP Nuances:** David Cramer cautioned against simply [proxying OpenAPI specs via MCP](https://x.com/zeeg/status/1913289979368796671), arguing it misuses the protocol and ignores LLM limitations in tool selection and JSON handling.
* **MCP Transport Clarification:** Harald Kirschner clarified the newer "Streamable HTTP" transport ([MCP Spec](https://modelcontextprotocol.io/specification/2025-03-26/basic/transports)), noting it's primarily HTTP-based, with Server-Sent Events (SSE) as an optional upgrade for streaming server messages, deprecating the older standalone SSE transport. Discussions on implementation and future streaming enhancements are ongoing ([TS SDK Issue #220](https://github.com/modelcontextprotocol/typescript-sdk/issues/220), [MCP Roadmap](https://modelcontextprotocol.io/development/roadmap)).
* **Agent2Agent (A2A) Protocol Announced:** Google unveiled the Agent2Agent (A2A) protocol ([Google Developers Blog](https://developers.googleblog.com/en/a2a-a-new-era-of-agent-interoperability/)), an open standard aiming to allow AI agents from different frameworks and vendors to communicate and collaborate. This is a significant step towards agent interoperability. Akshay Pachaar also provided a clear explanation contrasting [MCP vs. A2A](https://x.com/akshay_pachaar/status/1912123161883345321). ([GitHub Repo](https://github.com/google/A2A))
* **New MCP Tools:** Ishan Goswami released an open-source [Twitter MCP Tool using Exa](https://x.com/TheIshanGoswami/status/1914226695118168331) (GitHub: [exa-labs/exa-mcp-server](https://github.com/exa-labs/exa-mcp-server)) for searching tweets/profiles without the official API. Eyal Toledano shipped [TaskmasterAI v0.12](https://x.com/EyalToledano/status/1914084622918455572) with multi-PRD parsing, smarter breakdowns, and easier MCP setup, alongside a new [website](http://task-master.dev).
* **Cursor Workflows & Tips:** Ryo Lu shared [12 tips for using Cursor effectively](https://x.com/ryolu_/status/1914384195138511142), emphasizing clear rules, specific prompts, and iterative testing. Prajwal Tomar detailed his ["Context Boundary Method"](https://x.com/PrajwalTomar_/status/1913961327711633601) using structured docs (PRD, Flow, Tech Stack etc.) to guide AI coding tools like Cursor and reduce errors. Andrew Jefferson noted potential issues with [Markdown tab completion](https://x.com/EastlondonDev/status/1913967886521053688) in Cursor.
* **AI UI Builders Compared:** Damien shared a comparison of [Replit, Bolt.new, Lovable.dev, and v0](https://x.com/damienghader/status/1913248328302608482) for generating a multi-step onboarding flow from the same prompt.

## Building AI Agents: Frameworks & Best Practices

* **Agent Framework Discussion:** Harrison Chase (LangChain) published a deep dive on [how to think about agent frameworks](https://blog.langchain.dev/how-to-think-about-agent-frameworks/), sparking discussion comparing approaches (agents vs. workflows, declarative vs. non-declarative) and frameworks like LangGraph, OpenAI's Agents SDK, CrewAI, and AutoGen. Key themes include balancing flexibility vs. predictability and choosing the right abstraction level. OpenAI also released its own [practical guide to building agents](https://cdn.openai.com/business-guides-and-resources/a-practical-guide-to-building-agents.pdf).
* **Production Agent Principles:** Dex Horthy shared principles for production-grade LLM apps in ["12-factor-Agents"](https://github.com/humanlayer/12-factor-agents), arguing successful systems often use focused LLM loops within well-engineered software rather than fully autonomous, monolithic agents. This resonated with many builders facing reliability challenges.
* **Agent Memory:** Taranjeet from [mem0.ai](https://x.com/taranjeetio/status/1912185506559053855) emphasized the importance of persistent memory (beyond session history) for truly intelligent agents that learn and adapt over time ([Blog Post](https://mem0.ai/blog/memory-in-agents-what-why-and-how/)).
* **Key Best Practices:** Common advice emerging from various discussions includes: start simple (maybe not even an agent), use structured outputs (like JSON schemas) to reduce hallucinations, prefer multiple specialized agents over one complex agent, own your control flow rather than relying solely on the framework, and carefully manage context windows. ([Anthropic Guide](https://www.anthropic.com/research/building-effective-agents), [Dev.to Summary](https://dev.to/isaachagoel/read-this-before-building-ai-agents-lessons-from-the-trenches-333i)).

___

**✨ Workshop Spotlight: Conquer the Complexity of Cursor ✨**

Master practical AI development workflows in Cursor. This hands-on workshop covers Agents, Ask, Custom Modes, multi-file analysis, effective prompting, Cursor rules, and strategies for handling AI failures.

* **When:** **This Thursday!** April 24, 2025, **9:00 AM - 2:00 PM (PDT)**
* **Where:** Zoom (Live Q&A included)
* **Investment:** $249

Unlock Cursor's potential and streamline your development process.

