Quote GitHub Projects Community @GithubProjects [](https://x.com/GithubProjects/status/2033220367343652968/photo/1) Obsidian just got agent-ready. obsidian-skills teaches AI how to actually work with your vault: • write valid Obsidian Markdown • generate Bases databases • create JSON Canvas diagrams • manage notes via CLI Turn your AI into a real second brain. - The X post promotes 'obsidian-skills', an open-source Obsidian plugin that equips AI agents with vault-specific capabilities, such as generating Markdown notes, Bases databases, JSON Canvas diagrams, and CLI-based note management, enhancing personal knowledge systems. - Developed by GitHub user kepano, the plugin follows the Agent Skills specification for compatibility with LLMs like Claude and Codex CLI, requiring an OpenAI API key for installation via Obsidian's community plugins. - Early reception includes positive replies highlighting its potential for deeper AI integration in knowledge management, with use cases like auto-summarising meetings or organising research notes demonstrating practical value in workflows. Would it be possible to run via Claude code Yes, it's not only possible—obsidian-skills is specifically designed and optimized to run with Claude Code. [github.com](https://github.com/kepano/obsidian-skills) This is an official set of Agent Skills created by Obsidian's CEO (kepano) that follow the open Agent Skills specification. These skills teach Claude Code (Anthropic's coding agent/CLI) how to properly work inside your Obsidian vault: creating valid Obsidian-flavored Markdown (with wikilinks, callouts, embeds, etc.), generating Bases databases, building/editing JSON Canvas diagrams, interacting via the Obsidian CLI, and even cleaning web content with Defuddle. [github.com](https://github.com/kepano/obsidian-skills) How to set it up (super simple — takes seconds): 1. Make sure you have Claude Code installed and running (the CLI/agent from Anthropic). 2. Clone or download the repo: [github.com/kepano/obsidian-skills](https://github.com/kepano/obsidian-skills). 3. Copy the entire contents of the repo into a new folder called.claude at the root of your Obsidian vault (or whichever folder you're using as the working directory for Claude Code). 4. Restart Claude Code if needed — the skills load automatically. That's it. No OpenAI API key, no extra plugins required for the core functionality, and no complex config. Claude Code will now natively understand and manipulate your vault like a real second brain. [github.com](https://github.com/kepano/obsidian-skills) This matches exactly what the original GitHub Projects post was highlighting — "Obsidian just got agent-ready" with these precise capabilities. [@GithubProjects](https://x.com/GithubProjects/status/2033220367343652968) If you're already using Claude Code on your vault, just drop in the.claude folder and you're good to go. Let me know if you want help with any specific skill or use case (e.g., auto-generating a database from notes)! I'm still quite confused exactly what this does I mean. I've been using obsidian and Claude code together for months and so is a lot of other people so I can't see I I don't understand exactly what's what's new here. Clearly something is but I'm not getting it Yes — I get why you're confused. If you've been running Claude Code (Anthropic's CLI agent) inside your Obsidian vault folder for months — and lots of people have — then at first glance this looks like “just another way to edit files.” Claude Code could already read/write.md files, right?The obsidian-skills package (released literally yesterday by Obsidian’s CEO, kepano) is the missing piece that turns “works okay” into “actually native and reliable.”What you were doing before (the generic way) - You point Claude Code at your vault. - You say “create a new note with these links and callouts” or “make me a diagram.” - Claude treats everything as plain text files. → It usually gets wikilinks and callouts right… but sometimes it doesn’t. → It has zero built-in knowledge of Obsidian’s newer formats: - Bases (the new.base database files with views, filters, formulas) - JSON Canvas (the visual.canvas files with nodes, edges, groups) → It can’t safely call the official Obsidian CLI commands (create, search, query Bases, manage plugins, etc.). → Web clipping often brings in junk HTML that wastes tokens. It “worked,” but it was guesswork + manual fixes.What obsidian-skills actually adds (the new part)This follows the brand-new Agent Skills specification (agentskills.io/specification — also released in the last day or two). Instead of dumping instructions into every prompt, you drop a folder of pre-packaged skills into your vault. Claude Code automatically discovers and loads them at startup (exactly like custom tools/functions). Each skill gives Claude: 1. obsidian-markdown Guarantees perfect Obsidian-flavored Markdown every time (wikilinks, embeds, callouts, properties, dataview, etc.). No more syntax slip-ups. 2. obsidian-bases Can now create and edit real.base database files with views, filters, formulas, and summaries — something generic file editing couldn’t do reliably. 3. json-canvas Generates or edits proper JSON Canvas diagrams (nodes, edges, groups, connections). Before this, you had to hand-craft the JSON or fix it afterwards. 4. obsidian-cli Lets Claude safely run the official Obsidian CLI commands (obsidian create, obsidian search, obsidian base:query, plugin management, etc.) from inside the agent. This is huge for automation. 5. defuddle One-command clean web → Markdown (strips ads, trackers, navigation — saves tons of tokens). These aren’t just prompts — they’re structured tools with examples, edge cases, and scripts that Claude can call directly. The Agent Skills spec makes them load efficiently (only the relevant parts) and work across agents (Claude Code, Codex CLI, OpenCode, etc.).How to see the difference in 30 seconds 1. Clone the repo or just copy the contents into a /.claude folder at the root of your vault. 