Inspired by kepano/obsidian-skills.
Source: kepano/obsidian-skillsReverse-engineered from real GitHub workflow.
Agent skills for Obsidian. Teach your agent to use Markdown, Bases, JSON Canvas, and use the CLI.
Agent skills library enabling AI agents to create and manipulate Obsidian files, bases, canvas diagrams, and CLI operations
gemini -p "Sync my project notes to Obsidian"I'm creating an agent skills library for Obsidian that teaches AI agents how to work with Obsidian's ecosystem. The goal is to make it easy for any compatible agent (Claude Code, Cursor, or other AI tools) to create and manipulate Obsidian files and vaults.
The core idea is following the Agent Skills specification, which means each skill is self-contained and can be discovered and used by any compliant agent. I need five main skill modules:
The first skill teaches agents to work with Obsidian Markdown files. This includes creating and editing .md files with Obsidian-specific features like wikilinks, embeds, callouts, and YAML properties. Agents should understand how to write valid Obsidian Flavored Markdown syntax.
The second skill is for Obsidian Bases, which are a newer database-like feature. Agents need to understand how to create and edit .base files with views, filters, formulas, and summaries. This is structured data inside Obsidian.
The third skill handles JSON Canvas, which is a separate open standard for creating visual diagrams. Agents should be able to create and edit .canvas files with nodes, edges, groups, and connections to represent knowledge graphs visually.
The fourth skill interfaces with the Obsidian CLI, letting agents interact with vaults programmatically, including managing plugins and themes, and doing development-related tasks.
The fifth is a utility skill that wraps the Defuddle tool to extract clean markdown from web pages, removing clutter and noise to save tokens when processing web content.
Each skill should be a folder containing a SKILL.md file that describes its interface and capabilities in a format that agents can parse and understand. The project should support multiple installation methods: it can be added via package managers, manually into Claude Code's .claude folder, copied to Codex's skills path, or cloned into OpenCode's skills directory.
The whole thing should be organized so that each skill is self-contained and discoverable. The skills follow a standard specification so they work across different agent platforms. Think of this as building a comprehensive toolkit that bridges the gap between AI agents and the Obsidian ecosystem, making it natural for agents to work with markdown, databases, canvas diagrams, and CLI operations.