Claude Skill for Obsidian: Full Setup Guide

A Claude skill for Obsidian is a folder of instructions that teaches Claude how to read, tag, file, and link the notes in your vault the same way every time — instead of you re-explaining your folder structure and tagging conventions in every conversation. Point Claude at your vault through Cowork or Claude Code, drop in one skill, and the messy inbox of half-written notes you have been meaning to organize for six months becomes a five-minute weekly pass.

This guide covers both ways to give Claude folder access to a real Obsidian vault, a complete note-processing skill you can copy in today, and a daily-notes workflow built around it. If you have not built a skill before, start with our how to build a Claude skill walkthrough — this article assumes you know what a skill is and wants to show you a specific, working one.

What a Claude skill for Obsidian actually does

It is not a plugin, and Claude does not run inside Obsidian itself — the skill is a set of markdown instructions that Claude loads when it detects you are working on vault-related notes, and it operates on the files directly through whatever tool has folder access to that vault. For the full definition of what a skill is and how Claude decides when to load one, see our what are Claude skills guide.

Concretely, a note-processing skill for Obsidian tells Claude: here is my folder structure, here is my tagging taxonomy, here is how I want a raw note turned into a filed, linked, properly-tagged one. You write that logic once. After that, saying "process my inbox" triggers the same routine every time, with the same judgment calls, instead of Claude guessing at your conventions from scratch in each session.

Two ways to open your vault to Claude: Cowork vs Claude Code

You need something with real file access to your vault folder before any skill can act on it — Claude in a browser tab cannot read or write local markdown files. There are two practical routes for non-developers: Cowork, Anthropic's agentic desktop workspace, or Claude Code run from a terminal pointed at your vault directory.

For most people who just want their notes organized without touching a terminal, Cowork is the easier front door — you scope it to the vault folder once, and it keeps that access across sessions. If you already have Claude Code installed for other work, running it from inside your vault directory costs you nothing extra and gets you the same skill support.

The note-processing skill: full code you can drop in today

Below is a complete skill for processing raw Obsidian notes: it adds frontmatter, assigns tags from a fixed taxonomy, decides which folder a note belongs in, links it to related existing notes, and appends a reference in today's daily note. Save this as SKILL.md inside a folder named obsidian-note-processor, and put that folder under .claude/skills/ in your vault (or your user-level skills folder if you want it available in every vault you open).

---
name: obsidian-note-processor
description: Use when the user asks to process, clean up, file, tag, or
  link raw notes in their Obsidian vault. Triggers on phrases like "process
  my inbox", "file this note", "clean up my notes", "tag and link this".
---

# Obsidian Note Processor

You are processing raw notes from this vault's Inbox/ folder into
properly tagged, filed, and linked permanent notes.

## Vault structure
- Inbox/        — unprocessed notes land here, filename = rough title
- Areas/         — ongoing responsibilities (Work, Health, Finances, ...)
- Projects/      — time-bound efforts with a defined end state
- Reference/     — reusable knowledge, no action attached
- Daily/         — one note per day, named YYYY-MM-DD.md

## Tag taxonomy (only use these — do not invent new top-level tags)
#project #area #reference #idea #decision #followup #person #meeting

## For each file in Inbox/
1. Read the note. If it has no frontmatter, add:
   ---
   created: [today's date]
   tags: []
   status: inbox
   ---
2. Read the content and assign 2-5 tags from the taxonomy above based on
   what the note is actually about — not what folder you think it should
   land in. A meeting note about a project still gets both #meeting and
   #project.
3. Decide the destination folder:
   - Tagged #project -> Projects/[Project Name]/
   - Tagged #area -> Areas/[Area Name]/
   - Tagged #reference -> Reference/
   - Everything else stays in Inbox/ and gets flagged for the user to
     decide manually — never guess on ambiguous notes.
4. Search the vault for existing notes sharing 2+ tags or obvious keyword
   overlap. Add a "## Related" section at the bottom of the note with
   [[wikilinks]] to the 2-4 most relevant matches. Do not force links that
   are not genuinely related.
5. If the note references a specific date, event, or decision made today,
   append one line to Daily/[today].md under a "## Notes filed today"
   heading, linking back to the processed note.
6. Move the file from Inbox/ to its resolved folder. Update the
   frontmatter status field from "inbox" to "filed".
7. Append one line to Reference/Processing-Log.md:
   [date] — filed [[note title]] -> [destination folder], tags: [list]

## Rules
- Never delete a note. If you are unsure where something belongs, leave
  it in Inbox/ with a comment rather than guessing.
- Never overwrite existing frontmatter fields you don't understand —
  add to them, don't replace them.
- Process one note at a time and report what you did before moving to
  the next, so mistakes are easy to catch.
- If Inbox/ has more than 15 unprocessed notes, process the 10 oldest
  first and tell the user how many remain.

