Another take on the problem of fetch() sometimes returning a response and sometimes throwing an error: what if it always returned an object containing either response or error? https://github.com/human...

safe-fetch 1.1.0 has been released! ✨ Errors now serialize as JSON in Response body Details: https://github.com/human...

Crosspost 1.0.2 has been released! 🔧 Replaced dotenv with native process.loadEnvFile for env handling Details: https://github.com/human...

Did you know that fetch() can reject errors? Network errors or AbortSignals cause errors, so you never get a Response object and need to handle these separately. That's annoying so I created Safe Fetch, a wrapper that always returns a Response: https://github.com/human...

This is super weird. When I test Bredbox locally, it's able to correctly fetch and extract meta data from a YouTube URL. Once deployed to Cloudflare, I get nothing. (thinking emoji)

I currently have three VS Code windows open: 1. My local file system 2. My WSL instance 3. A remote server via SSH It's nearly impossible to tell the difference between them. Amazing work by the VS Code team.

If you look closely, you may notice someone familiar in the day 2 keynote at GitHub Universe. youtu.be/q1IxyisKcZI?...

I've switched my go-to models now: Claude Haiku 4.5 - general coding, implementation, incremental changes GPT-5-Codex - designing a feature, debugging difficult problems

It's almost time for my monthly newsletter. What should I write about?

Pro tip: The GitHub MCP server has 94 tools. You probably don't need all of them. Enable just the ones you need to avoid confusing the LLM.

Doctor: Your LDL is high. I think it’s time for a statin. Me: I’d like to try adjusting my diet to first. Him: We can try but I don’t think it will come down enough. Me:

Temporarily paused sending out new Bredbox invites after a user reported a large Pocket file failed to import. Got everything figured out and invites are going out again. If you loved Pocket, sign up for a Bredbox invite: https://bredbox.app

While I generally enjoy using Cloudflare infrastructure, the worker subrequests limit is a pain to work with. Very little visibility into what counts as a subrequest and the limits aren’t enforced locally so it’s a huge time sink to debug “too many subrequests” errors.

Excited to see VS Code Copilot also reading in CLAUDE.md for instructions. 🎉

React Email is such a well-thought-out project. Absolute joy to use and my emails already look better.

Pro tip: Did you accidentally `rm` a file you had open in this VS Code session? Open the command palette --> "Local History: Find Entry to Restore". You're welcome.

One of the hidden benefits of ESLint going multi-language is pulling in a new group of contributors. Since releasing the JSON, CSS, and Markdown plugins, we had over a dozen new consistent contributors and four became team members.

I’ve been saying this for a while: any linter that wants to meaningfully compete with ESLint needs to support ESLint plugins. First Deno lint, now Oxlint: voidzero.dev/posts/announ...

I still use ChatGPT for most non-coding AI stuff. I feel like it always give the right level of detail. With Claude, it's like that friend who you ask what time it is and they tell you the history of timepieces.

Shout out to @nicr.dev for creating this utility to bulk update npm packages to trusted publishing: https://github.com/nicol...

Pretty excited that full-text search just shipped on Bredbox! Feature parity with Pocket for managing bookmarks is getting very close. Only major feature left to implement is highlights. Still invite-only while I'm building it out, invites sent weekly. Join the waiting list: https://bredbox.app

If you want your open source project to grow, you need to get comfortable with other people reviewing and merging pull requests without your involvement. This didn't come naturally to me and still feels weird, but there's no other way to keep up project velocity.

Early thoughts on Claude Haiku 4.5: It seems just as fast as GPT-4.1 but with more agency and more capacity to understand complex situations. It has less of a tendency to go off on tangents than Sonnet 4.0 but still does from time to time.

I am quickly approaching the need to declare GitHub notification bankruptcy. Can't get under 100 on a daily basis. 😓

My current biggest time sink is Supabase's lack of true down migrations. `migration down` deletes all data, which essentially is like doing `db reset`.

