second-brain

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

/second-brain — Karpathy LLM Wiki workflow

/second-brain — Karpathy LLM Wiki工作流

Wraps an existing Second Brain in Obsidian (or any markdown-based vault). The wiki vault's CLAUDE.md is the authoritative schema — the skill orchestrates the operations the user has been doing manually.

本工具封装了Obsidian（或任何基于markdown的笔记库）中的现有第二大脑。笔记库中的CLAUDE.md是权威架构文件——本技能将用户手动执行的操作进行编排。

Mental model

心智模型

Three layers, each with a clear role:

raw/      →  wiki/         →  outputs/
sources      compiled         generated
                              artifacts

raw/ — unprocessed source material. Articles, highlights, ideas, braindumps, tweets. Type-prefixed (
```
article-
```
,
```
idea-
```
,
```
highlights-
```
,
```
braindump-
```
,
```
note-
```
,
```
resource-
```
,
```
tweet-
```
). Never deleted — source of truth.
wiki/ — AI-compiled topic pages. One page per concept, not per source. Interlinked via
```
[[wikilinks]]
```
.
```
INDEX.md
```
at root.
outputs/ — generated artifacts from queries: research summaries, analyses, slide decks. Named descriptively.

Folders to leave alone during wiki ops:

Projects/

Daily/

Templates/

Inbox/

Notes/

Tasks.md

Kanban.md

Home.md

分为三层，每层职责明确：

raw/      →  wiki/         →  outputs/
原始源文件      编译后页面         生成的产物

raw/ — 未处理的源素材。包括文章、高亮内容、想法、头脑风暴记录、推文。文件以类型前缀命名（
```
article-
```
、
```
idea-
```
、
```
highlights-
```
、
```
braindump-
```
、
```
note-
```
、
```
resource-
```
、
```
tweet-
```
）。永不删除——是事实来源。
wiki/ — AI编译的主题页面。每个概念对应一个页面，而非每个源文件对应一个页面。通过
```
[[wikilinks]]
```
实现相互关联。根目录下的
```
INDEX.md
```
为索引文件。
outputs/ — 查询生成的产物：研究摘要、分析报告、幻灯片。命名需清晰描述内容。

在wiki操作过程中，请勿修改以下文件夹：

Projects/

、

Daily/

、

Templates/

、

Inbox/

、

Notes/

、

Tasks.md

、

Kanban.md

、

Home.md

。

Step 1 — Load vault config + schema

步骤1 — 加载笔记库配置与架构

Read

references/vault-config.md

for the vault path (default:

${SECOND_BRAIN_VAULT:-$HOME/Documents/SecondBrain}/

)

Read
```
<vault>/CLAUDE.md
```
for the authoritative schema. If present, trust it over
```
references/schema.md
```
— the user's vault is the source of truth.
If no
```
<vault>/CLAUDE.md
```
, fall back to
```
references/schema.md
```
.

读取

references/vault-config.md

获取笔记库路径（默认：

${SECOND_BRAIN_VAULT:-$HOME/Documents/SecondBrain}/

）

读取
```
<vault>/CLAUDE.md
```
获取权威架构。如果存在该文件，则优先信任它而非
```
references/schema.md
```
——用户的笔记库是事实来源。
如果不存在
```
<vault>/CLAUDE.md
```
，则回退使用
```
references/schema.md
```
。

Step 2 — Parse mode

步骤2 — 解析模式

Invocation	Mode
`/sb capture` / `/second-brain capture` / "capture this" / "save this to my brain"	capture
`/sb compile` / "compile the wiki" / "process raw notes"	compile
`/sb query <question>` / "ask my brain X" / "what does my brain say about Y"	query
`/sb lint` / "lint the wiki" / "health check my brain"	lint
`/sb connect` / "find connections" / "suggest wikilinks"	connect
`/sb search <term>` / "search my notes for X"	search

