News for hackers. Stuff that matters.
Re:Jujutsu Primer by Okkef on Friday November 20, 2026 @01:47AM (#50571015)
While jujitsu is superior to git in many ways, I found it hard to get Claude and other agents to use it consistently and more effectively than git.

Git and jj work trees are not interoperable, so any tool that assumes git forces you back. Looking at you, Claude app…

Re:Jujutsu Primer by bigstrat2003 on Friday November 20, 2026 @01:51AM (#50571040)
You should really be doing your own commits anyway. It's not hard, and it's important to make sure you review what is going into the repo.
Re:Jujutsu Primer by danpalmer on Friday November 20, 2026 @01:56AM (#50571072)
I used to think this even just 6 months ago, but I no longer do. With the right prompting it's possible to get near-human results on commit chains. That said, I'm not suggesting you shouldn't review code, you absolutely should, it's just that I am fine letting agents manage the commits now.
Re:Jujutsu Primer by danpalmer on Friday November 20, 2026 @01:51AM (#50571041)
Yeah it took me a while to get agents comfortable with it. Mostly I had access to Gemini, and 2.x and early 3.x models struggled. By 3.7 it was much better, and with all the 4 models it's a complete non-issue. Separately I (and others) worked on improving our JJ skills and they're in a much better place now with some scripts that wrap up more complex operations (e.g. maintaining file copy/move markers).

This hasn't been an issue for me for a while now, and I've actually had Opus 5.5 and Gemini Hydrogen build and submit a 28 commit deep stack, with all the reviewing/rebasing/shifting around that entails.

Re:Jujutsu Primer by p_l on Friday November 20, 2026 @02:05AM (#50571129)
During late 3.x timeframe Google has been explicitly adding Jujutsu support across the agent stack in Antigravity, but it's not fully exposed unlike Git yet
Re:Jujutsu Primer by danpalmer on Friday November 20, 2026 @02:17AM (#50571233)
That's also true, the Antigravity harness may have been a contributing factor here. I use the internal Antigravity fork which uses JJ heavily because JJ is a core supported VCS interface for us.
Re:Jujutsu Primer by p_l on Friday November 20, 2026 @03:03AM (#50571556)
Lots of JJ in public antigravity changelogs, it just seems not wired into UI well
Re:Jujutsu Primer by danpalmer on Friday November 20, 2026 @04:53AM (#50572436)
That tracks. I've not used JJ in public Antigravity, only in the internal fork, and that's not JJ for git, but JJ for Piper, so I don't know how much commonality there is.
Re:Jujutsu Primer by ajkavanagh on Friday November 20, 2026 @02:18AM (#50571237)
Recently JJ supports colocated workspace/worktrees so you get a .git directory in the workspace so git is happy. Also workmux has initial support for JJ colocated workspaces and sesh (tmux session handler) also has experimental support. Whilst there are still some sharp edges, none of them are around data integrity, and I'm using it every day. YMMV.
Re:Jujutsu Primer by Okkef on Friday November 20, 2026 @02:53AM (#50571480)
Oooh thanks for pointing that out! :)
Re:Jujutsu Primer by BeetleB on Friday November 20, 2026 @02:29AM (#50571323)
I have hooks that prevent repository modifications with git commands. And I have a global AGENTS.md file that says to prefer jujutsu if present.

It works totally fine.

Re:Jujutsu Primer by hdjrudni on Friday November 20, 2026 @02:32AM (#50571348)
> I found it hard to get Claude and other agents to use it consistently and more effectively than git.

Write yourself a very good JJ skill or reference file from your AGENTS.md.

My LLMs struggled with it a lot too, but I slowly covered everything they were messing up and they haven't made a mistake in months now. Even when there are "divergent" issues, it knows how to recover now.

Re:Jujutsu Primer by photios on Friday November 20, 2026 @02:37AM (#50571374)
> I found it hard to get Claude and other agents to use it consistently and more effectively than git.

That was the biggest hurdle for me too. I was a 2+ year jj user, then I had to switch back to git because of all the agentic tools relying on it.

I recently went back to jj, after finding a decent skill [1] and discovering Emacs' majutsu plugin. Claude still tries to do a git command every once in a while, but it's a lot more bearable.

[1] https://github.com/netresearch/jujutsu-workflow-skill/

Re:Jujutsu Primer by FranGro78 on Friday November 20, 2026 @03:06AM (#50571577)
I use jj with open weight models and Claude Opus without much trouble, even without any explicit skill loaded for them - maybe they've made it into the training data? They seem to understand how the two differ conceptually and when told that the repo it is working in uses jj they stop using stash (or at least claim to).

I can say this hasn't always been the case, and until about 2 (maybe 3?) months ago I would handle all the VCS related parts of the work. I would let the agents write code, which I've found less and less necessary to tweak as time has gone on, and write the commit messages and create the PRs by hand.

My colleagues (who use git) let Claude write commit messages, but I find those awfully long-winded and my eyes just glaze over when I open their PRs - but to each their own I guess.

At least the Claude commit messages are an improvement over the commit messages before - I gave up trying to get them to apply the recommendations of how to write good commit messages given in https://cbea.ms/git-commit/

Re:Jujutsu Primer by indeyets on Friday November 20, 2026 @03:55AM (#50571984)
I wrote an extensive skill some time ago. Works really nice https://codeberg.org/thisalex/jj-skill

It's a tad opinionated: my agents are not allowed to push to external repos — but that's easy to change if you need

Re:Jujutsu Primer by jon-wood on Friday November 20, 2026 @06:32AM (#50573250)
Use hooks. Prompts are for guiding an agent towards a goal. If you want it to never, ever, do a thing then write some hooks that detect an attempt to use git and refuse the tool use. "Is the `git` executable in this command line" is entirely deterministic and doesn't require throwing an LLM running on 15 GPUs at the problem.