A Meta-Tool Is Born
Imagine working in Obsidian — double-bracket links, graph view, Canvas, Bases (database views). The surface area is already complex. Now you want an AI agent to manage the vault. When Claude Code opens your vault, it has no idea what a .canvas file looks like, or how a [[wikilink]] differs from a regular Markdown link.
That is the gap kepano/obsidian-skills fills. The author is not a random hobbyist — it is Steph Ango (kepano), CEO of Obsidian. This is not a third-party plugin; it is Obsidian's own operational manual for AI agents.
What Does It Provide?
Five Skills, each mapping to an Obsidian-specific data format or tool:
| Skill | Purpose |
|-------|---------|
| obsidian-markdown | Handles Obsidian-flavored Markdown: wikilinks, embeds, callouts, properties |
| obsidian-bases | Operates on .base files — Obsidian's database views with filters, formulas, summaries |
| json-canvas | Operates on .canvas files — the JSON Canvas visual board format |
| obsidian-cli | Invokes the Obsidian CLI for plugin and theme development |
| defuddle | Extracts clean Markdown from web pages, stripping noise to save tokens |
Think of it this way: plain Markdown is a "lingua franca" the AI already speaks. Obsidian's wikilinks, Canvas, and Bases are dialects. Without learning the dialect, an agent treats [[Note A]] as plain text instead of a link and .canvas files as opaque JSON.
The Agent Skills Specification: A Trend Worth Watching
The most important thing here is not any single Skill — it is compliance with the Agent Skills specification. The README states explicitly:
> These skills follow the Agent Skills specification so they can be used by any skills-compatible agent, including Claude Code, Codex, and Open Code.
In other words, these Skills are not locked to Claude Code. Codex, OpenCode, or any compliant agent can use the same Skill files. Installation paths differ (Claude Code uses .claude/, Codex uses ~/.codex/skills/, OpenCode uses ~/.opencode/skills/), but the Skill content is portable.
A useful analogy: Skill files are to AI agents what device drivers are to operating systems. Once upon a time every OS needed its own driver; then generic driver frameworks emerged. The Agent Skills specification is doing the same for AI capability packs — making them reusable across agents.
Why Did Obsidian's CEO Write This Personally?
The signal matters more than the project.
Obsidian's business model has always been selling Sync and Publish services while keeping the client free and local. The CEO personally authoring Agent Skills signals that Obsidian treats "AI operability" as infrastructure-grade investment.
The motivation is straightforward: if AI agents cannot correctly read and write Obsidian's file formats, users will migrate to AI-friendlier tools. Rather than waiting for third parties, Obsidian is defining the standard itself. This mirrors the logic of database vendors writing their own ORM drivers — whoever controls the interface controls the ecosystem.
Defuddle: An Underrated Skill
Among the five, defuddle is easiest to overlook but may deliver the highest practical value. Its job sounds simple: extract clean Markdown from a web page, stripping navigation, ads, and scripts.
The keyword is "save tokens."
When an agent fetches a web page directly, raw HTML for a complex page can consume 20K+ tokens on navigation and scripts alone. Defuddle distills the meaningful content into roughly 2K tokens. For agents billed per token and constrained by context windows, that is not a nice-to-have — it is real cost savings.
The functionality overlaps with generic web-reader skills, but Defuddle is native to the Obsidian ecosystem and plugs directly into Markdown-note workflows.
Relationship to Anthropic Skills
Trending the same day is anthropics/skills, Anthropic's official Skill repository. The two trending together is not coincidence — the Agent Skills ecosystem is forming quickly.
The distinction:
- anthropics/skills — Anthropic-maintained, general-purpose Skill collection
- kepano/obsidian-skills — first-party Skills for a third-party application (Obsidian)
- Repository: kepano/obsidian-skills
- Trending +411 stars on launch day
- Author: Steph Ango (Obsidian CEO)
- Specification: agentskills.io/specification
- License: not specified in the README
The relationship is analogous to "the standard library bundled with an OS" versus "an SDK shipped by an application vendor." You need the former to make agents run; you need the latter to make agents operate specific tools.
Lessons for Developers
For anyone building a tool product, obsidian-skills offers a clear template:
1. If your data format is not a standard format, it needs a Skill — otherwise the AI will treat it as unrecognized text. 2. Skills should follow the Agent Skills specification — do not lock to a single agent; give users choice. 3. First-party authorship carries more credibility than third-party work — users trust the vendor's understanding of its own formats.
Data
*The emergence of the Agent Skills specification signals a shift in the AI agent ecosystem from "every agent for itself" to "portable capabilities." Obsidian's CEO personally authoring Skills is an early signal that application vendors are actively adapting to the agent ecosystem.*