Zettelclaw
Follow the canonical Zettelclaw vault model: capture externally, write durable notes manually, and use the agent for navigation/synthesis.
This is an instruction-only skill. It does not install software by itself. qmd is optional when it is already available in the environment, and rg is the supported fallback for search.
Vault structure
CODEBLOCK0
Note types
Use YAML frontmatter on every note:
-
type
-
tags
-
created
- - Required on
doc and content notes:
-
status (
queued | in-progress | done | archived)
- - Optional content metadata:
-
author
- INLINECODE9
Use these primary types:
- -
note: durable atomic thinking note; no INLINECODE11 - INLINECODE12 : non-atomic working/reference note; uses INLINECODE13
- INLINECODE14 : daily log note; no INLINECODE15
- content types:
article, book, movie, tv, youtube, tweet, podcast, paper (and extensible additional content types); uses INLINECODE24
Templates
Always read the matching template in 03 Templates/ before creating a note:
- - INLINECODE26
- INLINECODE27
- INLINECODE28
Use core templates/date syntax. Do not require Templater.
Inbox workflow
- - Web captures land in
00 Inbox/ via clipper-capture.json. - Clipper sets
type by URL (tweet, youtube, else article) and status: queued. - Process inbox captures by keeping/moving, converting into a
type: note, or deleting. - Do not auto-write durable thinking notes from captured content unless explicitly asked.
Bases workflow
- -
00 Inbox/inbox.base is the canonical queue view. - Grouping is by
note.type for scan-by-content-type triage. - Prefer creating/editing
.base files over Dataview.
Titles as APIs
A note's title is its interface. Use complete, declarative phrases: "Spaced Repetition Works Because of Retrieval," not "Spaced Repetition." A well-titled note can be linked and understood without opening it. When creating or renaming notes, always prefer a full declarative statement.
Tag conventions
Tags live in the frontmatter tags array, not inline in the body.
- - Lowercase, hyphenated:
spaced-repetition, not Spaced Repetition or spacedRepetition. - Topic-oriented, not structural:
learning (what it's about), not important (how you feel about it). - Nest only when a hierarchy is genuinely useful:
ai/transformers is fine, deep nesting is not. - Suggest tags based on the vault's existing taxonomy rather than inventing new ones.
Editing rules
- - Preserve existing prose unless user asks to rewrite.
- Do not add or maintain an
updated frontmatter field. - Use dense wikilinking (
[[Note Title]]) and allow unresolved links as stubs. - Do not create top-level folders unless explicitly requested.
- Do not assign/change
status, move notes, or delete notes without explicit instruction. - Agent write surface is limited to:
-
/ask response callouts
- optional daily briefing callout in journals
Search patterns
Default QMD collections:
- - INLINECODE51
- INLINECODE52
- INLINECODE53
- INLINECODE54
CODEBLOCK1
OpenClaw integration
If configuring OpenClaw memory paths, use:
Do not write legacy top-level memorySearch.
Only change OpenClaw config when the operator explicitly wants this vault wired into an OpenClaw installation.
Zettelclaw
遵循规范的 Zettelclaw 知识库模型:从外部捕获内容,手动编写持久笔记,并使用智能体进行导航/综合。
这是一项纯指令技能。它本身不会安装任何软件。当环境中已存在 qmd 时,其为可选项,而 rg 是受支持的搜索回退方案。
知识库结构
/
├── 00 收件箱/
├── 01 笔记/
├── 02 日志/
├── 03 模板/
├── 04 附件/
└── README.md
笔记类型
每篇笔记均使用 YAML 前置元数据:
- type
- tags
- created
- status(queued | in-progress | done | archived)
- author
- source
使用以下主要类型:
- - note:持久的原子化思考笔记;无 status
- doc:非原子化的工作/参考笔记;使用 status
- journal:每日日志笔记;无 status
- 内容类型:article、book、movie、tv、youtube、tweet、podcast、paper(以及可扩展的其他内容类型);使用 status
模板
在创建笔记前,始终先读取 03 Templates/ 中对应的模板:
- - note.md
- journal.md
- clipper-capture.json
使用核心模板/日期语法。无需依赖 Templater。
收件箱工作流
- - 网页捕获通过 clipper-capture.json 落入 00 Inbox/。
- 剪藏工具根据 URL 设置 type(tweet、youtube,否则为 article)和 status: queued。
- 处理收件箱捕获内容:保留/移动、转换为 type: note,或删除。
- 除非明确要求,否则不要根据捕获内容自动编写持久思考笔记。
基础工作流
- - 00 Inbox/inbox.base 是规范的队列视图。
- 按 note.type 分组,以便按内容类型进行分类扫描。
- 优先创建/编辑 .base 文件,而非使用 Dataview。
标题即 API
笔记的标题就是它的接口。使用完整、陈述性的短语:Spaced Repetition Works Because of Retrieval,而非 Spaced Repetition。一个标题得当的笔记,无需打开即可被链接和理解。在创建或重命名笔记时,始终优先使用完整的陈述性语句。
标签规范
标签位于前置元数据的 tags 数组中,而非正文内联。
- - 小写,连字符连接:spaced-repetition,而非 Spaced Repetition 或 spacedRepetition。
- 面向主题,而非结构:learning(关于什么),而非 important(你的感受)。
- 仅在层级关系确实有用时使用嵌套:ai/transformers 可以,深度嵌套则不可取。
- 建议基于知识库现有分类体系来使用标签,而非发明新标签。
编辑规则
- - 除非用户要求重写,否则保留现有行文。
- 不要添加或维护 updated 前置元数据字段。
- 使用密集的维基链接([[Note Title]]),并允许未解析的链接作为占位符。
- 除非明确要求,否则不要创建顶级文件夹。
- 未经明确指示,不要分配/更改 status、移动笔记或删除笔记。
- 智能体的写入范围仅限于:
- /ask 响应标注
- 日志中可选的每日简报标注
搜索模式
默认 QMD 集合:
- - zettelclaw-inbox
- zettelclaw-notes
- zettelclaw-journal
- zettelclaw-attachments
bash
qmd(已安装时优先使用)
qmd query spaced repetition and retrieval -c zettelclaw-notes
qmd search status: queued -c zettelclaw-inbox
qmd vsearch what have I been learning about memory -c zettelclaw-notes
ripgrep 回退方案
rg -l type: note 01 Notes/
rg -l type: article 00 Inbox/ 01 Notes/
rg -l status: queued 00 Inbox/ 01 Notes/
OpenClaw 集成
如果配置 OpenClaw 记忆路径,请使用:
- - agents.defaults.memorySearch.extraPaths
不要写入旧的顶级 memorySearch。
仅当操作员明确希望将此知识库接入 OpenClaw 安装时,才更改 OpenClaw 配置。