I don't understand the people who hate Claude Code adding attribution markers to commits.

It's actually extremely valuable for open source to see when someone's contribution was done by Claude, it isn't an advertisement, it's just honesty ¯\_(ツ)_/¯

I feel like people should definitely be using Claude Code's sandbox mode, and I'm a bit shocked I rarely see people discussing it.

without the tedium of needing to figure out the exact Docker commands or YML config options to use via trial-and-error got pushes. And *that* has been absolutely massive.

The main improvements have just been from avoiding unnecessary layer rebuilds for docker containers, using a lot of caching, reducing image sizes, avoiding rebuilding images entirely if things haven't changed, etc. None of it is groundbreaking, but Claude enables me to do it a lot faster

Claude also can't seem to write Terraform files to save its life, so the infra folks have been crucial as well, because I'm also incompetent with Terraform :)

A ton of the work is only possible because I have a very good understanding of the CI pipeline itself, but Claude being able to get the Docker command syntax / CI YML syntax right on the first try is a *massive* improvement over how I used to do this. It's like night and day, it's completely insane.

I've been using Claude Code to help improve our CI pipelines at Employer, and along with some excellent work by the infra team, we've gotten the CI runtime from 15-18 minutes down to to 8-12 minutes.

I need someone to solve all the problems with git hooks for me, thanks

Which parts are laborious right now? I've tried to make the migration tool very accurate with help from the other maintainers over the last 4-5 months. Would be happy to hear about any problems that still exist :)

Includes Oxlint/tsgolint, Oxfmt, Vite, and more :) I'm very glad to see it become a fully open source tool!

Oxlint JS Plugins Alpha A collection of high-performance JavaScript tools written in Rust

Oxlint JS Plugins alpha has been released :) I helped a little bit with some docs stuff, kudos to the Oxc team on many months of work!

This is compatible with ESLint's plugin/rule API and a *lot* of plugins will work out of the box with it. Very stable alpha at this point.

oxc.rs/blog/2026-03...

I hate the orphans so much

We've working on it in small pieces over the last few months, it was probably 10 or 12 total merge requests of adjusting settings and upgrading a few packages that misbehaved on TS7. Only one was particularly huge/difficult. The final upgrade MR was only 2 file changes outside of package.json

Claude: The script needs a fundamental redesign - it should work per-vec, not per-file, to avoid mixing types. Let me rewrite it with a Python script for the actual transformation logic. Me: you are NOT allowed to use python Claude: Got it, no Python. Let me rewrite the script using only shell and perl. First let me revert everything and start fresh. Me: NO PERL EITHER

how do I stop Claude from trying to write Python scripts, I'll pay extra

A post by the OpenAI Developers twitter account from 1 hour ago, it says "Codex Security is now also available on ChatGPT Pro accounts."

Anthropic: This is incredibly dangerous in the wrong hands, so we are releasing Claude for Security to trusted partners only until we have enough safeguards.
OpenAI: We have invented the Torment Nexus and you can all have it for free for a month with your $20 subscription!!!!

Got TypeScript 7 merged for the work codebase!!

Thank you to @jakebailey.dev and the TS team for all the work on this. Went from a 35s typecheck to 8s on an ~11k file React app 🥳

Good. Don't stand down, Anthropic. www.anthropic.com/news/stateme...

Review it like you would a PR from Linus Torvalds

Please report any bugs you've found so that we can fix them 🙏 github.com/oxc-project/...

@ericmbudd.com undefined

"if you accept arbitrary regexes you can make the system slow" is such a funny class of "vulnerability", like... yeah, no shit.

JS Plugins A collection of high-performance JavaScript tools written in Rust

Depending on the specific custom rules you've got, JS Plugins in Oxlint may be usable without much trouble :) oxc.rs/docs/guide/u...

Also see oxc.rs/docs/guide/u...

Proof of Concept: Script for comparing JS-generated upstream test cases by connorshea · Pull Request #19394 · oxc-project/oxc DISCLAIMER: Fully written by Claude Code / Opus 4.6. I'm opening this now because I think it's really useful and wanted to share it before I had to hop off for the day. I have not reviewed ...

I have forced Claude to build a proof of concept for this that should make it a lot easier to find dynamically-generated tests from the original rules :) github.com/oxc-project/...

test(linter): Add more tests for the `eslint/no-empty-function` rule. by connorshea · Pull Request #19392 · oxc-project/oxc This ports various tests from the original rule, all were added upstream in eslint/eslint@dcd95aa Most of them pass, but a few needed to be commented out. All of this work was done manually, no Cla...

I already went through and did this work via automation for a lot of rules, but our rulegen tooling doesn't support retrieving tests that are generated dynamically, so I know for certain that some are missing cases still.

For example: github.com/oxc-project/...

If anyone wants to give me tons of credits to use for LLM nonsense, I think running one in a loop to find Oxlint rules that are missing test cases from the original plugins would be neat.

LLMs seem to absolutely love generating GitHub Actions configs that use versions that are multiple major releases behind, like @actions/checkout v4 instead of v6. The GitHub team should probably be proactive about mitigating this somehow...

Sometimes I think semantic versioning is worse for software than not using it. Like, it'd sometimes be better to have smaller, incremental jumps that break things rather than major version bumps that have 20 breaking changes all at once. You *can* do this with semantic versioning, but no one does.

Building a C compiler with a team of parallel Claudes Anthropic is an AI safety and research company that's working to build reliable, interpretable, and steerable AI systems.

Prediction for this year: AWS earns $10 trillion in 2026 as Claude rewrites everything in Rust. In January 2027, we run out of things to rewrite and the economy collapses entirely. www.anthropic.com/engineering/...

I dunno about underhyped, but a lot of things are definitely about to change.

The software you can create with these models has beaten by expectations repeatedly, as has unfortunately the image generation.

A dropdown that says "Rules" with a search box where "arrow" has been input. The dropdown has two results: "eslint/arrow-body-style" and "vue/no-arrow-functions-in-watch".

The oxc playground UI with a sidebar that has the "eslint/no-console" rule selected. In the main view there is a code editor with the following example code: ``` console.log("foo"); // error because we set the no-console linter rule. debugger; // allowed because no other rules are enabled. ```

I have updated the oxlint playground with a linter rule picker, so now you can easily use the playground for reproduction/issue reporting when there are bugs :) Or just play around with oxlint a bit more easily!

playground.oxc.rs