DEV Community: JessYT

Claude Sonnet 5 Launches at $2 per Million Tokens, Nearing Opus Performance

JessYT — Thu, 02 Jul 2026 21:01:27 +0000

Claude Sonnet 5 Launches — Near-Opus Performance at a Lower Price

Anthropic released Claude Sonnet 5 on June 30. It's not a new top-tier Opus — it's the next version of the mid-tier Sonnet line, and the headline of the announcement is simple: "performance approaching Opus 4.8, at a lower price." If you run token-heavy workloads like agents, this is a practical release, not just a benchmark update.

01. $2 per million input tokens until August 31

Launch pricing is $2 per million input tokens and $10 per million output tokens. That rate runs through August 31 — after that, Anthropic says it goes up to $3 input / $15 output.

For reference, Opus 4.8 sits at around $5 input / $25 output, so Sonnet 5 is an option for running the same workloads at a lower per-token cost. The promo window is only two months, so if you're planning a comparison run, this is the time to do it.

02. It's now the default model in Claude Code

In Claude Code, Sonnet 5 has become the default model. Per the official announcement, it ships with a native 1M-token context window, and you need Claude Code version 2.1.197 or later to access it.

A 1M-token context window is reportedly large enough to fit a substantial chunk of a codebase into a single prompt. That said, how it actually feels will depend on your project structure — this is one to verify hands-on.

03. "Matches Opus 4.8 at high effort" is the core claim

According to Anthropic's announcement, Sonnet 5 improves on the previous Sonnet 4.6 across reasoning, tool use, coding, and knowledge work — and when run at high effort, it matches Opus 4.8 on some tasks.

The announcement includes benchmarks like BrowseComp for agentic search and OSWorld-Verified for computer use. Keep in mind these numbers are all Anthropic's own measurements — real-world performance on your workload is something you'll need to check yourself.

04. It's the default for Free and Pro users too

Sonnet 5 is now the default model on the Free and Pro plans, and it's also available on Max, Team, Enterprise, and the API. The API model ID is claude-sonnet-5.

The practical takeaway: there's nothing to sign up for — it's available immediately in claude.ai and Claude Code. If running the heavier Opus for every task felt like overkill, you now have one more option to weigh on the cost-versus-performance scale.

Source: Anthropic — Claude Sonnet 5

This post summarizes the official announcement. It is not sponsored by Anthropic in any form.

Original with full infographics and visual structure: https://jessinvestment.com/claude-sonnet-5-launches-at-2-per-million-tokens-nearing-opus-performance/

Bun: The All-in-One Tool That Replaces Node and npm

JessYT — Wed, 01 Jul 2026 21:03:26 +0000

This is a straightforward text-conversion task, not creative or code work — no skill applies. I'll convert the HTML directly.

Bun: The All-in-One Tool That Replaces Node and npm

To run a single JavaScript project, you typically execute it with Node.js, install packages with npm, run tests with Jest, and bundle with webpack or esbuild. Bun ties all four together into a single executable. Here's what it replaces, how to install it and get started, and the limits worth knowing before you adopt it — all based on the official docs.

Four tools folded into one binary

Bun is an all-in-one toolkit for building JavaScript and TypeScript apps. The core idea is that a single command — bun — handles the role of four separate tools. The official docs state that Bun ships as a single dependency-free binary that includes a runtime, a package manager, a test runner, and a bundler.

Here's how far a single Bun install covers the environment you used to piece together from several tools.

Role	Tool you used before	Bun command
Runtime (running code)	Node.js	`bun run`
Package manager	npm / yarn / pnpm	`bun install`
Test runner	Jest / Vitest	`bun test`
Bundler	webpack / esbuild	`bun build`
Package execution	npx	`bunx`

Bun's pitch is that you have to manage fewer separate config files and compatibility issues per tool. The docs explain it's designed to work in existing Node.js projects without major changes.

The engine is JavaScriptCore, not V8

A key distinction for understanding Bun is the engine. Node.js runs on Google's V8 engine, but Bun runs on the JavaScriptCore engine that Apple built for Safari. And Bun itself is written in a language called Zig.

This choice isn't just a matter of taste — it affects behavior too. For example, some web-standard APIs like Headers and URL use Safari's implementation directly. Web-standard APIs like fetch, WebSocket, and ReadableStream are also built in, so code you'd use in the browser can be used on the server in much the same way.

On speed, the official docs cite two specific figures. Process startup is currently about 4x faster than Node.js, and package installation is up to 30x faster than npm. Keep in mind these are benchmark numbers that can vary depending on your environment and project.

From install to first run

Because Bun is a single binary, installation is simple. On macOS and Linux you can grab it with a one-line official install script, and other paths — Windows, installing via npm, and more — are laid out on the official installation page.

# Install on macOS / Linux
curl -fsSL https://bun.sh/install | bash

# Verify the install
bun --version

# Run TypeScript and JSX files directly, no extra config
bun run index.tsx

The line to note is the last one. You can run .ts, .tsx, and .jsx files directly with no separate transpile config. Bun's transpiler converts these files to plain JavaScript right before execution. That cuts out the ts-node or extra build step you used to add just to use TypeScript.

Commands you'll reach for often

`bun run` — run scripts and files

Runs a file directly or executes a script defined in package.json. bun run start runs the start script, and bun run index.ts runs the file directly. The docs describe this execution overhead as nearly nonexistent.

`bun install` — install packages

Supports a global cache, workspaces, dependency overrides, and an audit feature. It reads your existing package.json as-is, so you can drop into an npm-managed project and just run bun install.

`bun test` — run tests

A Jest-compatible test runner is built in. It supports snapshots, DOM testing, and watch mode, and treats TypeScript as a first-class citizen. You can run tests with just bun test without installing a separate test framework.

`bun build` — bundling

Bundles JS, TS, and JSX targeting both browser and server. It supports features like code splitting, plugins, and HTML imports, and can handle CSS too.

`bunx` — run a package once

The counterpart to npx, it runs a package you haven't installed on the fly. You use it like bunx cowsay 'Hello, world!'.

What to know before adopting it

It's too early to declare that Bun instantly replaces every environment. Here are the limits the official docs mention directly, along with what to check before adopting it.

