Token导航 LogoToken导航

Agent Skill · 研究检索

kb-query(知识库查询)

kb-query 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 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。

kb-query 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为只读;许可证信息还不明确,第三方扫描已通过,按安装量在研究检索类的 73,675 个 Skill 里排第 71,838。

总安装 192周安装 8GitHub Stars 2下载 64核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
unknown

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:kb-query(知识库查询)
来源仓库:https://github.com/bahayonghang/obsidian-notes-karpathy
仓库路径:skills/kb-query
安装命令:
npx skills add https://github.com/bahayonghang/obsidian-notes-karpathy --skill kb-query
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 npx skills 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。

skills.shnpx skills
npx skills add https://github.com/bahayonghang/obsidian-notes-karpathy --skill kb-query

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

kb-query 用于查找、检索和筛选相关信息,适合快速定位候选结果。适用于关键词搜索、任务场景匹配或来源线索筛选等研究检索场景。通过 npx skills add 命令从 GitHub 仓库安装并使用。安装前需确认权限范围、维护状态及是否触发联网或文件操作。建议结合原始 README 核验具体用法和功能边界。

来源仓库内路径:skills/kb-query

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/bahayonghang/obsidian-notes-karpathy --skill k…;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它标注为只读,先用只读任务试一次,确认行为符合预期再扩大使用范围。

常见问题

6 条

kb-query 怎么安装?

在终端执行 npx skills add https://github.com/bahayonghang/obsidian-notes-karpathy --skill kb-query。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

kb-query 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

kb-query 需要什么权限?

权限级别是只读。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。

kb-query 是免费的吗?

公开资料没有标注明确许可证,收费和商用限制都无法确认。正式使用前请到来源仓库核对授权说明。

kb-query 用的人多吗?

公开数据是累计安装 192 次,近一周安装 8 次,下载 64 次,GitHub 2 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

kb-query 安全吗?

第三方扫描结果为通过,但这只覆盖已知的静态检查项,不能替代自己读一遍源码和安装命令。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 第三方扫描已通过,常见的高风险行为没有命中。
  • 标注为只读权限,试用阶段不容易误改本地文件。

先想清楚的地方

  • 许可证没有明确标注,能不能商用、能不能再分发都不确定。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。
  • 只想先试试效果、不希望它改动本地文件的人。

不太适合

  • 需要明确授权才能用到公司项目里的场景。
  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
kb-query

知识库查询 · 本页

1922只读未明确通过

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

KB Query

Search, answer, and generate outputs from the approved live layer.

Minimal loop

  1. locate the best approved live pages and briefings
  2. answer or draft the artifact from wiki/live/
  3. archive the result when it is substantive
  4. record the smallest durable delta that would improve the wiki next

When this compounds the wiki

kb-query should not behave like disposable chat over live notes. Strong outputs should either become reusable archived artifacts or leave behind explicit candidates for a better concept page, synthesis page, relationship edge, governed question, or curated hub.

When not to promote

Do not treat a polished answer as approved truth by itself. If the result discovers durable knowledge, it must still re-enter through draft -> review -> live.

Read before querying

Read these files first:

  • local AGENTS.md
  • local CLAUDE.md if present
  • ../obsidian-notes-karpathy/scripts/skill-contract-registry.json
  • ../obsidian-notes-karpathy/references/file-model.md
  • ../obsidian-notes-karpathy/references/lifecycle-matrix.md
  • ../obsidian-notes-karpathy/references/briefing-template.md
  • ../obsidian-notes-karpathy/references/qa-template.md
  • ../obsidian-notes-karpathy/references/content-output-template.md
  • ../obsidian-notes-karpathy/references/activity-log-template.md
  • ../obsidian-notes-karpathy/references/search-upgrades.md
  • ../obsidian-notes-karpathy/references/questions-and-reflection-policy.md
  • ../obsidian-notes-karpathy/references/query-writeback-lifecycle.md
  • ../obsidian-notes-karpathy/references/memory-lifecycle.md
  • ../obsidian-notes-karpathy/references/graph-contract.md
  • ../obsidian-notes-karpathy/references/render-template.md
  • ../obsidian-notes-karpathy/references/profile-contract.md
  • ../obsidian-notes-karpathy/references/episode-template.md

Treat skill-contract-registry.json as the canonical source for required references, baseline script, and expected write surfaces.

