What a Second Brain Lacks Isn't Collection—It's the Filing Step
Obsidian 1.15 early access lets Claude Code write content from the terminal under a specified heading. The filing step in a second brain loses one manual step, but you still have to design the heading structure yourself.
Obsidian 1.15 early access adds a small feature: Claude Code can write text from the terminal under any heading in any note, while leaving the rest of the file unchanged. The CLI has been built in since 1.12; 1.15 adds the “locate the heading” step. Obsidian is a local Markdown note-taking tool, Claude Code is Anthropic's terminal AI agent, and the two are connected via the CLI.
The previous workflow is familiar: copy the agent output, find the corresponding note, paste, and fix the formatting. Now the agent puts the content under the specified heading itself. The original post gave three scenarios: after Opus 5.5 runs, logs go under ## log in today's note; meeting summaries go under ## decisions in the project note; new sources from a research pass go under ## reading. Open Obsidian in the morning, and the previous day's work is already under the corresponding heading.
This feature sounds small, but it touches the most common way a second brain dies: it gets built and then sits gathering dust. There are many collection tools, but the filing entry point is too far away. A commenter in the replies said they only append content to the bottom of the daily note, and after a few days it becomes unreadable; being able to target a specific heading is what makes this usable. Another person asked whether it automatically creates the heading when it doesn't exist or fails. The original post doesn't elaborate; with the early access version, you have to try it yourself.
Karpathy has mentioned the idea of having models maintain an LLM wiki: the model saves notes, and humans only read. Mr. Buzzoni says that's exactly the part most setups skip. Links below. Some also did the math and said a $2,100-a-month second brain sounds crazy. Another comment was more direct: a second brain is only valuable if you actually open it; most people build a vault and never look at it again. Harsh, but consistent with experience. Tools solve where to write, not whether you open it.
If you're already using Obsidian and Claude Code, you can try this: confirm the CLI is available, upgrade to 1.15 early access, have the agent write under a specified heading, then open the note and check the formatting. Don't expect it to organize automatically. The heading structure has to be designed first; the clearer fixed entry points like ## log, ## decisions, and ## reading are, the less hassle later. Otherwise you're just moving the junk drawer somewhere else.
Links:
- Obsidian: https://obsidian.md
- Obsidian CLI: https://obsidian.md/help/cli
- Claude Code: https://claude.ai
- Karpathy's LLM wiki: https://gist.github.com/karpathy/442a6bf555914893e9891c11519de94f
Quoted original post: https://x.com/Mr_Buzzoni/status/2116431980862570724
Quoted discussion: https://x.com/Mr_Buzzoni/status/2115443090068516935
发布时间: 2026-10-31 16:29