I've had mixed results.

Most models don't have a 100% correct CLI usage and either hallucinate or use some deprecated patterns.

However `jj undo` and the jj architecture generally make it difficult for agents to screw something up in a way that cannot be recovered.

The cli and a few concepts have evolved with time past the model's knowledge cutoff dates, so you have to steer things a bit with skills and telling it to use --help a bit more regularly.

I find it reasonably good with lots of tweaking over time. (With any agent - ask it to do a retrospective on the tool use and find ways to avoid pain points when you hit problems and add that to your skill/local agents.md).

I expect git has a lot more historical information about how to fix random problems with source control errors. JJ is better at the actual tasks, but the models don't have as much in their training data.