Great concept, but the implementation doesn't match the pitch...yet!

[ElliotWood]

Great concept, but the implementation doesn't match the pitch...yet!

Really love the concept the idea of giving your Claude Code session/git copilot etc - some studio structure is something I've been thinking about too.

The folder layout and slash commands are genuinely smart, and the README sold me immediately. We could add MCP for each engine -- ive added seperate discussions for that.

One thing I'd love to see evolve: the agent hierarchy reads more like a vision than an implementation right now. Agents can't actually delegate to each other in Claude Code unless you've solved this part? If not the coordination still lives with the user and the Opus/Sonnet tier assignments don't map to anything the tool can enforce yet.

For a solo dev it also feels like 48 agents is a lot of ceremony; I reckon 15 focused ones would actually get more use. I know a lot of agent work in my day job for enterprise software, and I've found this to be the case. Also, I worked on AutoGen and was the Azure Dev lead in Microsoft Aus.

None of that takes away from what this is as a starting point though.

Are you planning to iterate on the hook enforcement and agent depth? Would love to contribute if so?

[whatchamacallem]

As an old game dev I have to admit I haven't seen this much hierarchy actually add to the creative process, and I haven't seen AI be creative either. But I love the idea. It is the science fiction we read about in the 1980s.

Check out this article on how to use adversarial AI setups to get better quality. You are currently running inside a single chat:

https://www.anthropic.com/engineering/harness-design-long-running-apps

[drhouse]

Check out this article on how to use adversarial AI setups to get better quality. You are currently running inside a single chat:

anthropic.com/engineering/harness-design-long-running-apps

interesting, I generated a notebookLM cinematic out of the article. I wonder when Ollama or similar model would be able to create an adversarial checklist and proceed, no matter how much time it took, with quality output

[Javi1994]

I have been using this tool for the past week, and one thing I found really useful is the design phase combined with prototypes. I don't think I will ever start a game without it again — it just makes your life so much easier. You can prototype any idea in a matter of hours and then keep building on it in another workflow, or however you want, once you have a clear idea.

[drhouse]

are you using Claude Opus/Sonnet or another model... is it expensive?

[Javi1994]

Yeah, I'm using Sonnet most of the time and Opus at High when prototyping and building code. I don't think it's expensive at all, but I guess it depends on the project size. For now all my projects are quite small, so maybe that changes in the future.

[Javi1994]

I also find it really amazing how often it just one-shots the prototypes in just one prompt if you have a basic GDD setup.