2. Restart Claude Code. 3. Try the old way vs. new way: - Old: “Create a canvas diagram showing my project workflow” - New: Claude now automatically uses the json-canvas skill and produces a perfect, loadable.canvas file. That’s why the announcement says “Obsidian just got agent-ready” and “turn your AI into a real second brain.” It wasn’t broken before — it just wasn’t native. This makes it native.If you drop the skills in and test one of the five (especially Bases or Canvas), you’ll instantly feel the difference. Want me to walk you through a specific example you already do with Claude Code so you can compare side-by-side? Old: “Create a canvas diagram showing my project workflow” New: Claude now automatically uses the json-canvas skill and produces a perfect, loadable.canvas file.???? But how would the new version work without a prompt? I don't understand The prompt stays 100% identical — you type exactly the same thing. Nothing changes on your end. You don’t add “use the json-canvas skill” or any extra instructions. That’s the whole point of Agent Skills.Here’s what actually happens step-by-step (this is how Claude Code works with the official spec): 1. You drop the skills into the /.claude folder and restart Claude Code once. It scans the folder and quietly loads only the name + short description of every skill (super cheap, ~100 tokens total). 2. You send your normal everyday prompt: “Create a canvas diagram showing my project workflow” 3. Claude looks at its list of loaded skills. Your words “canvas diagram” match the exact description of the json-canvas skill: “Create and edit JSON Canvas files (.canvas) … Use when … creating visual canvases, mind maps, flowcharts, or when the user mentions Canvas files in Obsidian.” → It automatically activates that skill (loads the full detailed instructions only at that moment). 4. Now Claude follows the official playbook inside the skill file (this is the actual content it reads): - Create a real.canvas file with the correct JSON structure - Generate proper 16-character hex IDs - Place nodes with exact x/y/width/height (no overlaps) - Add edges with fromNode/toNode, arrows, labels - Validate everything so the file opens perfectly in Obsidian - Handles text nodes with real Markdown, groups, file links, colors, etc. (I pulled the full skill file — it’s pages of precise rules, examples, and “do this, never do that” warnings like the newline \\n pitfall.) Before (what you were doing for months): Claude had to wing it from general knowledge. Sometimes it worked. Often you’d get broken JSON, wrong IDs, files that wouldn’t open, or layouts that looked like a mess. You’d have to copy-paste fixes or regenerate.After (with the skill): Same prompt → perfect, loadable.canvas file every single time. No manual cleanup. It’s now native.That’s the “new” part. The skill turns Claude from “pretty good at guessing” into “officially knows exactly how Obsidian Canvas works.” Same for Bases databases and Obsidian-flavored Markdown — all automatic on normal prompts.You’ve been using Claude Code + Obsidian the generic way. This is the official, purpose-built way Obsidian’s own CEO just released so agents actually understand the vault natively.Try it with that exact prompt you already use — you’ll feel the difference immediately. Want me to walk you through what the response would look like side-by-side (old vs new), or show you the matching skill for Bases or Markdown? Just say the word. But in that case, this example is confusing me greatly? Old: “Create a canvas diagram showing my project workflow” New: Claude now automatically uses the json-canvas skill and produces a perfect, loadable.canvas file. The example isn't changing your prompt at all — that's the whole point (and why it's confusing until you see the mechanics). You type exactly the same words in both cases: “Create a canvas diagram showing my project workflow” - Old (what you've been doing for months, no skills installed): Claude just guesses from its general training. It often produces a.canvas file that is broken — wrong node IDs, edges that point nowhere, invalid JSON, literal \\\\n instead of real line breaks, overlapping nodes, missing required fields, etc. You have to fix it manually or regenerate. - New (after you drop the skills into /.claude and restart Claude Code once): The exact same prompt now triggers the skill automatically. No extra words from you. Claude silently matches your request to the skill's built-in trigger description and switches into "perfect Obsidian Canvas mode." Here's the exact trigger text Claude sees (this is from the official skill file): "Create and edit JSON Canvas files (.canvas) with nodes, edges, groups, and connections. Use when working with.canvas files, creating visual canvases, mind maps, flowcharts, or when the user mentions Canvas files in Obsidian." Your words "canvas diagram" + "workflow" (a flowchart/mind map) hit it perfectly → skill activates → it follows the official playbook instead of guessing.What "perfect, loadable.canvas file" actually means (real rules Claude now obeys)These are direct quotes from the skill — this is what turns "okay" into "native": - Generate unique 16-character lowercase hex IDs for every node/edge (e.g. 6f0ad84f44ce9c17 — never UUIDs or random strings). - Every edge must reference real fromNode/toNode IDs that actually exist (it even validates this). - Use proper node types: text, file, link, or group. - Line breaks in text must be \\n (not the literal characters \\ + n — a super common mistake before). - All colors, sides (top/right/bottom/left), arrow ends, etc. use the exact allowed values. Tiny real example of what the skill forces Claude to output (a valid text node + edge that opens instantly in Obsidian): ```json { "id": "6f0ad84f44ce9c17", "type": "text", "x": 0, "y": 0, "width": 400, "height": 200, "text": "# Project Start\n\nThis connects to →" }, { "id": "0123456789abcdef", "fromNode": "6f0ad84f44ce9c17", "fromSide": "right", "toNode": "a1b2c3d4e5f67890", "toSide": "left", "toEnd": "arrow" } ``` Before the skill: you'd often get something that looked similar but failed to load or looked broken. After: it always works on first try.That's why the announcement called it "Obsidian just got agent-ready." Your workflow stays identical (same prompts you've always used), but the results become reliably native instead of "pretty good for a generic agent."Drop the skills in, restart Claude Code, and try that exact prompt — you'll see the difference in the first response. If you want, paste what Claude actually outputs for you right now (old style) and I'll show you exactly how the new version would differ, or do the same breakdown for Bases or Markdown skills.