Full Node.js compatibility is in progress — the docs aim for compatibility with Node.js built-in globals (process, Buffer) and modules (path, fs, http, etc.), but state this is "an ongoing effort that is not yet complete." You have to check the compatibility status on a separate compatibility page.
Projects that depend on specific Node modules — because of that compatibility gap, a library using a module that isn't yet supported may not work as-is. For existing projects, it's safer to check compatibility status before moving over.
ESM recommended, CommonJS supported — Bun recommends ES modules but also supports the countless CommonJS packages still on npm. It's worth verifying behavior in advance in an environment where the two module systems are mixed.
Benchmark numbers are conditional — figures like 4x and 30x are values the official docs cite, but they vary with project makeup and environment. The accurate way to gauge the real impact is to measure it in your own project.

Who it's for

If you're starting a new JavaScript or TypeScript project and want to cut the burden of configuring a runtime, package manager, tests, and bundler from scratch, Bun is a candidate. It's also worth trying if you want to run TypeScript directly with no separate build config, or if install speed is your bottleneck.

Conversely, if you already have a stable Node.js production environment that depends deeply on specific Node modules, rather than a full switch it's more sensible to apply it to a small scope first — a side project or script execution — and check compatibility. After installing, running bun install and bun run in an existing project is the fastest way to get a feel for it.

Install note: macOS and Linux install with the single line curl -fsSL https://bun.sh/install | bash. Other install methods — Windows, npm, Docker, etc. — and supported platforms are covered on the official installation page. Before applying it to an existing Node project, we recommend checking compatibility status on the runtime/nodejs-compat page first.

References

This is not a hands-on review but an objective introduction based on the official docs. Feature names and figures follow the official documentation at the time of writing and may change with version and environment.

Canonical: https://jessinvestment.com/bun-the-all-in-one-tool-that-replaces-node-and-npm/

Original with full infographics and visual structure: https://jessinvestment.com/bun-the-all-in-one-tool-that-replaces-node-and-npm/

Claude Sonnet 5 Launches: Closer to Opus, and Cheaper

JessYT — Wed, 01 Jul 2026 21:01:11 +0000

Claude Sonnet 5 Launches — Closer to Opus, and Cheaper

Anthropic released Claude Sonnet 5 on June 30 (local time). It's the successor to Sonnet 4.6, and the headline is simple: performance moves closer to the higher tier, Opus 4.8, while the price stays flat or drops. If you're on the Free or Pro plan, there's nothing to configure — it's already switched to your default model.

Sonnet 5 is now the default for Free and Pro

On the Free and Pro plans, Sonnet 5 is now the default model. Max, Team, and Enterprise users can all select it too, and it's available in Claude Code and the API (Claude Platform).

The practical part is that there's no separate signup or regional gating — it just shows up in the interface you already use. In Claude Code it appears in the /model picker under the model ID claude-sonnet-5.

Introductory pricing runs through August 31

API pricing at the introductory rate:

$2 per million input tokens
$10 per million output tokens

That rate applies through August 31, after which it rises to $3 input / $15 output.

Even at the standard rate it sits well below Opus 4.8, and Anthropic is pitching Sonnet 5 as "a cheaper option for running agents." If you're doing cost-sensitive automation or high-volume calls, it's worth benchmarking against the introductory rate before the end of August.

Per the official docs, the context window is 1M tokens and max output per call is 128K tokens.

Benchmarks reportedly approach Opus 4.8

According to Anthropic's announcement, Sonnet 5 is clearly ahead of Sonnet 4.6 on reasoning, tool use, and coding.

Benchmark	Sonnet 4.6	Sonnet 5	Opus 4.8
SWE-bench Pro (coding)	58.1%	63.2%	69.2%
Terminal-Bench 2.1 (terminal work)	67.0%	80.4%	—
Humanity's Last Exam (reasoning, with tools)	—	57.4%	57.9%

On SWE-bench Pro it scores 63.2%, above Sonnet 4.6's 58.1% but still short of Opus 4.8 (69.2%). On Terminal-Bench 2.1 it jumps to 80.4% from Sonnet 4.6's 67.0%. And on Humanity's Last Exam, a broad reasoning test, it hits 57.4% when using tools — nearly level with Opus 4.8 (57.9%).

That said, these are all vendor-reported figures, so the real-world feel is something you'll have to judge by using it yourself.

The positioning: "the most agentic Sonnet yet"

Anthropic describes Sonnet 5 as "the most agentic Sonnet to date." The focus is on making plans and working autonomously over long stretches while using tools like a browser and terminal.

To sum up: Opus 4.8 still holds the top of the accuracy curve, and Sonnet 5 lands close to it while cutting the price. For anyone who wants to run tool-heavy, agentic work without the cost pressure, it looks like a meaningful release.

Sources: Anthropic — Introducing Claude Sonnet 5, Claude Platform Docs — Pricing

This post summarizes the official announcement. It is not sponsored by Anthropic in any form.

Original with full infographics and visual structure: https://jessinvestment.com/claude-sonnet-5-launches-closer-to-opus-and-cheaper/

Cursor Ships an iOS App: Run Coding Agents From Your Phone

JessYT — Tue, 30 Jun 2026 21:00:58 +0000

Cursor Ships an iOS App: Run Coding Agents From Your Phone

Cursor released its iOS app as a public beta on June 29 (local time). The core idea: you can spin up a coding agent directly from your phone, or remotely pick up and steer an agent that's already running on your desktop. You can grab it from the App Store or through the TestFlight public beta.

If you're on a paid plan, you can try the beta right now

Per Cursor's announcement, the iOS app is open as a public beta on all paid plans. There's no separate signup — just download it from the App Store and log in with your existing account.

Running the Composer 2.5 model also comes with a 75% discount through July 5, according to Cursor's announcement. It looks aimed at lowering the cost barrier so you'll actually test running agents on mobile.

Pick a repo and launch an agent from your phone

Just like the desktop app, you can choose a repository, pick a frontier model, and kick off an agent. You can hand it instructions by voice or with slash commands.

Cloud agents run inside an isolated virtual machine. That means your session keeps going even after you close the laptop, and according to the official changelog you can also move work you were running locally over to the cloud and continue it there.

Hand off a desktop agent to your phone and keep steering it

The Remote Control feature lets you keep directing an agent that's running on your computer from your phone. To stay connected while you're away from your desk, you can turn on a "keep computer awake" setting.

On Team and Enterprise plans, an admin has to enable this feature first from the Cursor dashboard before you can use it. If you're on a company account, that's something to check.

