claude-obsidian — Cheat Sheet

Version installed: 1.4.3 · Source: AgriciDaniel/claude-obsidian · License: MIT

The plugin turns Claude into a persistent wiki maintainer. You feed it sources, questions, or conversations — it files, cross-references, and maintains the knowledge base. Knowledge compounds across sessions instead of evaporating.


The 10 Skills at a Glance

#SkillSay / typeOne-line purpose
1wiki/wiki, “set up wiki”, “scaffold vault”Setup check + router. Use first after install.
2wiki-ingestingest [file], ingest this url, batch ingestRead source → create 8-15 cross-referenced wiki pages
3wiki-querywhat do you know about X?, query: X, based on the wikiAnswer a question using only the wiki, with page citations
4wiki-lintlint the wiki, wiki audit, find orphansHealth check: orphans, dead links, stale claims, frontmatter gaps
5save/save, save this, file thisFile the current chat as a structured wiki page
6autoresearch/autoresearch [topic], deep dive into [topic]Autonomous research loop: web → synthesize → file
7canvas/canvas, canvas add image, canvas zoneBuild Obsidian visual canvases (images, PDFs, notes)
8defuddledefuddle, clean this page, strip this urlStrip ads/navigation from web pages before ingest (saves 40-60% tokens)
9obsidian-markdownAuto — reference for other skillsCorrect Obsidian Flavored Markdown (wikilinks, callouts, embeds, properties)
10obsidian-basescreate a base, add a base file, database viewNative Obsidian .base files: dynamic tables, filters, formulas

What Each Skill Reads and Writes

This is the part most docs skip. If you want to know what a skill will actually do to your vault, read this table.

SkillReadsWrites / ModifiesTypical scale
wiki-ingestsource file/URL, .raw/.manifest.json, wiki/index.mdwiki/sources/{slug}.md, multiple wiki/entities/ + wiki/concepts/ pages, updates wiki/index.md + wiki/log.md, creates/updates .raw/.manifest.json1 source → 8-15 pages touched
wiki-querywiki/hot.md, wiki/index.md, relevant wiki pagesNothing by default. If you say “file this answer”, also writes a wiki/questions/ page and logs itRead-only unless you ask to save
wiki-lintWhole wiki/ tree, wikilinks graphCreates/updates Dataview dashboards in wiki/meta/, may update frontmatter gaps, generates canvas health mapsCan touch dozens of pages if there’s cleanup to do
saveThe current conversationOne new page in wiki/concepts/, wiki/questions/, wiki/sources/, or wiki/meta/ depending on content type. Updates wiki/index.md + wiki/log.md + wiki/hot.md1 page + 3 updates
autoresearchprogram.md config you write, web search resultsMany pages — sources, entities, concepts, the driving question, a synthesis. Updates index + log + hotHighest impact. A single topic can create 20+ pages
canvasImages, PDFs, wiki pages.canvas files (JSON-based) in Obsidian vaultNo markdown pages affected — only canvases
defuddleA URLNothing in your vault — emits cleaned markdown to be piped into wiki-ingestStaging step; no writes on its own
obsidian-basesWiki pages with frontmatterNew .base files (Obsidian database definitions)1 base file per view
obsidian-markdownNothingNothing — pure reference skillN/A
wikiVault stateScaffolds missing dirs/files on first run; otherwise reports statusFirst run: touches scaffold. Subsequent: read-only.

What Changes in Your Vault the First Time You Use Each

Run this as a mental checklist. These files/dirs appear as a side effect of using the plugin — they’re normal and expected.

First use ofNew on disk
any skill(already done — WIKI.md, wiki/{entities,sources,domains,comparisons,questions,meta}/, hot.md, overview.md exist)
wiki-ingest.raw/.manifest.json (tracks hashes so re-ingesting a file is idempotent)
wiki-lintDataview dashboards in wiki/meta/, possibly a wiki/meta/orphans.md
canvas.canvas files (Obsidian’s native JSON format) at the vault root or wherever you specify
obsidian-bases*.base files — require Obsidian v1.9.10+ to render natively
autoresearchA program.md config file at whatever path you specify

Legacy content (wiki/concepts/, wiki/connections/, wiki/courses/, wiki/summaries/ — 140+ files) is never auto-migrated. Plugin skills read them fine but write new content into the new-schema dirs.


Frontmatter Every New Page Needs

When plugin skills create a page, they use this schema. If you hand-write a page in the new dirs, match it.

---
type: source | entity | concept | domain | comparison | question | overview | meta
title: Human-Readable Title
status: seed | developing | mature | evergreen
created: YYYY-MM-DD
updated: YYYY-MM-DD
tags: [tag1, tag2]
related: [[Other Page]]
sources: [[.raw/file.md]]  # only for derived content
---

Status ladder:

  • seed — created but barely filled in
  • developing — being actively built up across sessions
  • mature — substantially complete, stable
  • evergreen — reference material, only updated when facts change

Common Recipes (I want X → Do Y)

Ingest a PDF or textbook chapter

ingest raw/ADMN201/ch14-money-banking.pdf

Creates wiki/sources/ch14-money-banking.md + entity/concept pages for every named thing. Log entry appended.

Ingest a bunch of sources at once

ingest all the files in raw/PHIL252/

Batch mode. One at a time with cross-referencing.

