We have something coming just for this

Current rule of thumb that has worked well for agent setups around OpenClaw:

• separate workspaces when you need hard boundaries (different memory, different file trees, different tools/policies, different ownership)
• keep a shared workspace when teams mostly use the same knowledge/files and only the routing/persona differs

In practice, splitting too early creates overhead: duplicated memory, duplicated docs, more migration friction, and more chances to edit the wrong context file.

So my default would be:

1. one shared workspace first
2. route by channel / agent / role
3. split into per-team workspaces only when the boundaries become operationally real

If the feature you have coming is something like scoped memory + scoped file visibility without full workspace duplication, that would probably hit the sweet spot.