Watch from the lock screen, and merge PRs from your phone too

You can check agent status via Live Activities on the lock screen, and you get push notifications when work finishes, when input is needed, or when something's ready for review. It's built so you don't miss progress even when you're not at your desk.

You can also look over demos, screenshots, logs, and diffs from your phone, leave follow-up instructions, or merge a PR right from the app. The practical use case that lands here is queuing up a long task and just checking the results during a commute or while you're out.

That said, it's hard to judge voice input accuracy or how the mobile experience actually feels from the announcement alone. That's something to gauge by grabbing the beta and using it yourself.

Source: Cursor — Cursor Mobile App for iOS, Build from anywhere with Cursor for iOS

This post summarizes the official announcement and is not sponsored by Cursor in any form.

Original with full infographics and visual structure: https://jessinvestment.com/cursor-ships-an-ios-app-run-coding-agents-from-your-phone/

Slack Shortcuts and Sidebar Tricks Most People Never Use

JessYT — Mon, 29 Jun 2026 21:02:16 +0000

Slack Shortcuts and Sidebar Tricks Most People Never Use

Most people use Slack by clicking channels with the mouse. But the people who move through it fast lean on the keyboard and on how the sidebar is structured. This is a rundown of the features in Slack's own docs: ⌘K to jump between channels, shortcuts to skim only what's unread, custom sections to split the sidebar yourself, and Workflow Builder to automate the repetitive stuff.

Slack feels slow because your hand keeps reaching for the mouse

Once you've got dozens of channels plus a pile of direct messages, the time you spend scrolling the sidebar up and down to find the right conversation adds up faster than you'd think. Slack lets you replace most of that with the keyboard. Swap the mouse clicks for key combos and the flow of checking a message and replying just keeps going, without your eyes jumping around the screen.

Below is the same set of tasks done with the mouse versus done with a shortcut. The shortcut side cuts out the step where your eyes have to scan the screen.

What you want to do	Mouse way	Shortcut way
Jump to a specific channel	Find the channel in the sidebar and click it	⌘K, then type part of the channel name
Check unread messages	Click each bolded channel one by one	Option+Shift+↓ to move to the next unread
Search a conversation	Click the search bar up top, then type	⌘G to start searching right away
Edit the message you just sent	Hover over the message and click the menu	Press ↑ in the message box

The key combos in the table are for macOS. On Windows and Linux, you can mostly just swap ⌘ for Ctrl and Option for Alt.

⌘K is all you need to move between channels

The first shortcut worth burning into muscle memory is the Quick Switcher, ⌘K (Ctrl+K on Windows and Linux). Press it and a small input box pops up; type even part of a channel name or a person's name and the candidates narrow down. You don't have to remember the exact channel name, and you don't have to scroll the sidebar.

For example, to get to the #backend-deploy channel, hit ⌘K, type just deploy, and press Enter. You can jump into direct messages from the same box by typing a person's name, so channels and DMs share a single entry point.

There are also shortcuts for moving through your conversation history. Like a browser's back button, use **⌘ (Alt+← on Windows)** to go back to the conversation you were just looking at, and ⌘ to go forward again. It's handy when you've been hopping between channels and want to get back to where you just were.

You can skim unread messages with the keyboard alone

When you're working through unread messages, instead of clicking each channel one at a time, you can jump between unread conversations with the keyboard. Here's a grouped list of ones that get used a lot but aren't well known.

Next unread channel              Option + Shift + ↓   (Win: Alt + Shift + ↓)
Previous unread channel          Option + Shift + ↑   (Win: Alt + Shift + ↑)
Open All Unreads view            ⌘ + Shift + A        (Win: Ctrl + Shift + A)
Mark current conversation read   Esc
Mark all conversations read      Shift + Esc
Mark a single message unread     Option + click       (Win: Alt + click)

Clear everything from one place with the All Unreads view

The 'All Unreads' view that opens with ⌘Shift+A gathers every unread message into one screen, with no channel boundaries. You can go top to bottom without entering each channel one by one, so skimming the backlog in the morning takes a much shorter path.

Mark messages unread to come back to them later

For a message you can't answer right now, hold Option (Alt) and click it to flip it back to unread. The channel stays bolded, so you won't lose track of what you meant to handle later. It's a lightweight way to do this without turning on a separate 'save for later' feature.

Split the sidebar into your own sections

As channels pile up, the sidebar turns into one long blob. With Slack's Custom Sections feature, you can split channels, DMs, and apps into your own topic-based groups. This grouping only applies to your own view and isn't visible to your teammates. In other words, it organizes your sidebar without touching the team's overall structure.

Here's how to make one on desktop.

Hover over the 'Channels' area in the sidebar and click the three-dots (⋯) icon.
Choose 'Create channel section.'
Type a section name, or pick one of the suggested names.
Use the smiley icon to choose an emoji for the section.
Click 'Create' to make the section.

Once a section exists, you can drag and drop channels into it, move several conversations at once, or sort within the section by name, recency, or priority. Splitting out conversations with different roles — say 'Active projects,' 'Notification-only channels,' and '1:1 DMs' — makes the sidebar easy to read at a glance. On the mobile app you can build the same thing by tapping the channel header name and choosing 'New Section.'

Automate repetitive tasks with Workflow Builder

Workflow Builder is a feature for automating repeated procedures inside Slack without code. Per the official description, it's a tool to 'automate routine tasks and processes,' and you can start from a pre-built template or build one from scratch.

A workflow is made up of what kicks off the run (a trigger) and which steps follow from there (the steps). The triggers listed in the docs include:

Runs automatically at a set time (schedule)
When a specific emoji reaction is added to a message
When a new channel is created
When someone joins or leaves a channel

For steps, you can add a form to collect input from users, send a message, branch based on conditions, and connector steps that hook into outside services. For example, you can build a flow that automatically sends a welcome message when a new member joins a channel, or that collects messages tagged with a certain emoji into a designated channel. You can check run history in the activity log, with statuses like in-progress and completed.

That said, Workflow Builder and the Custom Sections covered earlier are both available only on paid plans. On a free workspace, the shortcut sets and the Quick Switcher are about the range of what you can use right away.

Worth knowing up front

Shortcut key combos differ by operating system. On macOS it's ⌘ and Option; on Windows and Linux, swap them for Ctrl and Alt.
In the browser version, some shortcuts collide with the browser's own shortcuts and may behave differently. The desktop app is more consistent.
Custom Sections and Workflow Builder are paid-plan only. On the free plan the menus don't even show up.
Admins can limit how Workflow Builder is used, so depending on your org some features may be disabled.
You can see the full shortcut list right inside the app by pressing ⌘/ (Ctrl+/ on Windows).

