Post

this sucks 🙏 you all handled it super professionally

Short version of the story: somebody clicked on a link. Don't click on links.

That's an inaccurate assessment of what happened. None of our maintainers were compromised; nobody took direct action to incur this incident. Instead, a misconfiguration of our CI allowed this to happen fully autonomously.

Fair. I was more just using this as a humourous opportunity to remind everybody that we don't click on links, we don't think about clicking on links, and we certainly don't forward links to others for them to also click on.

I'd be happy to help roll out `@lavamoat/allow-scripts` and some good defaults from lavamoat.github.io/guides/harde... to tanstack/start. It's make repositories using it immune to the entire class of attacks like the one used in malware that hit you.

We'd be super down, thank you. Let me know how I can help facilitate

I'm leaving for JSHeroes tomorrow, but maybe that's an opportunity to kickstart it in-person. I know @tkdodo.eu is gonna be there. Maybe someone else too? Regardless, I need someone who'd like to work with me and has enough expertise in how the project is set up to be able to push back /

"more resistant" !== "immune"

Thanks, I meant immune 😁 With the right tools and configurations you can prevent all known install-time attack vectors. As a result, the malware would be forced to rely on people running the software from the package, which is much less common than simply having it in the dependency tree.

This post mortem reads like it was written nearly entirely by Claude. It’s constantly repeating itself, and I feel like parts of it don’t make sense. It melodramatically insists that the attack isn’t novel three different times.

Like, inside the entirely redundant “Why none alone is enough” section, Claude’s bullet point list of the ‘three’ vulnerabilities begins, “1. pull_request_target alone is fine for trusted operations (e.g. comments, labeling).” I don’t see how it makes sense to present that as a vulnerability at all.

The ‘open questions’ section is baffling, too. Why are we reading about ‘Tanner’ having access to post-job logs? Is he not supposed to be the author? Why are we reading vague instructions to “cross-reference” the activity history of the malicious GitHub account?