I have some questions about best practices for deployment? * Do people commit

baml_client

to their repo? Or do you rebuild it as part of your ci/cd deployment pipeline? * How do you prevent coding agents from modifing baml_client and only update

baml_src

?

Great questions! We do encourage you to check in baml_src and to .gitignore baml_client, to reduce PR churn. This comes with some tradeoffs - if you deploy to some platform that wants to run your source code, then your build script will need to generate baml_client on the platform. As for keeping agents off your baml_client - those files do have headers that say "Don't edit me directly!" but probably the assistant isn't reading the entire file, just a chunk. I'd recommend mentioning baml_client by name in your AGENTS.md or CLAUDE.md file, if you use those. If you do, could you let us know if you have success with that?

Sounds good! I've added > Never edit

baml_client

only edit

baml_src

and run

npm run baml:generate

to regenerate the client. to agents.md. Lets see if that works

You can also use this reference if it’s helpful! https://docs.boundaryml.com/guide/development/deploying/docker