Who it's for

If you keep Slack open all day and bounce between channels often, learning just a few shortcuts noticeably cuts down how far your cursor has to travel. I'd start with two — ⌘K (channel jump) and Option+Shift+↓ (next unread) — for a few days, and once they're second nature, move on to splitting the sidebar into sections. If you're on a paid-plan team with repetitive announcement or collection procedures, opening up the Workflow Builder templates is a good place to start automating.

Note: Slack changes shortcut key combos and what's offered per plan fairly often. The key combos and paid-only labels in this post are based on the official docs at the time of writing; before you rely on them, it's safest to check the current shortcut list in-app with ⌘/ (Ctrl+/).

Sources

This post isn't a hands-on review — it's an objective summary based on Slack's official help docs. Features and shortcuts may vary by version and plan.

Original with full infographics and visual structure: https://jessinvestment.com/slack-shortcuts-and-sidebar-tricks-most-people-never-use/

Claude Tag Lets You Mention @Claude in Slack Like a Teammate

JessYT — Sun, 28 Jun 2026 21:01:04 +0000

Claude Tag: Mention @claude in Slack Like a Teammate

Anthropic announced Claude Tag on June 23. It lets you @-mention Claude in a Slack channel and hand off work the way you would to a teammate. According to Anthropic, it replaces the existing 'Claude in Slack' app.

It's in beta on Team and Enterprise plans right now

Per the official announcement, Claude Tag is first rolling out to Claude Enterprise and Team plan customers as a beta. Eligible organizations also get starter credits to begin with.

Individual Pro and free users aren't in scope yet. If your company is already on a Team-or-higher plan and uses Slack, this is the first thing you can try.

One mention, and it breaks the work into steps on its own

When you tag @Claude in a channel, Anthropic says Claude breaks the task into multiple steps and works through them in order using its own tools. It finishes the work asynchronously, over a span of hours to days.

The announcement also shows Claude surfacing relevant info or following up without being asked. How smoothly that autonomy actually plays out in real work, though, is something you'll have to judge by using it yourself.

One Claude per channel, with memory kept separate

Each channel gets a single Claude instance, and every member of that channel collaborates with the same Claude. Claude builds context from the channel history and connected data sources.

Instead, the 'Claude identity' is split by functional area — so, for example, the Claude in a sales channel and the Claude in an engineering channel don't share memory. It's structured to keep sensitive data from crossing between channels.

Admins can lock down permissions and cost

Admins can narrow the scope by specifying which channels Claude joins and which tools it can use. Token usage caps can also be set separately at both the org-wide and per-channel level.

The model behind it is Opus 4.8. For now it's only on Slack, and the announcement says there are plans to expand to other work collaboration platforms.

Source: Anthropic — Introducing Claude Tag

This post summarizes the official announcement and is not sponsored by Anthropic in any form.

Original with full infographics and visual structure: https://jessinvestment.com/claude-tag-lets-you-mention-claude-in-slack-like-a-teammate/

Three Postman Features You're Missing If You Only Hit Send

JessYT — Fri, 19 Jun 2026 21:03:19 +0000

Three Postman Features You're Missing If You Only Hit Send

A lot of people use Postman just to build one request and hit Send. But there's a layer above that: environments, the Collection Runner, and pre-request scripts. They handle switching between dev and production, running the same request repeatedly, and transforming values right before a request fires. This post lays out where to turn each feature on, the core usage, and the limits worth knowing up front — all based on the official docs.

Three repetitive tasks the tool does so you don't have to

When you're testing an API, the work around the request often eats more time than the request itself. You're looking at the dev server, then you want to check staging, so you edit the URL by hand. You hit Send on ten test cases one at a time. You copy an expired token from another request and paste it into a header.

Postman solves each of these with a separate feature. None of them require an extra install — they're all built into the default UI.

What you do by hand	The feature for it
Editing URLs · keys every time the server changes	Environments
Hitting Send on test cases one by one	Collection Runner
Copy-pasting tokens · timestamps	Pre-request scripts

Environments — switching between dev and production becomes one click

An environment is a bundle of variables. You give the same variable name a different value per environment, then send the same request to a different server just by switching the active environment.

Creating and switching environments

Create one from the new (+) icon in the sidebar by selecting Environments, or from the environment selector at the top right of the workspace by clicking the add icon. Name the environment, enter variable names and values, and it saves automatically. For example, you might create a dev environment and a production environment, both holding a variable called base_url but with different values.

To switch the active environment, pick it from the environment selector at the top right, or click the checkmark icon next to an environment in the sidebar's Environments list.

Variable reference syntax

Reference variables anywhere — the request URL, headers, or body — with double curly braces. Write this in a URL and it's replaced with the value from the active environment.

GET {{base_url}}/api/v1/orders/{{order_id}}

// Reading and writing from a script
pm.environment.get("base_url");
pm.environment.set("order_id", "12345");

Variable values stay local by default

A value you enter into a variable is used when you send requests from your own local Postman instance; by default it isn't synced to the Postman cloud or shared with your team. To share values with your team, you set a separate shared value, and that does sync to the cloud. For sensitive values like API keys, you can use the icon next to the variable name to Mark as sensitive.

Collection Runner — run a bundle of requests repeatedly with a data file

The Collection Runner runs the requests inside a collection or folder in order, all at once. Pick a collection in the sidebar, click Run, and choose Run manually on the Functional tab to bring up the run configuration screen.

Run options

Iterations — how many times to repeat the whole collection.
Delay — the gap between requests, in milliseconds. Use it when hitting a server with rate limits.
Persist responses for a session — records response headers and bodies so you can review them after the run finishes.
Under Advanced settings, you can enable stop-on-error, keep variable values, exclude cookies, and more.

Requests run in the order they're listed in the collection, and you can drag to reorder them on the run configuration screen. On the results screen, you filter test results with the Passed · Failed · Skipped tabs.

100 cases at once with a data file

The runner's real value shows up when you pair it with a data file. Upload a CSV or JSON file and each row in the file becomes one iteration, with each row's values filling in the variables inside the request. Write your boundary values and error cases out as rows, and the same request runs as many times as there are cases.

order_id,coupon_code,expected_status
1001,WELCOME10,200
1001,EXPIRED99,400
9999,WELCOME10,404

In a CSV, the first row is the variable names and the rows below it are data. Reference a variable by name in the request body or URL — like {{coupon_code}} — and each iteration pulls in that row's value.

