here’s my harebrained claude system idea: primary agent doe…

Samuel

@samuel.fm

here’s my harebrained claude system idea: primary agent does a plan->clear->implement loop, to keep context fresh. however, it then forgets things this is where a team of “scholar claudes” come in. they each have a chunk of the full context, review the plans, and chip in with helpful insights

5 1 29

Samuel 28 Jan

the scholar claudes *must* believe themselves to be wearing long scholarly robes, and speak in a dreary tone

3 26

Samuel 29 Jan

when a primary claude finishes their task, they join the scholars, and become the expert on their plan

futur 🔹29 Jan

get-shit-done's extensive documentation + letta claude-subconscious is basically my hacked together version of this

2 5

JauntyWunderKind 29 Jan

I was both impressed by and intimidated by get-shit-done / gsd skills when I ran into it. Impulse to try but man there was so much stuff there! Pretty direct path for humans at least. Cool to hear a report from the field that you're using it. github.com/glittercowbo...

GitHub - glittercowboy/get-shit-done: A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code and OpenCode.

A light-weight and powerful meta-prompting, context engineering and spec-driven development system for Claude Code and OpenCode. - glittercowboy/get-shit-done

github.com

1 2

futur 🔹29 Jan

yeah it's the first thing of its kind that I haven't found excessively gastowny, errs verbose still but to a reasonably unreasonable degree

Josh 🌊🌊29 Jan

hm. and here i am just figuring out ralph loops

1 2

futur 🔹29 Jan

ralph is still scary to me, I've just been trying to maximize claude-not-forgetting-stuff

Derridon't 29 Jan

Agent orchestration but it's just meta-textual Claude commentary like the Talmud

asura 29 Jan

I mean that's actually valid. If you want to play with a similar arch goose kind of does that out of the box with a local ollama instance. block.github.io/goose/docs/g...

Julee Burdekin 28 Jan

Yes — but costs. Anyway to ensure context outside of (between) LLMs? Like separation of concerns and to each its specification (declarative declaration of what the end state should look like)? And, or... ???