➡️ [**Read More about the Workshop**](https://egghead.io/workshop/cursor) ➡️ [**Register Now ($249)**](https://buy.stripe.com/9AQaG5ftF5P22eAfZ3)

_(Team training also available)_

___

## Notable Prompts

* **o3 Fact-Checking:** A prompt structure for using o3 to thoroughly fact-check an entire article by parsing individual facts. (See [Matt Shumer's Tweet](https://x.com/mattshumer_/status/1914412812891570199) for the prompt start).
* **Multiple Experts Method:** A system prompt that instructs the AI to analyze problems from multiple expert perspectives (e.g., Systems Thinker, Risk Assessor) and synthesize their insights. (See [God of Prompt's Tweet](https://x.com/godofprompt/status/1914370848032497906) for the full prompt and example).
* **Personality-Infused Prompt:** A detailed persona prompt instructing the AI to adopt a specific, informal, and slightly critical tone using modern slang and abbreviations. (See [Kitze's Tweet](https://x.com/thekitze/status/1913719205007937825) for the full prompt, originally from [@threepointone](https://x.com/threepointone)).

## Community Buzz & Resources

* **New Community Site:** [AGIdevice](https://x.com/agidevice/status/1913666033870647325) is building [cursorcommunity.com](http://cursorcommunity.com) in public.
* **Industry Chatter:** Deirdre Bosa discussed the potential [OpenAI bid for Windsurf](https://x.com/dee_bosa/status/1912906057791447294) and the shift towards the app layer. AI Native Dev reported on perceived [friction between Microsoft and Cursor](https://x.com/ainativedev/status/1912869556973949430) ([Article](http://ainativedev.co/spx)).
* **Android XR Demo:** Google showed off prototype smart glasses with Android XR ([Demo Video via 9to5Google](https://9to5google.com/2025/04/17/google-android-xr-smart-glasses-live-demo/)).
* **New Open Dialogue Model:** A new small open model, [Dia-1.6B](https://x.com/altryne/status/1914421817139454341), was released. It's focused on dialogue disentanglement – figuring out who's talking to whom in complex conversations. ([GitHub](https://github.com/nari-labs/dia), [Hugging Face](https://huggingface.co/nari-labs/Dia-1.6B)). I was extremely impressed with the demos – definitely worth checking out.
* **Video Generation with Veo 2:** Google's Veo 2 video generation model is now available in Gemini Advanced and Google AI Studio ([Google Blog](https://blog.google/products/gemini/video-generation/), [Developers Blog](https://developers.googleblog.com/en/veo-2-video-generation-now-generally-available/)). I've tried out a few videos and was really impressed with the quality. Pro tip: In AI Studio, you can use an image as a reference when generating video ([API Docs](https://ai.google.dev/gemini-api/docs/video#generate_from_images)), though surprisingly, Gemini doesn't support that yet. It's heavily rate-limited at the moment, but still fun to play with for generating short 8-second clips.

___

That’s the scoop for this issue! The pace isn't slowing down, particularly around agent capabilities, specialized models, and the tooling ecosystem supporting them.

If you have any feedback or questions, hit reply! Always happy to chat about the latest in AI dev tools.

John Lindquist


AI Dev Essentials #3: Gemini Flash, o4-mini, Codex CLI, A2A Protocol & Agent Updates

AI Dev Essentials covers Gemini 2.5, O4-mini, Claude’s research tools, Codex CLI, and new workflows for building AI agents with MCP and Cursor.

Establish an automated, AI-driven workflow for implementing UI designs by comparing live development versions against target screenshots. This lesson integrates the Playwright MCP server for browser automation, a custom script using Pixelmatch for visual diffing, and Cursor Rules for persistent context. See how to instruct the Cursor Agent to iteratively implement design changes based on visual diff feedback, creating a robust loop where the AI refines the UI towards the target design.

**Workflow demonstrated in this lesson:**

*   Configure the Playwright MCP server within Cursor for browser control.
*   Use the Agent and Playwright MCP tools (`browser_navigate`, `browser_resize`, `browser_take_screenshot`) to capture baseline and current state PNG screenshots.
*   Install `pixelmatch` and `pngjs` dependencies.
*   Guide the Agent to create and run a TypeScript script (`compare-images.ts`) to generate visual diff images between screenshots.
*   Establish Cursor Rules (`visual-testing.mdc`, `project-setup.mdc`, etc.) to provide the Agent with necessary context about directories, scripts, dependencies, and the overall workflow.
*   Initiate an AI-driven design implementation loop:
    *   Agent captures the current state via Playwright.
    *   Agent runs the comparison script to get a visual diff.
    *   Agent analyzes the diff and implements code changes in the project (e.g., `app/page.tsx`, components, `globals.css`).
    *   Repeat the capture-compare-implement cycle, using the latest diff as feedback.
*   Use context files and prompts like "Please continue" to manage the multi-step process and guide the AI through potential interruptions.

**Key benefits:**

*   Creates an iterative, AI-driven workflow for implementing designs based on visual comparisons.
*   Combines browser automation (Playwright) and pixel-level diffing (Pixelmatch) within the Cursor environment.
*   Utilizes Cursor Rules to effectively manage context and guide the AI through complex, multi-step tasks.
*   Demonstrates how visual diffs can serve as direct feedback for AI-driven UI code generation and refinement.

Search

Search Results