Claude Skills for Obsidian: Steph Ango's Official Agent Integration

When Steph Ango — CEO of Obsidian and the designer behind the app's philosophy of "files over apps" — published kepano/obsidian-skills on GitHub in early 2026, it quietly marked a turning point. It was the first time a major productivity tool's creator had officially embraced the Agent Skills specification and published production-quality skills for their own platform.

The result: 12,900+ stars, and AI agents that can finally navigate Obsidian vaults the way a human expert would.

What kepano/obsidian-skills Contains

The repository ships five skills, each targeting a specific Obsidian file format or capability:

Skill	File Type	What It Teaches Claude
obsidian-markdown	.md	Obsidian Flavored Markdown — Wikilinks, callouts, Dataview, tags
obsidian-bases	.base	Obsidian Bases (database views over your vault)
json-canvas	.canvas	JSON Canvas for spatial mind maps and flowcharts
obsidian-cli	CLI commands	obsidian:// URI protocol and local REST API
defuddle	Web → Markdown	Strip web pages to clean Markdown before saving

Each skill is a self-contained folder with a SKILL.md file. Because they follow the open Agent Skills specification, they work with Claude Code, Codex CLI, Gemini CLI, and any other compatible agent — not just Anthropic's tooling.

Installing the Obsidian Skills

Getting started takes one command from Claude Code:

/plugin marketplace add kepano/obsidian-skills

Or install individual skills if you only need a subset:

/plugin marketplace add kepano/obsidian-skills/skills/obsidian-markdown
/plugin marketplace add kepano/obsidian-skills/skills/defuddle

After installation, the skills live in ~/.claude/skills/ (global) or .claude/skills/ (per-project). Claude only loads a skill when your request is relevant to it — a design pattern called progressive disclosure that keeps token costs low.

The obsidian-markdown Skill in Practice

The most important skill is obsidian-markdown. Standard Markdown editors write CommonMark. But Obsidian has its own dialect — Obsidian Flavored Markdown (OFM) — with features that vanilla AI agents get wrong:

• Wikilinks: [[Note Name]] and [[Note Name|display text]]
• Embedded content: ![[Image.png]] and ![[Other Note#Section]]
• Callouts: > [!NOTE] / > [!WARNING] / > [!TIP]
• Tags: #tag in-line and in frontmatter
• Dataview inline fields: field:: value

Without the skill, Claude might write standard Markdown links ([Note Name](note-name.md)) that break Obsidian's graph view and backlinks. With the skill, Claude produces native OFM:

---
tags: [meeting-notes, q1-2026]
date: 2026-03-09
project: "[[Project Alpha]]"
---

# Weekly Sync — March 9

> [!NOTE] Action Items
> Follow up with [[Jane Smith]] on the API timeline.

Related: [[Architecture Decision Records]], [[Sprint Planning Template]]

The skill teaches Claude to always use Wikilinks for internal references, write frontmatter that Dataview can query, and use callout syntax Obsidian renders natively.

The defuddle Skill: Web Clipping for AI Agents

defuddle is arguably the most novel skill in the set. It solves a friction point that anyone doing research in Obsidian knows: web pages are noisy. Navigation, ads, cookie banners, and boilerplate bloat every page.

With the defuddle skill installed, you can ask Claude to:

Research the paper at [URL] and save a clean summary to my vault as
Research/Papers/YYYY-MM-DD-paper-title.md

Claude fetches the page, uses defuddle's processing logic to strip non-content elements, and saves a clean Markdown note with proper OFM frontmatter. It's the same capability as Obsidian's web clipper browser extension — but automated and scriptable.

The obsidian-bases Skill: AI-Powered Database Views

Obsidian Bases, released in 2025, turned every vault into a lightweight database — letting you create table views, filtered queries, and linked records across notes. The obsidian-bases skill teaches Claude the .base file format.

A .base file is a JSON-like schema that defines what data to show and how to filter it:

{
  "filters": {
    "and": [
      {"field": "tags", "operator": "contains", "value": "project"},
      {"field": "status", "operator": "!=", "value": "archived"}
    ]
  },
  "columns": [
    {"field": "title", "width": 300},
    {"field": "status"},
    {"field": "due-date", "type": "date"}
  ]
}

With the skill, you can ask Claude to create new Bases views:

Create a Projects base that shows all notes tagged #project,
grouped by status, with columns for title, owner, and due-date.

Without the skill, Claude would have no idea .base files exist. With it, the file Claude generates works immediately on open.

The json-canvas Skill: Mind Maps and Architecture Diagrams

JSON Canvas is an open format for infinite-canvas visual notes — it's how Obsidian's canvas feature stores mind maps, flowcharts, and architecture diagrams. The json-canvas skill teaches Claude this format.

The skill enables agentic canvas creation:

Based on the meeting notes in [[Q1 Planning]], create a canvas
showing the dependency graph between our workstreams as a flowchart.

Claude generates a .canvas file with proper node positions, edge connections, and embedded note cards — ready to open in Obsidian's visual editor.

The obsidian-cli Skill: Automating Vault Operations

Obsidian exposes a URI protocol (obsidian://) and, with the Local REST API plugin, a full HTTP API for programmatic vault access. The obsidian-cli skill teaches Claude both interfaces:

# Open a specific note
open "obsidian://open?vault=MyVault&file=Daily Notes/2026-03-09"

# Create a new note via Local REST API
curl -X POST http://localhost:27123/vault/new-note.md \
  -H "Authorization: Bearer $OBSIDIAN_API_KEY" \
  -H "Content-Type: text/markdown" \
  -d "# New Note\n\nCreated by Claude."

With the skill, Claude understands both mechanisms and can orchestrate multi-step vault operations — create a note, add it to a canvas, update a Base view — as a single automated workflow.

Why This Matters Beyond Obsidian

The significance of kepano/obsidian-skills extends beyond Obsidian users. Steph Ango building official skills signals that tool vendors are taking Agent Skills seriously as the integration layer for AI agents.

Until now, most skills were written by individual developers or third-party enthusiasts. Having the tool's creator publish the authoritative skill set changes the equation:

1. Correctness guarantees: The obsidian-markdown skill accurately reflects OFM because the person who designed OFM wrote it (or supervised it).
2. Maintenance commitment: When Obsidian ships new features, the skills will be updated by the same team.
3. Precedent for other tools: What Obsidian did, Expo did (official expo/skills repository), and others will follow.

For developers and knowledge workers, the practical message is clear: AI agents can now be first-class Obsidian collaborators — not clumsy outsiders guessing at vault conventions.

Getting Started

To use Claude Code as your Obsidian AI assistant:

1. Install Claude Code and open it in your vault directory
2. Install the skills: /plugin marketplace add kepano/obsidian-skills
3. Try your first prompt: "Summarize all my meeting notes from this week into a weekly digest note in Weekly Reviews/2026-W10.md"

Claude will read your notes (with your permission), compose a summary using proper OFM formatting, and save it exactly where you asked.

For a broader collection of tools and skills, explore claudeskills.info — including skills for research, productivity, and knowledge management that pair well with Obsidian-based workflows.

---

Sources: kepano/obsidian-skills on GitHub · Steph Ango's announcement on X · Obsidian Skills deep-dive on Medium