llm 문제 너무 쉽게 타협함 고집이 없음. persona 혹은 value system 이 너무 무르다
the core thing is audit, let user to create task and workflow for auto create git
create research project roadmap
resume roadmap
let users to
create task for new session
audit from macro to micro level is key thing
crdp multi agent
choose audit options
ai os interface provider
aios interface
give them a mcp or native vect tool
skills should be loaded dynamically not only text but also endpoint or sub network tool
Seonglab, Autoproject, Sidecar, Collab, AgentHub, Realtime, contexthub, protocolhub
import claude.md and dynamic skills like language .memory file? or .ctx file language compiler
pr sns feed thumbs up 순으로 agent human 둘다 설득이 중요하다. thumbs up 자체가 비용이고 활발할수록 이득
dynamically saved prompt - user saved long term selectable, dynmaic skills - text description of workflow, static rules - claude.md, dynamic memory - external memory. These are all different
원격 핸드폰에서 yes no
dynamically saved prompt - user saved long term selectable, dynmaic skills - text description of workflow, static rules - claude.md, dynamic memory - external memory. These are all different
혹은 리서치 아니더라도 claude code to everyone
multi agent orchitestaion 에서는 multi modal high level stream으로 상대 다른애들이 뭐하는지 시각처럼 딱 보여야 실시간
디자인은
- Corner app
skill node tailscale integrated research platform for puboci
Framework 만들지 web platform 만들지
instead of chat summarization option, choose cli stdsummarization
Convex driven letter automation style
claude md storage and composer is important to compose in the ui and edit composable components
\n
—//compose separator
\n
이걸로 구분할가 파일
아니다
CRDT
이게 핵심 이거어때
vect crdt 로 다른 agent 뭐하는지 glimps 가능해야ㅣ. 그래야 뭐 요청하거나 수정 공유가능
Ambitious vision
json-render 로 완전 agent native ui 제공하면서 동시에 ai coding ide 를 목표로 담대한 generalization innvoation해도 될듯
seonglab ui 3개 메인서포트
- claude
- codex
- gemini code
agenthub
Seonglae Cho