The Second Brain Built by 70K Stars: The Hardest Step Isn't Installation

Four free repositories claim they can build your Claude-powered Obsidian second brain for you. At the time of posting, their combined star count was over 70k.

1. obsidian-skills, about 50k stars

Author kepano, CEO of Obsidian. Six skills that teach agents to write Obsidian-flavored Markdown, Bases tables, and JSON Canvas canvases, operate a vault through the official CLI, use Defuddle to extract clean Markdown from web pages to save tokens, and use Knap to batch-render templates from JSON or CSV. It follows the Agent Skills spec and can be installed in Claude Code, Codex, and OpenCode. Two plugin marketplace commands:

/plugin marketplace add kepano/obsidian-skills
/plugin install obsidian@obsidian-skills

https://github.com/kepano/obsidian-skills

It is the only one of the four that comes from the official Obsidian team.

2. claude-obsidian, about 16k stars, MIT

Four core commands: /wiki initializes or takes over an existing vault; /wiki-ingest turns sources in inbox/ into linked notes with citations; /wiki-query provides read-only Q&A; /wiki-lint checks for dead links, orphan pages, and missing fields. Behind it are 15 skills and 3 read-only agents, designed based on Andrej Karpathy's LLM Wiki pattern.

The write flow deserves a separate mention. It first reads every note to be changed, records a SHA-256 hash, merges all agents' drafts into a single plan, and commits it all at once as one transaction. If power is lost midway, the journal can roll back. If a note you edited in Obsidian has a mismatched hash, the write is aborted directly and won't overwrite you. Requires Python 3.11+. On Windows, you can only write through WSL; native Windows is view-only.

https://github.com/AgriciDaniel/claude-obsidian

3. obsidian-second-brain, about 5,000 stars, MIT

48 commands, builds for 8 platforms. What's more unusual is Crew: Claude Code, Codex, Cursor, and Grok Bot share one log file; each writes a line after finishing a task, and other agents read it first when they start.

It introduces a concept called OKM, Open Knowledge Metabolism. Every stored fact is either eternal, date-stamped, or a pointer to a live source. The goal is to keep the knowledge base from filling up with things that used to be true and no longer are.

A few commands: /obsidian-challenge uses history in your vault to argue against your new ideas; /obsidian-ingest rewrites 5 to 15 existing pages from a single URL; /research-deep first scans your existing notes and then fills the gaps.

Things to note: background agents are disabled by default; enabling them requires --dangerously-skip-permissions. Its tool surface is locked to Read, Write, Edit, Glob, and Grep; it cannot run a shell or access the network, and it only adds, never deletes.

https://github.com/eugeniughelbur/obsidian-second-brain

4. second-brain-starter

Author Cole Medin. Fill out a requirements template, and it generates a nine-stage personalized build plan.

It is a plan, not runnable software. The repository has no license.

https://github.com/coleam00/second-brain-starter

Read the warnings first

The original post ends with a few notes that are more useful than the list itself. Only the first comes from the Obsidian CEO; the rest are community projects. The fourth is a plan, not code, and it has no license.

One commenter said the warning section of posts like this is often worth more than the list, and most posts skip it entirely. Another gave a screening order: first look at stars, then look at who wrote it, and before installing, go through the skill files yourself to see what it actually tells the agent to do to your notes.

A few hard-to-answer questions

One commenter framed outsourcing memory in physiological terms: if the hippocampus stops exercising, synapses that don't fire get pruned, and the vault ends up as just a more respectable tombstone.

Another said their second brain is just a Markdown folder plus an index file. Their judgment: building it was never the hard part; deleting the notes that are no longer right is.

Someone else asked whether kepano's skills and claude-obsidian's wiki commands can run together. Both teach agents to work in a vault; whether stacking them is synergy or conflict has no clear answer.

One observation

Of the four repositories, two deal with how to undo bad writes, one deals with how old facts should retire, and one just gives you a blueprint.

Install them all, and what gets solved is still just setup. On the day you finish installing, your vault is empty. After that, whether every write is correct and whether every fact still holds—these repositories cannot do that for you.