What's the point to maintain main specs?

[sherrrrr]

I'd like to understand: Are main specs intended to help the AI understand existing business logic during the design phase?
I use AI to help read the source code and seems that openspec/specs/ is only mentioned in

1. "Check openspec/specs/ for existing spec names" (schema.yaml:17)
2. "Locate the existing requirement in openspec/specs/<capability>/spec.md" (schema.yaml:39)
3. Some post-design operations: openspec archive, /opsx:sync, /opsx:archive, openspec validate, and openspec show.

These didn't ask AI to think with the content of main specs.

Questions
What's the point to maintain main specs? Does main specs used somewhere else to help AI think? Or is it maintained only for human to read?

[m9dfukc]

I would like to know as well ... I just very recently started to deep dive into how openspec (and claude rules/skills/etc)) really works (besides using openspec heavily since months). It would be great it we could "do more" with these specs ... maybe autogenerate rules / skill for recurring patterns. Right now it seems to me, that I would have to reference them manually to the LLM if I want to gain any advantage out of these skills.

[marcindulak]

It's hard to make statements about nondeterministic agents that tend to ignore instructions, but it appears that today the value of maintaining the design documentation is mostly for humans to keep the agent on track.

See Do Coding Agents Need Design Documentation?. More serious research on that topic may appear at some point.