Biggest gap in the Claude Code harness is any sense of durable artifacts. It makes a huge difference to formalize persisting requirements, plans, current state against plans, and a durable backlog.
These days I can fire up a brand new session in a repo and ask “what should we get next” and get a synthesized and prioritized plan. And /clear is only about model context, not project context.
As strong as Claude Code is, it’s odd they haven’t embraced more of the meta stuff around software development. Fortunately its extension points are really very good.
I use this same technique as part of https://github.com/brookstalley/prawduct
Biggest gap in the Claude Code harness is any sense of durable artifacts. It makes a huge difference to formalize persisting requirements, plans, current state against plans, and a durable backlog.
These days I can fire up a brand new session in a repo and ask “what should we get next” and get a synthesized and prioritized plan. And /clear is only about model context, not project context.
As strong as Claude Code is, it’s odd they haven’t embraced more of the meta stuff around software development. Fortunately its extension points are really very good.
Seems worth to try.
Please try it
smart
thx