Claude Skill for Obsidian: Turn Claude Into a Note Agent

A Claude skill for Obsidian teaches Claude how to read and write real Obsidian Flavored Markdown — wikilinks, embeds, callouts, and frontmatter properties — instead of generic Markdown. You install it by adding the skill files to a .claude folder in your vault, then let Claude create and edit notes that behave correctly inside Obsidian.

Skill structure and install steps verified 31 July 2026 against Anthropic's Agent Skills docs.

Out of the box, an AI assistant writes plain Markdown that ignores everything that makes Obsidian powerful: [[wikilinks]], embeds, callouts, and properties. A Claude skill for Obsidian fixes that. It gives Claude a set of instructions for Obsidian's specific syntax, so the notes it produces link, embed, and render the way your vault expects. This guide explains what a Claude skill for Obsidian is, how to install one, and how to use it to turn Claude into a genuine note-taking agent. The skill works with the models Claude currently spans — the latest ones being Opus 4.8, Sonnet 5, and Haiku 4.5 — and setup takes under 10 minutes, so we review these steps regularly to keep them accurate.

What is a Claude skill for Obsidian?

A packaged folder of instructions that teaches Claude how your vault works — its structure, link syntax, frontmatter, and conventions — so it can act on notes correctly.

An Agent Skill is a small package of instructions and resources that teaches Claude to do a specific job well. A Claude skill for Obsidian is exactly that, focused on Obsidian's flavor of Markdown. The best-known one is the open-source obsidian-skills project from kepano (Steph Ango, Obsidian's own CEO), which teaches Claude to use Obsidian's file formats and CLI.

The point is fidelity. Without the skill, ask Claude to "link this note to my project note" and it writes a standard [text](path.md) link that Obsidian treats as a plain file reference. With a Claude skill for Obsidian, Claude knows to write [[Project]] — a real wikilink that shows up in the graph, in backlinks, and in autocomplete. The skill encodes dozens of these Obsidian-specific rules so the AI works with your vault rather than against it.

What the skill actually teaches Claude

Four things: wiki-link and tag syntax, your frontmatter schema, folder conventions, and the house rules for how notes should be written and filed.

A good Claude skill for Obsidian covers the syntax that generic Markdown misses:

• Wikilinks — [[Note Name]] for internal links, including [[Note#Heading]] and [[Note#^block-id]] for headings and blocks.
• Embeds — ![[Note]] to transclude a whole note, section, image, or PDF into another.
• Callouts — typed blockquotes like > [!note] that render as styled alert boxes.
• Properties (frontmatter) — YAML metadata such as title, tags, and aliases at the top of a note.
• Tags and comments — inline #tags and %% comments %% that Obsidian understands.
• Advanced formats — newer Obsidian features like Bases and JSON Canvas.

Crucially, the skill also teaches the one rule most AIs get wrong: use [[Note Name]], never [link](path/to/note.md), for internal links. That single habit is the difference between a vault that stays connected and one that quietly fills with broken references. In our own day-to-day work with Claude Code, this is the exact mistake we see most often when no skill is loaded — plain [text](path.md) links that look correct but never surface in the graph or backlinks — which is why we treat this one rule as the whole reason the skill earns its place.

How to install a the skill

Drop the skill folder where Claude can read it, point Claude at your vault directory, and confirm it can list and read notes before letting it write.

There are two common ways to get a the skill working, depending on whether you use Claude Code or want it inside the Obsidian app.

Option A — with Claude Code in your vault. Add the skill files to a .claude folder in the root of your Obsidian vault (or the folder you point Claude Code at). Cloning the obsidian-skills repository into that location is enough — the next time you run Claude Code there, it picks up the skill and applies Obsidian's rules automatically.

Option B — inside the Obsidian app. The Claude Code Skills community plugin brings the same idea into Obsidian's interface: highlight text in a note, right-click to pick a skill, and get a streaming Claude response in a side panel without leaving the app. That's the smoother route if you live in Obsidian and don't want to switch to a terminal.

Either way, once the the skill is in place, you don't invoke it manually — Claude reads the skill and follows Obsidian's conventions whenever it touches a note.

How to use it: Claude as an Obsidian agent

Give it whole jobs rather than single edits — summarise a folder, build an index note, retag a topic — and review the diff before accepting.

With the skill installed, you can hand Claude the kind of note work that used to be tedious:

• Draft and link. "Create a note on this topic and link it to my related notes" produces a note with proper [[wikilinks]] into your existing vault.
• Restructure. "Split this long note into an index plus linked sub-notes" gives you connected notes with embeds, not a wall of text.
• Add metadata. "Add tags and aliases to these notes" writes correct frontmatter properties Obsidian can query.
• Clean up. "Convert these callouts and fix the broken links" repairs syntax at scale.

Pair a the skill with Claude's memory and you get an assistant that understands both your vault's structure and your ongoing context. If you want to build your own skills for other tools the same way, our guides on Claude skills and making Claude Code do anything show the pattern — the Claude skill for LinkedIn is another real-world example of the same idea applied to a different platform.

Tips for getting the best results

Keep the skill short and specific, name real folders and tags, and version the vault in git so any bad write is one command from being undone.

A the skill works best with a little setup on your side:

• Point Claude at the whole vault. Run it from your vault root so it can see existing notes and link into them, rather than from a single-file view where it can't discover [[targets]].
• Name notes the way you link them. Obsidian wikilinks resolve by note title, so consistent, human-readable filenames make Claude's links land correctly.
• Ask for structure, not just text. Prompts like "create an index note that links to these five" or "add properties and tags to these notes" get the most out of the skill's Obsidian awareness.
• Review the graph after big edits. After a large restructure, open Obsidian's graph view to confirm the new [[links]] connected the way you intended.

Because a the skill only changes how Claude writes — not your vault's data — it's safe to experiment. Worst case, you undo an edit. That low risk is what makes it worth trying even on an established vault, and it's a big reason the pattern caught on so quickly among Obsidian power users.

One honest limit worth knowing

Claude edits files, not the running Obsidian app. Changes land on disk and appear when Obsidian re-reads them; plugins and live state are untouched.

A the skill changes how Claude writes going forward — it does not retroactively audit what is already in your vault. It won't silently repair the plain-Markdown links you created before you installed it; it only follows Obsidian's rules on the notes Claude touches from here on. To clean up the past, you have to ask for it explicitly with a task like "find and fix any [text](note.md) links that should be [[wikilinks]]." On a large vault we'd run that cleanup in small batches rather than one sweeping pass, opening the graph view between runs to confirm nothing linked to the wrong note. Set that expectation up front and the skill stops feeling magic and starts feeling dependable: a careful author for every new note, plus an on-demand janitor for the old ones whenever you point it at them.