wiki-warehouselisted
Install: claude install-skill Hefrock/agent-skills
# Wiki Warehouse
The vault is the **index**; a separate private GitHub repo (`knowledge-warehouse`) is the
**warehouse**. Originals and full extracted text live in the warehouse. The vault gets
only a lean **Source note** — summary, key excerpts, and a content-hash pointer — so it
stays fast and its git history stays small. The two are joined by a `doc_id` (a sha256
content hash), never a path, so warehouse files can be reorganized without breaking the
vault.
This skill owns the intake pathway. It does **not** replace `wiki-synthesizer` — the
synthesizer still compiles the resulting Source note into `Knowledge/` concepts. Think of
it as: **warehouse ingests the artifact → synthesizer distills its meaning.**
## Prerequisites
1. The `obsidian-vault` MCP server connected (same as wiki-operator). Verify with `/mcp`.
2. The `knowledge-warehouse` repo cloned locally, with push access. The intake tool is
`bin/intake.py` in that repo; the vault side is this skill.
3. For born-digital PDFs (the common case), one extractor available to `intake.py`:
`pip install pymupdf` is enough. Scanned documents additionally need an OCR tool
(`apt install ocrmypdf`) — `intake.py` detects and reports this per document.
If the MCP server or the warehouse clone is unavailable, stop and tell the user — do not
simulate ingestion or write a Source note with a pointer to a file that wasn't stored.
## Principles
1. **The binary never enters the vault.** Originals and full text live only in t