One of the authors here — happy to answer anything.
The short version: we built this for ourselves and our clients. When building data engineering projects, we wanted proper declarative config instead of click-ops and hand-rolled scripts that drift from reality the moment someone touches the UI. It's been a real productivity boost — everything's already set up and ready to go. Because the config is declarative, an LLM can read all of it easily and then verify the result, since the tools are already at hand. That's what makes the output solid.
It turned out pretty well, so we figured we'd share it. It's still early, and I'd genuinely like to hear where the model breaks for your workflows.
For now we've focused mostly on Claude. It should work with other agents too, but you'd likely need to do a bit of tweaking — adjusting skills, plugins, that kind of thing. Happy to help with that if you run into anything.
One of the authors here — happy to answer anything.
The short version: we built this for ourselves and our clients. When building data engineering projects, we wanted proper declarative config instead of click-ops and hand-rolled scripts that drift from reality the moment someone touches the UI. It's been a real productivity boost — everything's already set up and ready to go. Because the config is declarative, an LLM can read all of it easily and then verify the result, since the tools are already at hand. That's what makes the output solid.
It turned out pretty well, so we figured we'd share it. It's still early, and I'd genuinely like to hear where the model breaks for your workflows.
Is Claude Code a hard requirement here, or can I point any agent at the scaffolded project?
For now we've focused mostly on Claude. It should work with other agents too, but you'd likely need to do a bit of tweaking — adjusting skills, plugins, that kind of thing. Happy to help with that if you run into anything.
[flagged]