Steph Ango's Official Obsidian Skills: How They Work + How I Use Them

TL;DR

Steph Ango is the CEO of Obsidian. In late 2025 he published an official set of Claude Skills built specifically for Obsidian users: small, composable workflows that give Claude vault-aware capabilities — capture a thought into the right note, retrieve context from across the vault, summarize across notes, link related ideas. They aren't magic and they aren't a replacement for a curated second brain. They're the missing tooling layer that sits between "I have a vault" and "the AI uses my vault effectively." This post covers what's in the official set, when each skill is actually useful, and the two-file pairing that makes them work for me daily.

Who Is Steph Ango (And Why That Matters)

Steph Ango is the CEO of Obsidian. He's been Obsidian's product voice for years and writes regularly about local-first software, file over app, and the long-term values that make Obsidian what it is. Anything he ships for Obsidian carries weight because he sets the direction of the product itself.

When he released a set of Claude Skills for Obsidian users in late 2025, two things were notable:

  • It's an official-feeling endorsement of Claude as the AI companion for Obsidian. Not Microsoft Copilot, not ChatGPT, not Gemini — Claude. There are technical reasons (Claude Code reads files natively, Anthropic's privacy stance aligns with Obsidian's local-first philosophy), but the choice also signals direction.
  • It's a working example of how skills should be structured. The skills are short, focused, and composable — each one does a single thing well, and they layer on top of any second-brain setup you've already built.

    • If you've been Obsidian-curious and waiting for a clear signal on how Obsidian and AI should fit together, this is it.

    What's in the Official Skill Set

    The skill set is intentionally small. Each skill is a few hundred lines of carefully-written instructions plus optional supporting scripts. The headline skills (subject to updates — check the GitHub repo for current state):

  • Capture. Add a thought to the right note in your vault. The skill reasons about where the thought belongs (active project? decision? new note?) rather than dumping into an inbox.
  • Retrieve. Pull relevant context from across your vault for the current question. Smart enough to skip stale notes and prioritize files marked as canonical.
  • Summarize. Condense across multiple notes into a coherent paragraph. Useful for end-of-week reviews or "what did I decide about X" queries.
  • Link. Suggest connections between the current note and existing notes — a backlink-aware version of what Obsidian's graph view does, but written for an AI to act on.
  • Daily. A skill that runs at the start of a day's session, loads identity + active projects, and sets the AI up to be useful immediately.

    • The list will evolve. The pattern won't: each skill is small, focused, and reads from the same vault you already write to.

    Why Skills Beat Custom Instructions

    Before skills existed, the way to give Claude vault-awareness was to paste long instructions into every prompt or into a Custom GPT-style system prompt. That approach has three problems:

  • It bloats the context. Every session pays the token cost.
  • It's hard to update. A single big instruction is fragile; editing one bullet risks breaking the rest.
  • It can't compose. You can't "use capture without retrieve" if everything is one giant prompt.

    • Skills solve all three. Each skill is loaded only when invoked, edits are local, and they compose naturally — capture then link then summarize is a workflow, not a monolith.

    The Two-File Pairing That Makes Them Useful

    Skills are tools. Tools without context produce generic outputs. The pairing that makes the official set actually work for me:

    File 1 — Your three core second-brain files (identity / projects / decisions). Without these, the skills don't know what to capture as, what to retrieve for, or what to link to. See the 3-file template post for the structure.

    File 2 — A skills index file (.obsidian/ai-skills.md or similar). A short file that lists which official skills are installed and when each should be invoked. Example: "Use daily at the start of a session. Use capture when I share a one-off thought. Use summarize only when I explicitly ask." This index acts as a meta-prompt that helps Claude pick the right skill without you having to name it every time.

    With these two pieces in place, the skills start behaving as advertised. Without them, they're just well-written prompts looking for context that doesn't exist.

    How I Actually Use Them Day to Day

    A realistic day:

  • Morning. Open Claude Code in the project I'm working on. The root CLAUDE.md loads, plus any folder-specific CLAUDE.md. The daily skill ran when I opened Claude.ai for chat-mode work; Claude already knows what's active.
  • During work. I drop one-liner thoughts into Claude ("client X mentioned they're worried about Y"). The capture skill decides whether that belongs in projects.md under that client, in decisions.md, or as a new note. I almost never have to say where to put it.
  • End of week. I run summarize against the week's captures + the project file. The output is a Friday-review draft I edit, not write from scratch.
  • When stuck. I invoke retrieve with the current problem. The skill pulls the 3–5 most relevant notes (sometimes including past decisions I'd forgotten I made) and feeds them as context.
  • When linking. When writing a new long-form note, I run link to suggest existing notes I should reference. About 30% of suggestions are useful, 70% are noise — useful enough to keep using.

    • None of this is exotic. It's the same workflow I had before the official skills, but with less friction.

    What the Skills Don't Do

    Setting expectations honestly:

  • They don't fix a stale vault. If your active-projects file is from three months ago, the skills will confidently surface stale context.
  • They don't replace the three-file template. They sit on top of it. Without an identity/projects/decisions base, the skills don't have enough signal to be useful.
  • They don't work outside Claude. They're Claude Skills specifically. The capabilities they encode (capture, retrieve, summarize) are universal, but the official implementation targets Claude.
  • They don't preserve your taste. They follow the instructions in your context files; if your taste isn't documented there, the skills won't infer it.

    • Installation (High Level)

    Specifics will drift as the official repo updates. The pattern:

  • Clone or download the official skills repo (linked from Steph Ango's posts / the Obsidian community).
  • Add the skills to your Claude.ai workspace via the Skills UI, or reference them from Claude Code via your settings.
  • Add the skills index file to your vault.
  • Test with a small task. If capture doesn't pick the right destination, edit your context files (not the skill) until it does.

    • The setup is one-time, maybe 30 minutes, and then it's invisible.

    Where to Go Next

  • If you don't have a second brain yet, start with the 3-file template.
  • If you have a vault but it's noisy for AI, read the pillar second-brain post for the curation principles.
  • If you're using ChatGPT and wondering whether to switch, the Claude + Obsidian comparison covers the trade-offs.

    • The official skills are the easiest way to test whether AI-augmented Obsidian fits your workflow. Spend a week with them on top of a real second brain and you'll know.