The Best Obsidian and PKM Skills for Claude Code
The most interesting thing in the Obsidian skill ecosystem right now is not any single skill — it's who wrote them. When the person who ships the format also ships the skill, the scores move.
The milestone: the format owner came down to the field
For months, "Obsidian skills" meant community reverse-engineering. Then kepano — Obsidian's CEO/CTO — published an official trio, and it immediately redefined what the ceiling looks like in our six-dimension evaluation:
- obsidian-bases — 9.1. Not a note-taking helper. It teaches the agent
.basefiles, Obsidian's actual database layer: filters, formulas, views over your vault as queried data. This is the skill that turns a folder of markdown into something an agent can slice. - obsidian-cli — 8.7. Vault reads and writes through a CLI, which means batch operations that don't depend on the agent improvising file edits across thousands of notes.
- obsidian-markdown — 8.7. Obsidian Flavored Markdown as a spec: embeds, callouts, wiki-link resolution, the bits that CommonMark says nothing about.
Why does this matter more than the numbers? Because a vault is only useful to an agent if the agent shares the owner's model of the vault. kepano's skills encode that model at the source. When obsidian-bases describes how a .base file evaluates, it isn't summarizing documentation — it's restating intent. That shows up directly in our structure and content scores, where all three official skills are near the top of the entire PKM category.
The 1.2-point gap is an ownership signal, not a talent gap
The best community series — jykim's — is genuinely good work. obsidian-links at 7.9 handles backlinks and unresolved references carefully. obsidian-yaml-frontmatter, also 7.9, is the right tool for the single most automatable PKM chore: property hygiene across a vault. obsidian-mermaid (7.5), obsidian-canvas (7.3), and obsidian-markdown-structure (7.2) cover the remaining surface.
obsidian-mermaid 7.5 Create Obsidian-compatible Mermaid diagrams. Prefer square layouts (TB + subgraph LR), avoid mark… ⬇ 4 obsidian-canvas 7.3 Create and manage Obsidian Canvas files with automatic layout generation. Use when creating visua… ⬇ 4 obsidian-markdown-structure 7.2 Validate and enforce markdown document structure including frontmatter positioning, heading hiera… ⬇ 2But look at the shape of the gap. 9.1/8.7/8.7 versus 7.9/7.9/7.5/7.3/7.2. The community skills lose points in the same places: edge-case behavior, spec ambiguity, and what we bucket as engineering — how cleanly the skill tells an agent to verify its own output against the real format.
That pattern is diagnostic. A community author documenting wikilinks has to guess at precedence questions: what happens with a heading link into an unresolved file, how aliases interact with frontmatter, which characters break a block reference. The format owner doesn't guess. The difference between "a well-written summary of Obsidian" and "Obsidian, expressed as instructions" is exactly the difference between 7.9 and 9.1.
This is the argument for why official authors shipping skills is an ecosystem milestone rather than a nice-to-have: every format with a canonical owner — Obsidian, GSAP, Remotion — has a class of defects that community skills structurally cannot eliminate, no matter how diligent the author.
What I'd actually install
For a working PKM setup with Claude Code, the honest stack is mostly official:
- obsidian-cli as the foundation — safe, scripted vault mutation beats ad-hoc file surgery every time.
- obsidian-markdown so generated notes parse and render correctly the first time.
- obsidian-bases once your vault has enough notes that "query it like a database" beats "grep it like a folder."
- Add obsidian-yaml-frontmatter from the community series for bulk property cleanup — it's the one place where a community skill fills a real hole in the official set.
- obsidian-links earns its slot if your workflow depends on graph integrity — finding orphans, repairing broken references at scale.
obsidian-mermaid, obsidian-canvas, and obsidian-markdown-structure are situational: install when a task needs them, don't carry them by default. Their trigger descriptions are broad enough that extra context rarely pays for itself.
The takeaway
The Obsidian corner of the skill ecosystem is now a controlled experiment in what makes agent skills good. Same domain, same tooling, same evaluation rubric — and the skills written by the person who owns the format beat the best community efforts by a full point and more, precisely on the dimensions that require knowing why the format behaves the way it does.
If you build tools on top of a format, the lesson generalizes: the highest-leverage skill author in your ecosystem is whoever controls the spec. And if you're choosing skills, stop reading bullet lists of features — check who wrote the thing. Browse the full evaluation data in the Skill123 directory.