The rule at the end — never delete, never guess silently — matters more than it looks. A skill that files things wrong is annoying. A skill that deletes or overwrites something is a trust problem you do not get back easily, so the instructions are written to fail toward "leave it and flag it" rather than toward "make a confident guess."

A second thing worth deciding up front is scope. You do not have to connect your entire vault on day one — if your Obsidian setup mixes personal journaling with work notes, scope Cowork or Claude Code to just the Inbox/ and Daily/ folders while you get comfortable, then widen access once you trust the output. Folder access in both tools is granular by design specifically so you can start narrow and expand later rather than handing over the whole vault before you have seen a single run.

Wiring the skill into a daily notes workflow

The skill above processes what already exists. The daily-notes workflow is what feeds it — a short end-of-day routine where Claude reads today's daily note, pulls out anything that should become a permanent note, and hands it to the processor. Run this once in the evening and your inbox never grows past what a single pass can clear.

Read Daily/[today's date].md.

1. Pull out anything that looks like a standalone idea, decision, or
   task that deserves its own note rather than living only in today's
   log — a new project mention, a decision with reasoning attached, a
   person's contact context, a reusable idea.
2. For each one, create a new file in Inbox/ with a clear title and
   the extracted content, plus a one-line frontmatter note pointing
   back to today's daily note as the source.
3. Leave the daily note itself untouched except for a "## Extracted
   today" section listing what you pulled out, linked to the new
   Inbox files.
4. Then run the obsidian-note-processor skill on the Inbox/ files you
   just created.
5. Report a short summary: how many notes extracted, how many filed,
   how many left in Inbox for a manual decision.

In practice this turns a daily note from a write-once scratchpad into the front door of your whole vault. You write freely during the day, and the two routines together — extract, then process — do the organizing you would otherwise put off until the backlog got uncomfortable. It is the same principle behind good writing workflows generally: capture messy first, structure second. Our how to use Claude for writing guide covers that same split for longer-form drafts.

Guardrails before you trust Claude with your whole vault

Folder access means real write access to real files, so the failure mode to design against is silent damage — a note filed to the wrong place is recoverable, a note that got overwritten or merged incorrectly may not be, especially if you do not notice for weeks. A few checks are worth doing before you let any skill run unattended on your full vault.

• Test on a copy first. Duplicate your vault folder, point Cowork or Claude Code at the copy, and run the skill against a realistic Inbox/ before you trust it on the original.
• Use version control if you can. A vault that is a git repository (or synced through a tool with version history) means any bad file operation is one revert away from fixed.
• Start with read-heavy tasks. Ask Claude to summarize or tag-suggest before you let it move and rewrite files on its own — see the output for a week, then hand over more.
• Keep the taxonomy small and explicit. A skill that can invent new tags or folders on the fly will drift from your system within a month. The skill above deliberately restricts tags to a fixed list for this reason.

None of this is specific to Obsidian — it is the same discipline you would want for any skill with write access to files you cannot easily reconstruct. If you are still deciding what your first skill should even be, our best Claude skills roundup is a good place to see which ones are worth building versus which ones are worth skipping.

Frequently asked questions

Do I need Claude Code, or is Cowork enough on its own?

Cowork is enough for most people. It gives Claude scoped folder access to your vault, supports skills, and can run them on a schedule without a terminal in sight. Claude Code is worth adding if you already use it for other work, or if you want to run the vault skill as one more step in a larger automated setup.

Can the skill break my existing note-taking system?

Only if the taxonomy and folder rules in the skill do not match your real system — which is why the skill above spells out your folders and tags explicitly rather than letting Claude infer them. Copy it, edit the taxonomy and folder list to match your actual vault, and test on a duplicate before running it on the original.

Does this work with Obsidian plugins like Dataview or Templater?

Yes, as long as the frontmatter and file structure the skill produces are compatible with what those plugins expect — Claude is just reading and writing plain markdown files, the same files your plugins already read. If a plugin depends on a specific frontmatter field, add that field to the skill's instructions explicitly.

Where do I go to learn how to build my own skills from here?

Start with our what are Claude skills guide for the concept, then how to build a Claude skill for the step-by-step process of writing your own SKILL.md from a blank folder — the same structure used for the Obsidian skill above applies to any repeatable task you want Claude to handle consistently.

What if my vault does not use the Inbox-Areas-Projects layout?

The folder names in the skill above are a placeholder, not a requirement. If your vault follows a different system — a flat structure, a Zettelkasten-style numbered sequence, or a calendar-first layout with no separate Areas or Projects folders at all — replace the "Vault structure" section with your own folder names and skip step 3's folder-routing logic entirely if it does not apply. The parts worth keeping regardless of layout are the tag taxonomy discipline, the never-delete rule, and the daily note backlink, since those three are what keep a vault searchable as it grows past a few hundred notes.