If ../obsidian-notes-karpathy/scripts/scan_query_scope.py exists, run it first and treat its output as the deterministic baseline for live-layer retrieval boundaries.

Then start with:

  • wiki/index.md
  • wiki/live/indices/INDEX.md
  • wiki/live/indices/CONCEPTS.md
  • wiki/live/indices/SOURCES.md
  • wiki/live/indices/TOPICS.md when present
  • wiki/live/indices/QUESTIONS.md when present
  • relevant wiki/briefings/{role}.md when the request maps to a role
  • prior outputs/qa/

Hard boundary

kb-query must not treat raw/ or wiki/drafts/ as retrieval truth.

MEMORY.md is also outside the default knowledge retrieval boundary. Read it only when the user is asking about preferences, editorial priorities, or collaboration behavior rather than topic knowledge.

Those layers may be cited only as evidence if a human explicitly asks for source inspection. Default synthesis should stay on wiki/live/. outputs/episodes/ and outputs/health/graph-snapshot.json may help surface candidates, but they are not default truth surfaces.

Source-backed answer discipline

  • prefer approved summaries when grounding key conclusions
  • use concept and entity pages as navigation and synthesis aids, not as a shortcut around approved evidence trails
  • when sources disagree, say so explicitly instead of flattening the disagreement

Modes

  1. search mode for quickly locating approved pages and ranking local candidates before synthesis
  2. research mode for grounded answers archived into outputs/qa/
  3. web mode for static browseable web exports under outputs/web/
  4. reflect-lite mode for question resolution, synthesis notes, or gap reports that should stay outside live until re-reviewed

When a substantive answer or artifact creates durable follow-up work, archive explicit writeback_candidates, open_questions_touched, source_live_pages, writeback_status, and a followup_route so the next compile/review or maintenance pass can decide what should happen next.

Prefer the smallest durable delta that improves future reuse:

  • update an existing live concept or entity when the identity is already right
  • draft a missing concept, entity, or synthesis when the knowledge is durable but absent
  • strengthen related links or question tracking when the problem is graph quality rather than page absence
  • create or expand a curated hub when repeated outputs keep traversing the same approved cluster

Retrieval ladder

Default retrieval order:

  1. wiki/index.md
  2. wiki/live/indices/*
  3. governance indices such as QUESTIONS.md, GAPS.md, and ALIASES.md when present
  4. relevant wiki/briefings/{role}.md
  5. prior outputs/qa/
  6. local structured or metadata-driven search
  7. optional semantic retrieval only as candidate surfacing

Semantic retrieval may help discover candidate pages, but approved live pages remain the truth source.

If the user explicitly says kb-search, treat that as direct wording for kb-query search mode.

If the user wants slides, reports, charts, or canvas derivatives from approved knowledge, hand off to kb-render instead of stretching kb-query into deterministic rendering work.

Writeback contract

Compounding principle: every substantive query should leave the wiki at least slightly better — whether through a new draft page, a stronger relationship, an updated hub, or a promoted question. Writeback is the default posture, not an opt-in extra.

Use ../obsidian-notes-karpathy/references/query-writeback-lifecycle.md as the detailed contract.

At minimum, substantive Q&A or publish outputs should:

  • record source_live_pages when specific approved pages grounded the answer
  • record open_questions_touched when the answer materially advances or reframes a standing question
  • record writeback_candidates when the output discovers durable follow-up worth re-entering the wiki
  • record writeback_status so later passes can see whether the work is still pending
  • record compounding_value when future reuse, navigation value, or synthesis payoff should be explicit
  • record crystallized_from_episode when the output came out of a broader episodic chain
  • set followup_route to none, draft, or review

Use:

  • none when the output is grounded and does not create durable follow-up work
  • draft when the output suggests new or updated long-term knowledge that must re-enter draft -> review -> live
  • review when the next action is an immediate human decision or a governance pass on already-prepared maintenance work

Output to the user

Report:

  1. which live pages and briefings were used
  2. whether prior Q&A was reused
  3. where the new artifact was saved
  4. whether a query or publish log entry should be appended
  5. whether any follow-up should go back into drafts instead of directly mutating live

平台分布

Codex

31.83%
按下载量换算20

Claude

31.73%
按下载量换算20

Cursor

17.33%
按下载量换算11

Gemini CLI

9.68%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

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。目前只有一个公开来源,建议同时核对源仓库。

继续浏览同类 Skills