触发指令	模式
`/sb capture` / `/second-brain capture` / "capture this" / "save this to my brain"	捕获
`/sb compile` / "compile the wiki" / "process raw notes"	编译
`/sb query <question>` / "ask my brain X" / "what does my brain say about Y"	查询
`/sb lint` / "lint the wiki" / "health check my brain"	检查
`/sb connect` / "find connections" / "suggest wikilinks"	关联
`/sb search <term>` / "search my notes for X"	搜索

Step 3 — Run the mode

步骤3 — 运行对应模式

capture

捕获

Inputs: URL, pasted text, file path, or screenshot.

Detect type from content:
- URL →
```
article-
```
- Pasted text with quoted highlights →
```
highlights-
```
- User's own thoughts / brainstorm →
```
braindump-
```
  or
```
idea-
```
- Single tweet / X post →
```
tweet-
```
- PDF, video, podcast →
```
resource-
```
- Quick reference (recipe, command, fact) →
```
note-
```
- If ambiguous, ask.
Generate a descriptive filename:
```
<type>-<kebab-case-topic>.md
```
(e.g.,
```
article-andrew-wilkinson-tiny-manual.md
```
). Use the source title or topic — not the URL slug.

Add metadata to the top:

markdown

source: <URL if applicable>
captured: YYYY-MM-DD

Save to
<vault>/raw/
.
If the source is a URL, fetch the article content (via WebFetch or agent-browser for auth-walled) and save the readable text — not just the URL.
Report path + a one-line summary of what was saved.

Don't compile into the wiki here — capture is fast intake. Compilation is a separate, deliberate pass.

输入：URL、粘贴的文本、文件路径或截图。

根据内容检测类型：
- URL →
```
article-
```
- 带引用高亮的粘贴文本 →
```
highlights-
```
- 用户自身想法/头脑风暴内容 →
```
braindump-
```
  或
```
idea-
```
- 单条推文/X帖子 →
```
tweet-
```
- PDF、视频、播客 →
```
resource-
```
- 快速参考内容（食谱、命令、事实） →
```
note-
```
- 若类型不明确，则询问用户。
生成描述性文件名：
```
<type>-<kebab-case-topic>.md
```
（例如：
```
article-andrew-wilkinson-tiny-manual.md
```
）。使用源标题或主题命名——而非URL后缀。

在顶部添加元数据：

markdown

source: <URL（若适用）>
captured: YYYY-MM-DD

保存至
<vault>/raw/
。
如果源是URL，则获取文章内容（通过WebFetch或agent-browser处理需要权限访问的内容）并保存可读文本——而非仅保存URL。
反馈保存路径及内容的一行摘要。

请勿在此阶段将内容编译至wiki中——捕获是快速收录操作，编译是单独的、需主动触发的步骤。

compile

编译

The expensive but valuable operation. Process unprocessed raw files into wiki pages.

Find unprocessed raw files: grep
```
wiki/*.md
```
for
```
Sources
```
sections; the raw files NOT listed are unprocessed.
Read each unprocessed raw file + the existing
```
wiki/INDEX.md
```
.
For each raw file:
- Extract key concepts, facts, insights
- Default: merge into an existing wiki page if the topic overlaps. Only create a new page if the concept doesn't fit anywhere.
- One page per concept, not per source.
- Use
```
[[wikilinks]]
```
  for every related concept
- Add the raw file under the wiki page's
```
## Sources
```
  section with a one-line note on what was drawn from it
Update
wiki/INDEX.md
:
- Add new pages under their category (Creative / Health & Longevity / Faith & Personal Growth / Business / Personal Growth / Tech / Hobbies / Sci-Fi / Pets — or new category if needed)
- One line per entry:
```
- [[Page Name]] — brief description
```
Connections section is mandatory on every wiki page. If a new page has no connections, find one before saving.
Quality > quantity. If a page would be <100 words, hold the raw file for now and ask the user if it should be merged into an adjacent page.

