integrate OpenSpec into dev-brain #52
Reference in New Issue
Block a user
Delete Branch "%!s()"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
dev brain should come with prepackaged (or on-demand loaded?) openspec. the skills should be automatically be part of the dev brain default and the command line tool should automatically installed. The idea is to always work with proper specifications so that the weaker models we can run locally get as much guidance as possible for a given problem.
We need to bundle openspec skills into the bundle, but not inside strings in the source code, like it is currently done for dev brain. instead there should be a folder with files that are versioned as markdown in the source and are carried into the bundle and from there populated into the right places inside the dev brain. The external utility is better than internal code here, because that allows the use of that tool from outside agents that use our server component.
The command line utility requires additional work, because it is currently based on typescript and I don't want to bring such external dependencies into our project. We should rebuild the command line utility in Rust, so we can carry it inside our own project as source and build target and have a performant utility. There needs to be a button in preferences to create a link in ~/.local/bin/ for that utility and make sure it is integrated into the PATH on the machine with the typical ways that is done on macos.
there needs to be a way to refresh the openspec skills from the bundle when a new version is installed, via button in the def brain settings - just like the one for the refresh of purpose.md and schema.md for dev brain. So there should be a source of predefined skills that is then populated into the dev brain.
What should not happen is that stuff is installed via npm. This is still meant to be mostly offline and we want to control the openspec implementation we deliver with the agent, beause of our weaker models.
There needs to be preferences to activate agents other than ours with the default list being claude code, codex, opencode and pi, where those preferences make sure that there are symbolic links in the right places for those agetns to find and use the skills, too. This is to ensure that the cli tool we built and the openspec skills are aligned and outside agents do the same process as the inside agent.
nope. just nope. dev brain integrated stuff should be small and adhoc and not something big that is there, that should come from default locatgions.