Latest VS Code update breaks my world. Copilot chat doesn't work in WSL. zsh terminal doesn't seem to be working in WSL either, it just hangs on every command.

Twitter just renewed my annual Premium Pro plan without any advance notice for nearly $400. It's over 3x more expensive than last year and no way I would've knowingly paid that. 😠

Crosspost 1.0.1 has been released! 🐞 Downgraded dotenv to fix console.log breaking MCP output Details: https://github.com/human...

Mentoss 0.12.0 has been released! ✨ called() method now throws error if no routes match Details: https://github.com/human...

Crosspost v1.0.0 released! Only breaking change is dropping support for Node.js 18.x. This release also adds support for the Nostr protocol. https://github.com/human...

Claude's tendency to decide that a whole file is wrong so it should be deleted and recreated from scratch is the most troubling behavior I've experienced with AI. Replace "file" with "civilization" and you understand why some are terrified of where AI is heading.

I absolutely love that Cloudflare refuses to deploy a worker if the required secrets are not in production. Such a productivity boost.

Pretty excited to launch collections on Bredbox! Getting very close to feature parity with Pocket for managing bookmarks. Next batch of invites going out soon.

GitHub really needs to provide some way for us to update multiple npm packages at once. For those of us with over a decade of packages to wade through, manually checking each one's publishing settings is time-consuming and prone to missing something.

In hindsight, getting a laptop with non-expandable RAM was a mistake.

GitHub's response to npm supply chain attacks is too one-sided. By reinforcing maintainer responsibility, they're missing ingestion-point detection. For example, the registry could require packages that add a postinstall script for the first time to have a semver major bump.

ESLint depends on sponsorships to pay maintainers. We're operating at a deficit with roughly a junior engineer's salary supporting a dozen maintainers plus awarding outside contributions. Please talk to your company about sponsoring ESLint: https://eslint.org/donate

ESLint v10.0.0 is coming! Check out our latest blog post for what to expect. https://eslint.org/blog/...

🤝 The Open Source Pledge = $2,000/year per developer to OSS maintainers. For enterprises using open source, supporting these projects isn't optional—it's essential for mitigating risks and ensuring sustainability. Join us → opensourcepledge.com

Due to an unprecedented amout of AI traffic coming from China to our website, we are temporarily blocking all requests coming from China until we can deploy a more targeted solution. We apologize for the inconvenience and appreciate your understanding.

Tried using Grok Code Fast 1 as my Copilot implementer and it did a pretty good job. A bit too reliant on the command line, which slowed things down, and did skip a couple steps. The code it did right was spot-on.

Do any of my followers work at Netlify? The ESLint websites are seeing a massive spike in bandwidth usage starting September 25 without a corresponding increase in traffic (via Google Analytics). Can anyone help us figure this out?

My dream AI IDE supports Claude Code-like subagents that I can arrange in a LangChain-like chain to fully define my development process.

First day using an Insta360 Link 2 as my primary webcam and so far I'm impressed. I also like that it looks and acts a bit like a little droid sitting on top of my monitor.

Pro tip: Sometimes the thing doesn't work because it's broken.

For Bredbox, I've been trying to avoid site-specific configs to help with content extraction due to the maintenance overheard. But man, content sites do some truly bizarre things that I'm not sure how to generalize.

Web context extraction challenge: incorrect/missing Open Graph data. If og:type="article", it's definitely an article, but many sites set og:type"website" on all pages, including articles. So og:type "website" is useless for this purpose.

A bunch of Bredbox invites went out over the past week. If you're on the waiting list, be sure to check your spam folder!

For some reason, Bredbox wasn't extracting the article content from @jakearchibald.com's blog. This turned out to be an edge case bug that is now fixed, so more sites can be extracted. 🎉

Just sent out the next batch of invites to Bredbox! Had to fix a few bugs that popped up with the first batch of users before I felt comfortable about letting more in. Sign up for your invite here: https://bredbox.app