Output: list of pages created/updated, what merged where, anything held for clarification.

这是耗时但有价值的操作。将未处理的raw文件转换为wiki页面。

查找未处理的raw文件：在
```
wiki/*.md
```
的
```
Sources
```
章节中搜索；未被列出的raw文件即为未处理文件。
读取每个未处理的raw文件 + 现有
```
wiki/INDEX.md
```
。
针对每个raw文件：
- 提取关键概念、事实、见解
- 默认规则：若主题重叠，则合并至现有wiki页面。仅当概念无法适配任何现有页面时，才创建新页面。
- 每个概念对应一个页面，而非每个源文件对应一个页面。
- 对每个相关概念使用
```
[[wikilinks]]
```
- 将raw文件添加至wiki页面的
```
## Sources
```
  章节下，并附上一行说明从中提取的内容
更新
wiki/INDEX.md
：
- 将新页面添加至对应分类下（创意 / 健康与长寿 / 信仰与个人成长 / 商业 / 个人成长 / 科技 / 爱好 / 科幻 / 宠物 —— 若需要可新增分类）
- 每条条目一行：
```
- [[Page Name]] — 简短描述
```
每个wiki页面必须包含关联章节。如果新页面没有关联内容，在保存前需找到至少一个关联页面。
质量优先于数量。如果页面内容不足100字，暂时保留raw文件，并询问用户是否应将其合并至相邻页面。

输出：创建/更新的页面列表、合并情况、需澄清的内容。

query

查询

Answer a question using ONLY the wiki/raw corpus. Different from

deep-research

(which goes external).

Read
wiki/INDEX.md
to identify potentially relevant pages
Read those pages + traverse
```
[[wikilinks]]
```
1–2 hops
Compose the answer:
- Cite wiki pages by name: "Per [[Microplastics Detox]]..."
- If the wiki contradicts itself, surface both sides
- If the wiki doesn't contain the answer, say so and offer to run
```
/deep-research
```
  to expand
Save to
outputs/<YYYY-MM-DD>-<question-slug>.md
with:
- The original question
- The answer
- List of wiki pages consulted
Show the answer in chat + path to the saved output
Optional render: if
```
--render pdf
```
or
```
--render html
```
was passed, pipe the output through pandoc using the shared stylesheet. See
```
references/schema.md
```
→ "Publishing alternatives" for the commands.

仅使用wiki/raw语料库回答问题。与

deep-research

（调用外部资源）不同。

**读取
```
wiki/INDEX.md
```
**以识别潜在相关页面
读取这些页面 + 遍历1–2层
```
[[wikilinks]]
```
关联页面
撰写答案：
- 通过页面名称引用："根据[[Microplastics Detox]]..."
- 如果wiki内容存在矛盾，需同时呈现双方观点
- 如果wiki中没有答案，需告知用户并提议运行
```
/deep-research
```
  来补充内容
保存至
outputs/<YYYY-MM-DD>-<question-slug>.md
，包含：
- 原始问题
- 答案
- 参考的wiki页面列表
在聊天中展示答案 + 保存的输出文件路径
可选渲染：如果传入了
```
--render pdf
```
或
```
--render html
```
参数，使用共享样式表通过pandoc处理输出内容。详见
```
references/schema.md
```
→ "发布替代方案"中的命令。

lint

检查

Health check the wiki.

Check:

Orphan pages — wiki/*.md that aren't in INDEX.md
Connection orphans — pages with no
```
[[wikilinks]]
```
to other pages
Unprocessed raw — raw files not listed under any wiki page's Sources
Stale pages — most recent source >6 months old AND topic is volatile (AI, marketing, finance, health protocols)
Topic gaps — concepts mentioned in 3+ pages without their own dedicated page
Contradictions — wiki pages making opposing claims without flagging it
Missing connections — pages on clearly related topics with no
```
[[wikilink]]
```
between them (suggest
```
/sb connect
```
)

Output: prioritized list. Most important first (broken structure beats stale content).