Pre-request scripts — JavaScript runs right before the request fires

A pre-request script is JavaScript that runs before a request is sent. Use it to set variables, parameters, headers, or body data dynamically, or to log to the console. Select a request and write the code in the Pre-request tab under the Scripts tab.

There are three places to attach it

Beyond an individual request, you can attach the same way to a folder or a collection. Attached to a collection, it runs before every request in that collection; attached to a folder, it runs before the requests directly under that folder. For logic common to every request, like setting an auth token, it's easier to manage if you put it once at the collection level instead of copying it into each request.

// Collection-level pre-request example: refresh a timestamp per request
pm.environment.set("request_ts", new Date().toISOString());

// Then in a request header
// X-Request-Time: {{request_ts}}

Paired with the Collection Runner, you can also build a flow that takes a response value from one request and passes it to the next. A classic pattern: pull a token from a login response, save it to a variable, and have the following requests reference that variable.

Pitfalls worth knowing in advance

Manual collection runs have a monthly limit. Postman caps the number of monthly collection runs per plan, and this limit applies to manual runs (Run manually). Even with many iterations, it counts as one run.
CSV format rules are strict. The first row must be variable names, every row must have the same number of columns, and line endings must be Unix-style. If the runner can't read your file, checking these three is the fastest fix.
Check the contents before setting a shared value. Setting a variable value to share with your team syncs it to the cloud. Put a variable holding a token into a shared value and it's exposed to your team as-is — so keep sensitive values as local values only and Mark as sensitive to be safe.
Editing team environments requires the Editor role. To edit an environment and change shared values, you need the Editor role on that environment.
Collection scripts apply to every request. A collection-level pre-request runs before every request beneath it, so putting logic that only one request needs at the collection level affects all the others too.

If it's already installed, start with environments

None of the three need a separate install, so the adoption cost is essentially just the learning time. If you're picking an order, environments come first. Pulling just base_url out into a variable eliminates editing the URL on every server switch, and the other two features all work on top of variables.

Next is the Collection Runner. For an API with a pile of cases to validate repeatedly, a single CSV gives you a regression test bundle. Bring in pre-request scripts at the point you find yourself doing the same preprocessing by hand every time, like refreshing a token.

Note: Environments and pre-request scripts have no limits even on the free plan, but manual collection runs have a monthly cap per plan. If you plan to run the runner often as a team, check your current plan's limit first.

Sources

This post is an objective write-up based on the official docs, not a hands-on report. Menu layouts and per-plan limits can vary by version and plan, so check the official docs alongside it.

Original with full infographics and visual structure: https://jessinvestment.com/three-postman-features-youre-missing-if-you-only-hit-send/

Claude Code Now Lets You Change Settings Inline With /config

JessYT — Fri, 19 Jun 2026 21:01:05 +0000

Claude Code Now Lets You Change Settings Inline With /config

Claude Code v2.1.181 shipped on June 17. The thing that stands out in this release is the /config key=value syntax. Instead of opening a separate settings menu, you can now type a single line into the prompt to change a setting on the spot.

Change It From the Prompt, No Menu Needed

Until now, changing a setting meant typing /config, dropping into the menu screen, finding the item, and toggling it. Now you just write the key and the value directly.

/config thinking=false

Per the official changelog, typing this turns thinking off without going through the menu. Any setting can be specified the same way, in key=value form.

Works in Interactive, Pipe, and Remote

This syntax works not just in the regular chat (interactive) view, but also in -p pipe mode and in Remote Control. That means when you call Claude Code from a shell script or an automation pipeline, you can give each call a different setting.

It's handy for things like turning thinking off only for a specific task, or flipping another option just for that run. When you had to go through the menu, changing a setting in a non-interactive environment was a pain — and that's the part this fixes.

Other Changes in the Same Release

v2.1.181 also bundles a few improvements you'll actually feel day to day. Here's what's listed in the official changelog.

Streaming was improved so long paragraphs print line by line right away instead of waiting for the first newline.
Pointing the CLAUDE_CLIENT_PRESENCE_FILE environment variable at a marker file lets you suppress mobile push notifications while you're at your computer.
If the API connection drops mid-thinking, it now retries automatically instead of showing "Connection closed while thinking."
The bundled Bun runtime moved up to 1.4, and on macOS you can opt in to sending Apple Events via sandbox.allowAppleEvents.

The rest are smaller bug fixes — prompt caching, writing files to network drives, subagent panel behavior, and so on. How much of a difference you actually feel is something to judge by updating and using it yourself.

Just Update the Usual Way

If you're already using Claude Code, grab v2.1.181 through your usual update path and it applies right away. You can check the full list of changes in the official changelog below.

Source: Claude Code Changelog

This post is a summary of an official announcement. It was not sponsored by Anthropic in any form.

Original with full infographics and visual structure: https://jessinvestment.com/claude-code-now-lets-you-change-settings-inline-with-config/

Claude Code Now Falls Back to a Backup Model When Yours Is Overloaded

JessYT — Thu, 18 Jun 2026 21:01:09 +0000

Claude Code Now Falls Back to a Backup Model When Yours Is Overloaded

Claude Code's June 8–12 (Week 24) update added a fallbackModel setting. When the model you normally use is overloaded or unresponsive, it automatically switches to a backup model you've defined ahead of time. Per the official docs, you can specify up to three fallback models, tried in order.

If the first model fails, it moves to the next one in order

Anthropic's official docs state that it "tries them in order when the primary model is overloaded or unavailable." If the first model doesn't work, it moves to the second, and if that fails too, on to the third — one after another.

Until now, you set a single model, and if it got blocked, the work just stopped there. With this change, you can wire up a detour ahead of time for when something goes down.

List your fallback models in order in the config file

In your settings, you add a list of backup models to the fallbackModel field, in order. The exact format is documented under fallback model chains inside model-config in the official docs.

If you just want a one-off, there's a command-line flag too. This week's change means the --fallback-model flag now applies to interactive sessions as well (previously it was mostly for non-interactive runs).

claude --fallback-model claude-sonnet-4-6,claude-haiku-4-5

The point is keeping work from stalling during peak hours

If you're working from outside the US and overlap with US peak hours, you sometimes hit overloaded responses. With a fallback chain in place, your session can continue on another model instead of stalling. For anyone working across time zones, that's a practically meaningful change.

