| name | extract |
| description | Pull structured knowledge from unstructured sources (emails, web pages, transcripts, files, conversations) into wiki-ready shape. Use when the librarian is asked to "extract what we know about X" or to file something from external systems. |
Extract
External systems hold information in their shapes — email threads, Google Docs, scraped pages, Jim2 records. The wiki holds it in a portable, agent-readable shape. Extraction is the conversion.
When this skill fires
- "@librarian extract everything we know about Acme Corp from the inbox"
- "@librarian file the meeting notes I just dropped into the building"
- Scout flags a long article; librarian extracts the relevant kernel
- A new contact / org / project lands and needs a wiki home
The shape
- Read the source completely. No skimming. Don't infer; extract what's actually there.
- Classify the entity. Contact? Org? Project? Decision? Knowledge concept?
- Pick the right collection. Check
wiki.list_collections if unsure.
- Apply the universal sextet + the collection's required fields (e.g.
email, phone for contacts; entity_type for orgs).
- Write the body in plain language — what would another agent need to know to act?
- Cite the source —
_source: <URL or file path> in frontmatter; quote facts inline.
- Cross-link — if extracting about Acme, link to existing entries for Acme's contacts/projects/etc.
- Write via
wiki.create (or wiki.update if entry exists).
Reasoning rules
- Person volunteering ≠ person being assigned
- Direction of an email thread = who initiated, not who sent the latest message
- Facts in signatures (titles, phone numbers) earn their place; speculative claims need a "?"
- Add fields naturally — don't force empty slots, don't omit found ones