对wiki进行健康校验。

检查项：

孤立页面 —
```
wiki/*.md
```
中未出现在INDEX.md中的页面
关联孤立页面 — 未链接到其他页面的页面（无
```
[[wikilinks]]
```
）
未处理的raw文件 — 未被任何wiki页面的Sources章节列出的raw文件
过时页面 — 最新源文件已超过6个月且主题属于易变领域（AI、营销、金融、健康方案）
主题缺口 — 在3个及以上页面中被提及但无专属页面的概念
矛盾内容 — wiki页面中存在对立观点但未标注的情况
缺失关联 — 主题明显相关但未通过
```
[[wikilink]]
```
连接的页面（建议运行
```
/sb connect
```
）

输出：按优先级排序的列表。最重要的问题排在最前面（结构问题优先于过时内容）。

connect

关联

Find pages that should be linked but aren't.

Build a topic map from INDEX.md + page summaries
For each page, find 2–5 other pages with thematic overlap
Check whether each candidate is already linked
Suggest the missing links — and if the user approves, edit the pages to add them to their
```
## Connections
```
sections

查找应该链接但未链接的页面。

根据INDEX.md和页面摘要构建主题图谱
为每个页面找到2–5个主题重叠的其他页面
检查每个候选页面是否已链接
建议缺失的链接——如果用户同意，编辑页面将链接添加至其
```
## Connections
```
章节

search

搜索

Quick grep across

wiki/

raw/

for a term. Return matching files with a 2-line excerpt around the match. Faster than

query

when the user knows what page they're looking for.

在

wiki/

raw/

中快速搜索指定术语。返回匹配文件及匹配位置前后2行的摘录。当用户明确知道要找的页面时，此操作比

query

更快。

Composes with

可组合使用的技能

```
deep-research
```
— when
```
query
```
finds gaps in the wiki, route to deep-research to expand from external sources. Deep-research output can be captured back into
```
raw/
```
for future compilation.
```
paste
```
— capture content cleanly into
```
raw/
```
(especially for terminal/CLI captures).

business-brainstorm

— checks

Portfolio of Businesses

and

Entrepreneurship & Startups

wiki pages for relevant context before brainstorming.

```
decide
```
— pull from
```
Personal Philosophy
```
/
```
Productivity & Systems
```
wiki for principles when scoring Q34 ("what principles are we bending"). New: a
```
Decision Log
```
wiki page accumulates the narrative form of decisions over time (the
```
decide
```
archive is the structured form; the wiki page is the story).
```
jab-hook
```
— a
```
Content Ideas
```
wiki page hoppers hooks, frameworks, and stories.
```
/jab-hook
```
drafts pull candidates from there.
```
slide-deck
```
— content drafted in
```
outputs/
```
becomes deck source; speaker notes can reference relevant wiki pages.
```
pm
```
— Projects/ folder in the vault is off-limits to second-brain; pm owns it. But a
```
Workflow Docs
```
wiki page captures operational patterns that show up across multiple projects.

```
deep-research
```
— 当
```
query
```
发现wiki存在内容缺口时，可调用deep-research从外部资源补充内容。deep-research的输出可被捕获回
```
raw/
```
，用于后续编译。
```
paste
```
— 将内容干净地捕获至
```
raw/
```
（尤其适用于终端/CLI场景下的捕获）。
```
business-brainstorm
```
— 在头脑风暴前，检查wiki中的
```
Portfolio of Businesses
```
和
```
Entrepreneurship & Startups
```
页面获取相关背景信息。
```
decide
```
— 在评分Q34（"我们正在违背哪些原则"）时，从wiki的
```
Personal Philosophy
```
/
```
Productivity & Systems
```
页面提取原则。新增功能：
```
Decision Log
```
wiki页面会逐步积累决策的叙事形式（
```
decide
```
的存档是结构化形式；wiki页面是故事形式）。
```
jab-hook
```
—
```
Content Ideas
```
wiki页面存储钩子、框架和故事。
```
/jab-hook
```
草稿会从中选取候选内容。
```
slide-deck
```
— 在
```
outputs/
```
中起草的内容可作为幻灯片源；演讲备注可参考相关wiki页面。
```
pm
```
— 笔记库中的Projects/文件夹属于second-brain的操作禁区；由pm技能管理。但
```
Workflow Docs
```
wiki页面会记录多个项目中出现的操作模式。