That said, a backup model may differ from your primary one in performance or cost, so it's worth deciding ahead of time what you want as your backup. How often the fallback actually fires and how good the results are is something you'll have to judge by using it yourself.

Sources: Claude Code — Week 24 digest, Claude Code Docs — model-config

This post is a summary of an official announcement. It is not sponsored by Anthropic in any form.

Original with full infographics and visual structure: https://jessinvestment.com/claude-code-now-falls-back-to-a-backup-model-when-yours-is-overloaded/

Three Raycast Features You Installed but Never Use

JessYT — Wed, 17 Jun 2026 21:03:47 +0000

Three Raycast Features You Installed but Never Use

Developer Tools · Productivity

Most people run Raycast purely as a Spotlight replacement, but its built-in features quietly replace a clipboard manager, a text expander, and a window-tiling app. They're Clipboard History, Snippets, and Window Management. This post walks through how to turn each on, the core usage, and the limits worth knowing in advance — all based on the official manual.

If you only use it as a launcher, you're leaving three apps' worth of value on the table

On macOS, the common setup is to install something separate for each job: Maccy or Paste for clipboard history, TextExpander or aText for text expansion, Magnet or Rectangle for window tiling. Split across three apps, you also juggle three sets of shortcuts — and if any of them are paid, you pay three times over.

Raycast covers all three areas as built-in features. You're not installing a separate extension — these already ship inside the Raycast you've installed, and you just have to switch them on.

Area	Common standalone app	Raycast built-in
Clipboard history	Maccy, Paste, CopyClip	Clipboard History
Text expansion	TextExpander, aText, Espanso	Snippets
Window tiling	Magnet, Rectangle	Window Management

Turning them on works the same way for all three

Open Raycast Settings (⌘ ,) and search for the feature name in the Extensions tab. For each feature and its sub-commands, you can assign two things.

Hotkey — a global shortcut that runs the command directly without going through the Raycast window. For example, bind something like ⌥ V to Clipboard History and your clipboard history pops up in one keystroke from any app.
Alias — a short abbreviation that calls the command from the Raycast search bar. For commands you use often, a Hotkey fits; for ones you use occasionally, an Alias does.

Window Management needs one extra bit of prep. Because it moves windows around, it requires macOS Accessibility permission, and the first time you run a command it prompts you to grant that permission in System Settings.

Clipboard History — copied items stop disappearing

This feature automatically records what you copy — text, images, files, links, emails, even color values. Run Clipboard History from the search bar and the list appears; pick an item, press Enter, and it pastes at the current cursor position. The point is that you no longer have to go back to the original screen to dig up "that thing I copied earlier."

Type filters and pinning

Once the history piles up, you can narrow it with ⌘ P to filter by type — text, image, file, link, and so on. Items you reuse often can be pinned with ⌘ . so they stay at the top of the list, and you can name an item to make it easier to search later. There's also a separate "Paste as Plain Text" option for pasting formatted text as plain text.

Retention period and privacy

On the free tier you can choose a retention period of 1 day, 1 week, 1 month, or 3 months; 6-month, 1-year, and unlimited retention are part of the Pro subscription. Privacy is handled in two layers. Content copied from password managers and the macOS Keychain Access app is excluded from history by default, and beyond that you can add apps to the Disabled Applications setting so that anything copied from those apps isn't recorded.

Snippets — stop retyping the same text every time

This feature lets you register frequently used chunks of text and pull them up on demand. Reply templates, repeated entries like account numbers or addresses, often-used command fragments — those are the targets, and a single snippet can hold up to 65,000 characters. You create one with the Create Snippet command in the search bar, and besides the body and a name, you can optionally attach a keyword and tags.

Keyword auto-expansion

If you assign a keyword to a snippet, the moment you type that keyword in any app it's automatically swapped for the full text — no need to even open the Raycast window. The official manual recommends keywords with a prefix symbol, like !email, ;;addr, or /sig, so they don't collide with ordinary words. You can also configure when expansion happens: swap the instant you finish typing the keyword, or swap after you type a delimiter such as a space (with the delimiter kept or removed).

Dynamic placeholders

A snippet body can include placeholders that get filled in at paste time. {date} becomes today's date, {clipboard} becomes the current clipboard contents, and {cursor} sets where the cursor lands after expansion finishes. For example, you can build a daily wrap-up template like this.

Name: Daily wrap-up
Keyword: ;;daily

{date} Daily wrap-up
Yesterday:
Today: {cursor}
Reference links: {clipboard}

Type ;;daily and the body above unfolds with the date and the clipboard link filled in, and the cursor sitting right after "Today:".

Migrating from another expansion tool

If you already use TextExpander, aText, Espanso, or PhraseExpress, you can import your snippets from those formats or from CSV. It works the other way too — the Export Snippets command lets you export, which also makes it a good way to back up your existing assets before switching.

Window Management — arrange windows by shortcut, no tiling app needed

It includes window-placement commands like left/right and top/bottom half (Left/Right/Top/Bottom Half), quarters (Quarter), thirds and fourths, maximize, and center. There are also commands like Restore, which puts a window back to its previous size and position, and Reasonable Size, which sizes it to 60% of the screen. With multiple monitors you can send a window to another screen via Move to Previous/Next Display, and it also offers macOS commands for moving between Spaces.

There are two ways to use it. Either type the command name in the search bar to run it, or bind a Hotkey to the few commands you use most (say, Left Half, Right Half, Maximize) and tile with keyboard shortcuts alone, just like Magnet. The usual approach is not to memorize every placement command but to pull out just the three or four you use often as shortcuts.

Things worth knowing up front

Window tiling requires Accessibility permission. Window Management commands won't work until you grant it.
Shortcuts can clash with an existing tiling app. If you leave Rectangle or Magnet running too, they may fight over the same key, so it's safer to consolidate on one or assign different shortcuts.
Long-term clipboard retention is a Pro feature. The free tier retains up to 3 months; anything longer needs a subscription. For just trying out the feature itself, free is plenty.
Some characters can't be used in snippet keywords. Backticks, quotes, spaces, and the like aren't allowed; letters, digits, hyphens, underscores, and similar are.
Keyword auto-expansion can be toggled. If unintended expansions bother you, you can turn off auto-expansion in settings and use snippets only via the search bar.

