I answered this question for a member and felt would be helpful for others as a post.
Stop setting up Obsidian before you have anything worth putting in it.
The temptation is to set up Obsidian first and figure out what goes in it. I did it the other way and it worked better.
Every new project I spin up gets three folders from day one: thinking (research, decisions, context), building (active work, what's in progress, what's blocked), and product (what ships, outputs, deliverables). I can't take credit for this system; I learned it from another creator.
That being said, it is this structure that helps Claude always know where to look depending on what I'm doing. The routing is a claude.md at the project root that points to the right folder. Once a project hits checkpoints or wraps, the thinking folder is what migrates to Obsidian. That's your second brain. Not everything you ever touched, just the distilled knowledge worth retrieving later. The building and product folders stay with the project.
Where I started: project folder structure, not notes, not tasks, not a blank Obsidian vault. Getting the thinking/building/product separation right inside each project meant I actually had something worth moving to a second brain later. If you start with Obsidian you'll spend weeks organizing empty folders.
On integrating Obsidian with Claude to auto-summarize and pull past research: the real unlock isn't the Obsidian integration itself, it's the context handoff system. Past research lives in Google Docs. Those get converted to markdown files and loaded into Claude Projects as context. Claude works from those during active projects. Then at checkpoints, the distilled insights move to the Obsidian vault. The chain is Google Docs (raw research) → md files in Claude Projects (active context) → Obsidian vault (long-term second brain). That handoff is the integration.
What I wish someone told me: separate every project folder into thinking, building, and product from the start. I wasted time early on with flat project folders where research sat next to active work sat next to finished deliverables, and Claude couldn't tell what was current vs. done. The three-folder split fixed that overnight.