Sibling implementations (reference)

同类实现方案（参考）

Two other systems following the same raw → wiki → outputs pattern. Both are worth watching as upgrade paths.

Hermes'
llm-wiki
skill — off-the-shelf implementation of the 3-folder pattern. Pre-built workflows for compile / query / lint. Useful for comparing schema decisions.
Gbrain by Garry Tan — much more sophisticated. Treats the brain as a database (Postgres or PGLite) with synthesis, graph traversal, gap analysis, scheduled cron maintenance, and MCP integration. Powers a 146K-page deployment with 24K people entities. If the user's vault outgrows the markdown-only pattern, Gbrain is the upgrade direction. Borrows worth adopting today even without migrating: people-as-entities (the
```
person-
```
raw type +
```
People
```
wiki page) and scheduled maintenance (wire
```
compile
```
and
```
lint
```
to fire on a recurring schedule via the
```
loop
```
or
```
compound-engineering:schedule
```
skill).

另外两个遵循raw → wiki → outputs模式的系统。两者都可作为升级方向参考。

Hermes的
llm-wiki
技能 — 开箱即用的三文件夹模式实现。预构建了编译/查询/检查工作流。有助于对比架构决策。
Gbrain（Garry Tan开发） — 功能更复杂。将第二大脑视为数据库（Postgres或PGLite），支持合成、图谱遍历、缺口分析、定时cron维护和MCP集成。支撑着拥有14.6万页面、2.4万人物实体的部署。如果用户的笔记库超出纯markdown模式的承载能力，Gbrain是升级方向。即使不迁移，也可借鉴其现有功能：人物实体化（
```
person-
```
类型的raw文件 +
```
People
```
wiki页面）和定时维护（通过
```
loop
```
或
```
compound-engineering:schedule
```
技能定期触发
```
compile
```
和
```
lint
```
）。

Notes on quality

质量注意事项

Notes on quality

—

Quality over quantity. Fewer well-connected wiki pages beat many thin ones. Hold raw files for clarification if compilation would produce a thin page.
Don't flatten nuance. If two raw sources contradict, the wiki page should note the disagreement, not pick a side silently.
Connections section is mandatory — every wiki page must link to at least one other page.
Never delete raw files after compilation. They're the source of truth.
Never modify files in
```
Projects/
```
,
```
Daily/
```
,
```
Templates/
```
,
```
Notes/
```
,
```
Tasks.md
```
,
```
Kanban.md
```
,
```
Home.md
```
, or
```
Inbox/
```
during second-brain operations. Those belong to other workflows.

质量优先于数量。少量关联良好的wiki页面优于大量内容单薄的页面。如果编译会产生内容单薄的页面，暂时保留raw文件并向用户确认。
不要弱化细节差异。如果两个raw源内容矛盾，wiki页面应标注分歧，而非默认选择某一方观点。
必须包含关联章节 — 每个wiki页面至少链接到一个其他页面。
编译后永不删除raw文件。它们是事实来源。
在second-brain操作期间，绝不修改
```
Projects/
```
、
```
Daily/
```
、
```
Templates/
```
、
```
Notes/
```
、
```
Tasks.md
```
、
```
Kanban.md
```
、
```
Home.md
```
或
```
Inbox/
```
中的文件。这些属于其他工作流的范畴。