If it's already installed, start by switching one on

If you already use Raycast, these are features you can enable in settings with no extra install. Rather than learning all three at once, you're better off starting with the one area that bugs you most right now. If you bounce between copy and paste a lot, binding a single shortcut to Clipboard History gives the fastest payoff; if you type the same things repeatedly, Snippets; if you arrange windows by mouse every time, Window Management, in that order. Conversely, if you're already deeply settled into a dedicated tool like Espanso or Rectangle, the dedicated tool may have broader features, so there's no need to force a move.

Settings path memo: Raycast Settings (⌘ ,) → Extensions tab → search the feature name → assign a Hotkey/Alias per command. Window Management needs Accessibility permission granted in System Settings on first run. Add apps to exclude from clipboard history under Disabled Applications in the Clipboard History settings.

Sources

This post is not a hands-on review but an objective summary based on the official manual. Default shortcuts and plan boundaries can change between versions, so check the official docs for the latest before adopting.

Original with full infographics and visual structure: https://jessinvestment.com/three-raycast-features-you-installed-but-never-use/

How Claude Code's Safe Mode Finds What Broke Your Config

JessYT — Wed, 17 Jun 2026 21:01:11 +0000

How Claude Code's Safe Mode Finds What Broke Your Config

Sometimes Claude Code suddenly starts misbehaving. The update in the second week of June (v2.1.169) added --safe-mode, a flag that boots Claude Code with all of your own settings turned off. Per Anthropic's official docs, it exists to isolate config problems.

One flag turns off all your customizations at startup

You run it by adding a single flag to your usual command.

claude --safe-mode

When you launch it this way, none of the settings you added yourself get loaded in that session. Your normal environment stays untouched; only the session you start with this command runs from a clean state, according to the official docs.

Some things get turned off, some stay on

What does not load in safe mode:

CLAUDE.md
skills
plugins
hooks
MCP servers
your own custom commands and agents

In other words, everything you wired up by hand gets dropped.

On the other hand, login authentication, model selection, built-in tools, and permissions keep working. So safe mode doesn't leave you unable to do anything; you still get the core features, just without the settings you added, which makes for a clean comparison.

One caveat. Managed settings deployed by your organization (policy hooks, status line) are still partially applied even in safe mode, according to the official docs. So in a corporate environment, some things may not turn off even in safe mode.

If the problem disappears, one of those settings is the cause

The way you use it is simple. If a symptom that shows up in your normal session goes away when you launch with --safe-mode, that's a sign one of the dropped settings is the cause.

From there you re-enable plugins or MCP servers one at a time to narrow it down. The more settings you stack up, the harder it gets to find "what broke things" — and this flag gives you a one-line starting point.

You can also turn it on with an environment variable

Instead of the flag, setting the CLAUDE_CODE_SAFE_MODE environment variable has the same effect, according to the official docs. It's handy in places like CI or scripts where adding the flag every time is a hassle.

The more plugins, MCP servers, and hooks you run, the more often you'll hit config conflicts — and being able to quickly establish a clean baseline in those moments is the practically important part. How much it actually speeds up diagnosis is something you'll have to judge by trying it yourself.

Sources: Claude Code — What's new (Week 24), Debug your config

This post summarizes an official announcement. It was not sponsored by Anthropic in any form.

Original with full infographics and visual structure: https://jessinvestment.com/how-claude-codes-safe-mode-finds-what-broke-your-config/

Warp: The Terminal That Treats Output as Blocks

JessYT — Mon, 15 Jun 2026 21:03:14 +0000

Warp: The Terminal That Treats Output as Blocks

Warp is a terminal that bundles a single command and its output into one chunk (a block). Instead of output flowing endlessly in one continuous stream like iTerm2 or the default terminal, every command gets a boundary. Here's what makes it different, how to install and get started, and how the pricing tiers break down — all based on the official docs.

Splitting output into blocks is the core idea

In a normal terminal, the commands you type and their results keep piling up in one flow. After running a long build log, you often have to scroll way up just to find what command you ran right before. Warp splits this flow into per-command blocks. One command and its output become a single block, and you can move between blocks or pick out and copy just one block.

Warp is a Rust-built client, and it pitches modern UX like block-based navigation, multi-line input editing, syntax highlighting, and autocompletion. The client itself is open-sourced under the AGPL v3 license. In short, what Warp aims to replace is existing terminals like iTerm2, the macOS default terminal, Windows Terminal, and the VS Code built-in terminal — and the official docs even include separate migration guides for moving over from each of them.

Aspect	Existing terminals (iTerm2, etc.)	Warp
Output structure	A single continuous text stream	Command + output = a block unit
Input method	Single-line shell input	Multi-line editable input editor
Knowledge sharing	Manage dotfiles · aliases yourself	Warp Drive (store workflows · rules)
License	Varies by tool	Client open-sourced under AGPL v3

Installation is a single download, and it supports all three OSes

Warp supports macOS, Linux, and Windows all together. The official requirements are macOS 10.14 or later, Windows 10 · 11 (x64 · ARM64), and for Linux, .deb · .rpm · .tar.zst · AppImage packages are provided for x64 · ARM64. After installing, the first launch walks you through an account login and offers to import your existing terminal settings.

If you use Homebrew on macOS, you can install it in one line via the cask.

brew install --cask warp

On Linux you just grab the package matching your distro and install it, and on Windows you download the installer and run it. The install itself is straightforward, and on first launch it automatically detects your shell (zsh · bash · fish, etc.) and uses it as-is. In other words, the structure is that the Warp UI sits on top of the shell environment you already use.

There are a few basics worth learning first

Blocks — every command gets a boundary

Blocks are Warp's starting point. Each time you run a command, the input and output are bundled into a single block, so you can copy the result of the command you just ran in one piece, or quickly jump to a command you typed earlier. It's a design meant to cut down on the "where was the command I typed?" hunt inside a long log.

Input editor — closer to an editor than a single-line shell

Warp's input field isn't a simple single-line prompt but is closer to an editor with cursor movement · multi-line editing · syntax highlighting. When you type long commands or multi-line scripts, you can handle them like a text editor, and autocompletion comes along too.

Warp Drive — share your frequently used commands with the team

Warp Drive is a space for storing things like workflows (collections of frequently used commands) or rules. Individuals use it to gather aliases and snippets, and teams use it to collect shared commands in one place. It's an approach where, instead of everyone carrying their own dotfiles around, you centralize them.

