Super claude vision and future planning

[nicodiferd]

Hey Nomen! I wanted to talk about the new superclaude version, I picked up sc right around the beginning of the summer and loved it, have always had it on my backbone for random superclaude use cases but ended up trying out bmad for a bit there and then gave that up (for good reason) but i eventually came back and saw a newer version and decided to install the fresh version and see whats different, the serena mcp server and squential thinking mcp servers are great! they really seem to have made a huge difference in terms of making the user experience of claude code really not about claude code its self, but its insane agentic ability to use tool calls is next level, really great and easy setup as well. My only concern is that with all of the personas and claude md files that are preloaded into claudes context window as soon as a fresh conversation starts,

it takes up 44% of the context window without a single chat! Which I understand serena and the other mcp servers are essentially retaining alot of the context (and reading and writing the files!) and also claude can deploy sub agents as well (isolated context windows), so my question is, does the high context window affect usage overtime and/or with larger projects? I am more so asking for your vision on the expandability of superclaude, is it higher context windows to keep up with growing project sizes, general context, personas, general rules, tools (mcp severs)? or will the amount of objects kind of bottom out and claude code its self wont be affected by high context window usage and most production models will stay around 200k->400k tokens.

Sorry for the yap and looking for open discussion :)

[spirefyio]

I'd love to understand this too. With the context window being so small on CC.. and losing almost 1/2 of it for SC + initial context.. it limits how many calls you can make before you gotta reset.. which fills it with about 50% or more. It really sucks we're so limited and though many seem to argue 200K is plenty.. it really isnt for anything other than small tasks/projects. Some will argue that is all you should be using it for. I disagree. Many complex scenarios, in my case working on a project that pulls in a few other libraries.. require that knowledge to be useful. Hard to "extend" capabilities or use libraries that it knows nothing about. Similar to using CC or ChatGPT with no context of the latest Go or Java or Zig. It uses old functions, incorrectly, makes stuff up, etc without the context. So larger context would be ideal, but doubt Anthropic is going to expand that anytime soon given how much they are losing with how expensive it is to run AI. This BTW started me down a rabbit hole of how to compress context.. which google tried to do but turned out while it did make more context available, the performance for processing natural language vs compressed formats was resulting in more errors and hallucinations.

[NomenAK]

Hey guys !
First i want to point out most tokens originate from MCP Tools description, not SuperClaude, SuperClaude itself take only 14% and will hardly get smaller, already trimmed and optimize extensively.
However we could argue that since MCP are shipped with SuperClaude, but i have no way to reduce it.
My advise is to disabled unused MCP Server, and Tools when possible (Serena, Morphllm)
And i encourage you to use the main session as overseer and delegating to custom agents for each Task or Plan.

[loomster321]

I am very impressed with Super Claude so far, having worked on it for the past three weeks or so. What would make it even better is to have an agent that reviews the current plan and status and suggests the next 3 to 5 Super Claude commands and perhaps gives two or three options. The framework, while robust, is quite complicated, and having an agent suggest the next steps would be extremely helpful IMHO, especially for beginners. Making this framework easy to dive into would be a way to grow the community, rapidly.