setup-obsidianlisted
Install: claude install-skill vraj-ai/skills
# setup-obsidian — build the retrieval graph
A docs folder with no map is a **pile**: every question reads every file. The same folder with a router and an index is a **librarian**: two or three files read, answer back.
The pretty graph view is a poster. The index is the engine. Build the engine.
Four artifacts, each with one job:
| File | Job |
|---|---|
| `ROUTER.md` | Signpost. Which index holds this kind of answer. Never the answer itself. |
| `index*.md` | Catalogue. One line per node: link, title, one sentence. **Generated.** |
| `state.md` | What survives between sessions. |
| the nodes | The knowledge. Already written — you are not authoring these. |
## Step 1 — Refuse an empty vault
Confirm the repo already holds real written knowledge — decision records, specs, planning docs, issues. Count them.
Structure over an empty folder is busywork. If there is nothing to index, stop and say so.
**Done when:** you can state the file count and where it lives.
## Step 2 — Pick the vault scope
The vault is the folder where **you can say why every file exists**. Usually the repo's docs directory, not the repo root — a root vault indexes `node_modules` and drowns the signal.
Two checks before committing to the scope:
- **Junk ratio.** Count indexable files inside the candidate versus the whole repo. If the candidate is not ~100% signal, tighten it or plan exclusions.
- **Link collisions.** Generated links are relative to the vault root. If a folder name inside the vault