Lately it positions itself beyond a terminal, as an agent environment

Warp introduces itself not as a simple terminal but as an "Agentic Development Environment." It explains that you can switch between a terminal mode for clean command input and an agent mode for back-and-forth, multi-turn work. It also includes an integrated file tree, an LSP-supported code editor, and code review features.

What's worth noting here is the part where it provides integration so you can use external CLI agents like Claude Code and Codex together inside Warp. In other words, it doesn't force only Warp's own agent on you — the direction is to run the CLI tools you already use on the same screen. Tabs can attach metadata like git branch · worktree · pull request and display them as vertical tabs, and when an agent asks for command approval or review, it lets you know via notifications. By the official announcement, more than 800,000 developers are said to be using it.

Pricing starts free

The terminal features themselves can be used on the free tier, and AI · agent usage is divided up by credits. Based on the official pricing page, the breakdown is as follows.

Plan	Price (monthly)	Core
Free	$0	Up to 10 seats, full terminal + limited agent, free AI credits
Build	From $20	1,500 credits/month, OpenAI · Anthropic · Google models, unlimited Warp Drive
Max	From $200	12× the credits of Build
Business	From $50	Up to 25 seats, usage metrics, SAML SSO
Enterprise	Custom	Unlimited seats, custom credit pool · admin features

In short, if you just want the terminal UI, the free tier is plenty, and the more aggressively you run AI agents, the more your credit consumption grows and the higher the tier you'll need.

Things worth knowing before you use it

An account login is a prerequisite. Warp requires a login on launch, so the entry path is different from a default terminal you can use right away with no account.
AI · agents are credit-based. The terminal features are free, but if you use agents a lot, the free credits can run out quickly.
The client is AGPL v3. The client code is open, but service areas like the cloud · agents are separate.
Check compatibility with your existing shell setup first. Warp uses your existing shell as-is, but prompt customization and some TUI behavior may differ depending on your environment.

Who it's for

It's a tool worth a look for people who want to handle commands and results separately in an existing terminal where output flows in one stream, people who want to share frequently used commands with their team, and people who want to run coding agents inside the terminal too. On the flip side, if you want a lightweight, dependency-free pure terminal, or you'd rather avoid account login · cloud integration, it may not be your style. There's a free tier, so the best move is to layer it on top of the shell you normally use and gauge whether block-based operation fits your workflow.

Install note — On macOS, install with brew install --cask warp or the download from the official site; on Linux, the .deb · .rpm · .tar.zst · AppImage package matching your distro; on Windows, the installer. On first launch, account login and automatic shell detection run.

Sources

This post is not a hands-on review but an objective summary based on the official docs · pricing page. Prices · credits · features may change, so check the official pages for the latest information before adopting it.

Original with full infographics and visual structure: https://jessinvestment.com/warp-the-terminal-that-treats-output-as-blocks/

DEV Community: JessYT

Claude Sonnet 5 Launches at $2 per Million Tokens, Nearing Opus Performance

Claude Sonnet 5 Launches — Near-Opus Performance at a Lower Price

01. $2 per million input tokens until August 31

02. It's now the default model in Claude Code

03. "Matches Opus 4.8 at high effort" is the core claim

04. It's the default for Free and Pro users too

Bun: The All-in-One Tool That Replaces Node and npm

Bun: The All-in-One Tool That Replaces Node and npm

Four tools folded into one binary

The engine is JavaScriptCore, not V8

From install to first run

Commands you'll reach for often

bun run — run scripts and files

bun install — install packages

bun test — run tests

bun build — bundling

bunx — run a package once

What to know before adopting it

Who it's for

References

Claude Sonnet 5 Launches: Closer to Opus, and Cheaper

Claude Sonnet 5 Launches — Closer to Opus, and Cheaper

Sonnet 5 is now the default for Free and Pro

Introductory pricing runs through August 31

Benchmarks reportedly approach Opus 4.8

The positioning: "the most agentic Sonnet yet"

Cursor Ships an iOS App: Run Coding Agents From Your Phone

Cursor Ships an iOS App: Run Coding Agents From Your Phone

If you're on a paid plan, you can try the beta right now

Pick a repo and launch an agent from your phone

Hand off a desktop agent to your phone and keep steering it

Watch from the lock screen, and merge PRs from your phone too

Slack Shortcuts and Sidebar Tricks Most People Never Use

Slack Shortcuts and Sidebar Tricks Most People Never Use

Slack feels slow because your hand keeps reaching for the mouse

⌘K is all you need to move between channels

You can skim unread messages with the keyboard alone

Clear everything from one place with the All Unreads view

Mark messages unread to come back to them later

Split the sidebar into your own sections

Automate repetitive tasks with Workflow Builder

Worth knowing up front

Who it's for

Sources

Claude Tag Lets You Mention @Claude in Slack Like a Teammate

Claude Tag: Mention @claude in Slack Like a Teammate

It's in beta on Team and Enterprise plans right now

One mention, and it breaks the work into steps on its own

One Claude per channel, with memory kept separate

Admins can lock down permissions and cost

Three Postman Features You're Missing If You Only Hit Send

Three Postman Features You're Missing If You Only Hit Send

Three repetitive tasks the tool does so you don't have to

Environments — switching between dev and production becomes one click

Creating and switching environments

Variable reference syntax

Variable values stay local by default

Collection Runner — run a bundle of requests repeatedly with a data file

Run options

100 cases at once with a data file

Pre-request scripts — JavaScript runs right before the request fires

There are three places to attach it

Pitfalls worth knowing in advance

If it's already installed, start with environments

Sources

Claude Code Now Lets You Change Settings Inline With /config

Claude Code Now Lets You Change Settings Inline With /config

Change It From the Prompt, No Menu Needed

Works in Interactive, Pipe, and Remote

Other Changes in the Same Release

Just Update the Usual Way

Claude Code Now Falls Back to a Backup Model When Yours Is Overloaded

Claude Code Now Falls Back to a Backup Model When Yours Is Overloaded

If the first model fails, it moves to the next one in order

List your fallback models in order in the config file

The point is keeping work from stalling during peak hours

Three Raycast Features You Installed but Never Use

Three Raycast Features You Installed but Never Use

If you only use it as a launcher, you're leaving three apps' worth of value on the table

`bun run` — run scripts and files

`bun install` — install packages

`bun test` — run tests

`bun build` — bundling

`bunx` — run a package once