Ingest a web article (clean it first)

defuddle https://example.com/article
ingest the cleaned output

Or just say ingest https://... and it’ll defuddle automatically.

Ask the wiki a question

what do you know about Open Market Operations?

Reads hot cache → index → relevant pages → answers with See [[concepts/OpenMarketOperations]] citations.

Save the current conversation as a note

/save

Or save this as "Key takeaways from Ch14".

Run a health check

lint the wiki

Returns prioritized list: orphans, dead links, frontmatter gaps, stale claims, empty sections. You pick which to fix.

Kick off autonomous research

/autoresearch Canadian monetary policy 2024

Runs web search, fetches sources, synthesizes, files. Can produce 20+ pages. Configure depth/scope via program.md.

Make a visual map

/canvas
canvas add note [[concepts/Business/InvestmentVehicles]]
canvas add note [[concepts/Business/RiskManagement]]
canvas zone "Portfolio fundamentals"

Build a dynamic table

create a base file showing all concepts with status=developing

Produces an Obsidian .base — queries update live as pages change.


How It Interacts With Your Existing Skills

graph LR
    subgraph Sources
        PDF[PDF / Chapter]
        URL[Web article]
        Convo[Chat insight]
    end

    subgraph Plugin [claude-obsidian]
        INGEST[wiki-ingest]
        SAVE[save]
        QUERY[wiki-query]
        LINT[wiki-lint]
    end

    subgraph Personal
        TS[tutor-setup]
        TUT[tutor]
        LES[lesson-tutor]
    end

    subgraph Vault[wiki/]
        CONC[concepts/]
        SRC[sources/]
        ENT[entities/]
        TRACK[tutor/concepts/]
        DASH[ADMN201-dashboard.md]
    end

    PDF --> INGEST
    PDF --> TS
    URL --> INGEST
    Convo --> SAVE

    INGEST --> SRC
    INGEST --> ENT
    INGEST --> CONC
    SAVE --> CONC
    TS --> CONC
    TS -->|practice Qs| TRACK

    TUT -->|reads| CONC
    TUT -->|reads| TRACK
    TUT -->|writes| DASH
    LES -->|reads| CONC
    LES -->|reads| DASH
    LES -->|appends "In Your Own Words"| CONC
    LES -->|writes| TRACK

    QUERY -->|reads| CONC
    QUERY -->|reads| SRC
    LINT -->|audits| Vault

When to pick which

GoalUse
Read a textbook chapter and get concept noteswiki-ingest
Read a textbook chapter and get concept notes + practice quiz questionstutor-setup
Quick answer grounded in your noteswiki-query
Drill yourself with a quiz/tutor
Get taught a concept interactively/lesson-tutor
Preserve a good chat answer/save
Research a topic end-to-end from scratch/autoresearch
Vault-level hygienewiki-lint

What’s NOT Touched

  • .obsidian/ — your theme, plugins, hotkeys, workspace layout. Untouched. (You specifically declined to run setup-vault.sh.)
  • Existing 140+ files under wiki/{concepts,connections,courses,summaries,tutor}/ — read, not rewritten.
  • wiki/index.md, wiki/log.md, wiki/study-plan.md, wiki/ADMN201-dashboard.md — appended to, never overwritten.
  • Auto-memory at ~/.claude/projects/.../memory/ — separate system, plugin ignores it.
  • The three personal skills (tutor, lesson-tutor, tutor-setup) — live alongside the plugin’s 10 skills.

Gotchas

  1. Re-ingesting is free. .raw/.manifest.json hashes sources, so running ingest on an unchanged file short-circuits. If you want to force re-ingest, delete the entry in .manifest.json.
  2. URLs get defuddled by default. wiki-ingest calls defuddle under the hood for URLs. Token savings are real (40-60%), but if you need the raw HTML, call wiki-ingest with a --raw hint.
  3. Autoresearch writes a LOT. Budget 20+ new pages per invocation. Review with wiki-lint after.
  4. .base files need Obsidian 1.9.10+. Older versions render them as raw JSON. Check your Obsidian version before using obsidian-bases.
  5. Hot cache rewrites on every session. wiki/hot.md is a rolling ~500-word summary — don’t put anything permanent there. It’ll be overwritten.
  6. Quartz deploy is still manual. The plugin doesn’t touch ~/Documents/quartz/. After a session, if you want jrnolet.org updated, still run:
    cp -r ~/Documents/StudyWiki/wiki/* ~/Documents/quartz/content/
    cd ~/Documents/quartz && git add . && git commit -m "update wiki" && git push origin v4:main

Quickest Value: Three Things To Try First

  1. lint the wiki — baseline health check. Shows what’s already broken in the existing 140+ files (orphans, dead links, etc.) so you know what cleanup is worth doing.
  2. ingest raw/ADMN201/[any chapter].pdf — see how a source turns into 8-15 linked pages. Compare the output to your legacy summaries/ format and decide how you like the new schema.
  3. /autoresearch [an exam topic you feel shaky on] — the compounding effect shows up here. Twenty cross-linked pages on, say, “money supply mechanics” that you can then drill with /tutor.

One-Sentence Test for “Should I Save This?”

If it took you more than a minute to produce and you’d be annoyed to re-derive it, say /save. Let the wiki compound.