Token导航 LogoToken导航TokenDH.com
研究检索只读clawhub未标认证来源可访问clear审计通过

obsidian-karpathy-wikiObsidian karpathy wiki 搜索

Agent Skill

obsidian-karpathy-wiki 用于查找、检索和筛选相关信息,适合在 OpenClaw 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

4,692

周安装

188

GitHub Stars

1

下载量

1,519
OpenClaw

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

MIT-0

最后核验

2026-05-01

来源状态

来源可访问

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:obsidian-karpathy-wiki(Obsidian karpathy wiki 搜索)
来源仓库:https://github.com/ducheneadas-source/obsidian-karpathy-wiki
安装命令:
openclaw skills install obsidian-karpathy-wiki
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。该命令会通过 OpenClaw 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

ClawHubOpenClaw
openclaw skills install obsidian-karpathy-wiki

简介

将 Obsidian 知识库作为持久 LLM wiki 进行操作。

  • 适合原始笔记摄取与结构化知识沉淀。
  • 使用 openclaw skills install obsidian-karpathy-wiki 安装。
  • 需确认是否依赖外部向量数据库服务。
  • 建议先小规模试用再扩展应用场景。obsidian-karpathy-wiki 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

name
obsidian-karpathy-wiki
description
Operate an Obsidian knowledge base as a persistent LLM wiki using a raw-to-source-to-wiki pattern. Use when Codex or another agent needs to ingest raw notes, reconcile loose inbox material, maintain source summaries, compile formal wiki pages, write back valuable query outputs, run heartbeat maintenance, lint contradictions and stale claims, or hand off ongoing wiki operations to another agent.

Obsidian Karpathy Wiki

Overview

Use this skill to maintain an Obsidian vault as a persistent, compounding knowledge base instead of a passive note archive.

The core pattern is:

  • raw material enters the vault
  • the agent summarizes it into sources
  • stable conclusions are promoted into formal wiki pages
  • valuable query results are saved into outputs
  • outputs can later be written back into formal knowledge

This skill is designed for an agent that does recurring maintenance, not just one-off cleanup.

Read references/object-model.md before restructuring a vault. Read references/workflow.md before ingest, query, lint, or migration work. Read references/adaptation-checklist.md before adapting this skill to a specific user's vault. Read references/heartbeat.md before setting up recurring maintenance behavior.

Core Idea

Do not treat the goal as organizing files.

Treat the goal as compiling knowledge once and then keeping it current.

A good vault maintained with this skill should let the agent answer questions by reading a living wiki, not by rediscovering everything directly from raw documents every time.

Recommended Vault Shape

Prefer this logical structure, even if physical names differ:

  • raw/
  • compile/ or another staging layer
  • wiki/
  • an output layer such as outputs/ or content-outputs/
  • governance files such as CLAUDE.md and a maintenance skill

Within wiki/, prefer this formal structure:

  • index.md
  • log.md
  • questions/
  • methods/
  • decisions/
  • principles/
  • domains/
  • concepts/
  • entities/
  • sources/
  • outputs/

Questions, methods, and decisions are the main navigation axis. Principles explain why methods and decisions stay valid across situations. Sources are evidence bridges. Outputs are reusable query artifacts.

Operating Model

Raw

Raw material is source-of-truth input.

Do not treat raw notes as formal knowledge by default.

Loose notes in the vault root should usually be treated as temporary inbox material and moved into a raw inbox folder unless the user explicitly wants them to remain in place.

Sources

wiki/sources/ is the bridge between raw material and formal knowledge.

A source page should summarize:

  • what the material says
  • which questions it informs
  • which methods it supports
  • which decisions it changes
  • which principles it strengthens or challenges

Formal Pages

Promote stable knowledge into:

  • questions/
  • methods/
  • decisions/
  • principles/

Avoid creating many thin formal pages from a single source.

Outputs

Save high-value query results into wiki/outputs/.

Examples:

  • syntheses
  • comparison tables
  • structured analysis
  • decision memos
  • topic briefs

Outputs are not automatically formal knowledge, but they are often new source material for future write-back.

Core Operating Loop

1. Ingest

When new material appears:

  1. Read the source.
  2. If a loose note appears in the vault root, treat it as inbox material.
  3. Move loose inbox notes to a human raw inbox folder unless the user explicitly wants them to remain in place.
  4. Create or update a wiki/sources/ page.
  5. Extract only the stable upgrades:

- question - method - decision - principle - concept - entity

  1. Update only the few formal pages that truly benefit.
  2. Update wiki/index.md.
  3. Append a record to wiki/log.md.

Do not explode one source into many thin pages.

2. Query

When answering a user question:

  1. Read wiki/index.md first.
  2. Read the most relevant question, method, and decision pages.
  3. Read principle, concept, entity, and source pages as supporting context.
  4. Produce a structured answer.
  5. If the answer is reusable, save it under wiki/outputs/.
  6. Decide whether the output should later be written back into formal pages.

3. Weekly Compile

Use weekly compile to:

  • review the last 5-7 days of source and staging material
  • extract repeated questions
  • promote stable methods
  • update decisions if assumptions changed
  • promote principles when repeated patterns explain why methods remain valid
  • update domain maps, index.md, and log.md

4. Lint

Check for:

  • questions with no linked methods
  • methods with no scenario or caveat
  • decisions with stale assumptions
  • principles that duplicate lower-value pages
  • source pages not connected to formal knowledge
  • pages that contradict each other
  • stale claims superseded by newer sources
  • loose notes that never entered the ingest flow
  • high-value outputs that should have been written back

5. Backfill

Treat output folders as write-back sources.

Backfill only:

  • reusable methods
  • recurring questions
  • strategic decisions
  • principle-level lessons

Do not promote finished drafts directly into formal wiki pages.

Heartbeat Rule

Heartbeat should mean recurring maintenance, not empty status pings.

A good heartbeat checks:

  • whether new loose notes appeared
  • whether raw inbox has unread material
  • whether source pages are piling up without promotion
  • whether formal pages are missing principle-level explanation
  • whether the vault has new orphan pages, stale assumptions, or contradictions
  • whether query outputs are accumulating without write-back

Heartbeat should lead to either:

  • direct maintenance work
  • or a concrete maintenance findings page

Publishing Safety

When publishing or sharing this skill:

  • do not hardcode a specific local path
  • do not include private automation ids
  • do not include personal account names or bot nicknames
  • do not assume the output folder must use one particular language name
  • keep user-specific rules in a separate private operator layer if needed

Writing Rules

  • Use Markdown only.
  • Use YAML frontmatter on formal pages.
  • Use Obsidian wiki links.
  • Keep source material distinct from formal conclusions.
  • Prefer durable formal page titles over date-based titles.
  • Treat loose root notes as temporary inbox material.

Handoff Rule

This skill is suitable as a public foundation for persistent Obsidian wiki maintenance.

For a private deployment, pair it with a separate vault-specific governance file that defines:

  • the actual vault path
  • exact folder names
  • any automations or heartbeat schedule
  • user-specific output folders
  • domain-specific operating rules

适合场景

01

OpenClaw 用户查找和安装 Skill 时

02

用户想查找某类 Agent Skill 时

03

需要根据任务场景推荐可安装能力包时

04

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

补充不同宿主或平台的使用分布数据

能力 5

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

OpenClaw

72.14%
按下载量换算1,096

安全审计

VirusTotal

通过

ClawScan

通过

Static analysis

通过

权限和风险

只读

该 Skill 主要提供规则、说明或参考内容,本身偏只读;真正读写文件、联网或执行命令仍取决于宿主 Agent 的任务。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills