Obsidian Second Brain
Turns your Obsidian vault into a self-rewriting knowledge base that tidies itself up — even overnight
What it does & when to use it
An open-source (MIT) community skill that turns an Obsidian vault into a knowledge system built specifically for Claude to read and write — not just appending new content, but rewriting existing pages, reconciling contradictions, and giving every note a self-explaining "for future Claude" preamble. The approach builds on Karpathy's LLM Wiki pattern and extends it into 44 slash commands, calendar management, and background maintenance agents.
Ships 44 slash commands covering day-to-day vault operations: one command to save everything worth remembering from a conversation (auto-split across five parallel subagents for people/projects/tasks/decisions/ideas), daily-note generation and updates, kanban task management, dev-log entries, and tracking of recurring obligations (like a monthly bill) with computed next-due dates. A research toolkit (7 commands — /x-read, /research, /research-deep, /notebooklm, /youtube, /podcast) pulls from Grok, Perplexity, NotebookLM and other sources and saves findings straight into the vault. It also supports Google Calendar integration, scheduled background agents that maintain the vault while you're away, and a per-project-vault mode for multi-repo setups. Every write follows a consistent "AI-first" spec: a preamble summary, full frontmatter metadata, source attribution with recency markers, and mandatory bidirectional links.
- Say "save what we just discussed" after a meeting and have decisions, action items, and mentioned people filed into the right places automatically
- Have Claude generate a daily briefing that merges your calendar, overdue kanban items, and related vault notes
- Track a recurring obligation (tax filing, subscription renewal) with an automatically computed next-due date
- Run a deep-research session with the /research command family and have the findings land in the vault instead of getting lost in chat history
- Maintain separate vaults per project or client with the same consistent operating habits
Pros & cons
- MIT-licensed — unlike the anthropics official document skills, you can freely fork, extend, and redistribute it
- 44 commands genuinely cover a full workflow — notes, scheduling, recurring obligations, deep research — this isn't a toy example
- The "AI-first note format" design is genuinely thought through: every note carries a summary and recency markers, built specifically so a future Claude session can re-orient itself quickly, rather than being formatted for a human skimming it
- The surface area is large (44 commands + background agents + calendar integration) — expect to spend real time reading `_CLAUDE.md` and the vault-schema conventions before getting full value out of it
- Calendar features (the schedule/reconcile modes of /obsidian-calendar) depend on the Google Calendar MCP and are Claude-Code-only; other clients don't get this part
- The AI-first note format (heavy frontmatter, self-explaining preambles) isn't especially pleasant for a human to read directly — it's optimized for Claude, so if you also want the notes to be nice for manual browsing you may need to adapt it
How to install
One-line install and bootstrap:
curl -sL https://raw.githubusercontent.com/eugeniughelbur/obsidian-second-brain/main/scripts/quick-install.sh | bashIt asks for a vault path, your name, and an optional preset (executive/builder/creator/researcher — each with a different kanban structure and note emphasis). You can also run python scripts/bootstrap_vault.py --path ~/my-vault --name "Your Name" manually. Once installed, the skill operates on the vault through standard file read/write tools — no MCP server is required, though the repo ships an optional one mainly for non-Claude-Code clients.
How to use
In Claude Code or another skill-compatible client, use the slash commands or plain language — e.g. /obsidian-save to save what's worth keeping from the current conversation, /obsidian-daily to generate today's note, or just say "log this decision we just made." The first time it touches a vault, it reads _CLAUDE.md (the vault's operating-rules file) and offers to create one if it's missing.
Compared to similar skills
Compared to just letting Claude read and write plain Markdown notes, the core difference is that notes here are treated as living documents — new information rewrites existing pages by default instead of blindly appending, and mentions of the same person or project across the vault get cross-linked instead of scattered. Compared to a simple "scratchpad" note-taking skill, this is closer to a full personal knowledge management (PKM) system, with a correspondingly higher setup and mental-model cost.