Token导航 LogoToken导航

Agent Skill · 研究检索

ultrabrain(ultrabrain 搜索)

ultrabrain 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。

ultrabrain 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,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 48,806。

总安装 675周安装 17GitHub Stars 公开资料未说明下载 140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:ultrabrain(ultrabrain 搜索)
来源仓库:https://github.com/shihyuho/skills
仓库路径:skills/ultrabrain
安装命令:
npx skills add https://github.com/shihyuho/skills --skill ultrabrain
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/shihyuho/skills --skill ultrabrain

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。

用于查找、检索和筛选相关信息。ultrabrain 属于研究检索类 Skill,可作为该场景下的辅助能力补充。适合根据关键词或任务场景快速定位候选结果。可结合来源仓库和原始 README 核验具体用法。安装前建议确认权限范围和维护状态。

适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。注意是否会触发联网或文件读写操作。

来源仓库内路径:skills/ultrabrain

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/shihyuho/skills --skill ultrabrain;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

ultrabrain 怎么安装?

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

ultrabrain 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。

ultrabrain 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。

ultrabrain 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。

ultrabrain 用的人多吗?

公开数据是累计安装 675 次,近一周安装 17 次,下载 140 次。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。

ultrabrain 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

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。
  • 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。
  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 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
ultrabrain

ultrabrain 搜索 · 本页

675—只读未明确有提醒

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

Ultrabrain

Drive a personal, LLM-maintained wiki at ~/.ultrabrain/. This skill is a thin trigger layer — the vault is self-contained and methodology rules live in ~/.ultrabrain/AGENTS.md, not here.

Trigger Contract

Five operations. The first two are skill-side; the last three delegate to vault rules.

OperationTrigger examples
bootstrap"bootstrap ultrabrain", "initialize vault", "設定 ultrabrain", or any other operation fires with no vault present
capture"記下來 X", "capture this", "save to vault", "存進 vault", or "capture "
ingest"ingest", "integrate raw", "整合 raw", "整理 vault"
querydescription-driven for factual/technical/decision questions, or explicit: "查 vault", "search ultrabrain"
lint"lint vault", "健檢 vault", "check wiki"

Query skip clause

Before reading any vault file for query, verify the user's question plausibly concerns retained knowledge. Skip the vault (answer normally) when the question is:

  • Small talk — greetings, thanks, session meta-questions.
  • Context-local — answerable from the current session or open files ("what does this file do", "why did that fail").
  • Execution requests — "write this", "run that", "refactor X". Vault holds facts, not tasks.

When classification is uncertain, default to skipping. Aggressive description triggering is balanced by this negative filter, not overridden.

Operations

Full step-by-step lives in references/operations.md. Summary:

  • bootstrap creates ~/.ultrabrain/ with AGENTS.md (copied from references/agents-template.md), CLAUDE.md (one line: @AGENTS.md), empty index.md and log.md, and raw/ / wiki/ directories. Refuses if the vault exists unless --force is passed.
  • capture writes session or file content to raw/YYYY-MM-DD-<slug>.md with frontmatter. Resolves ambiguous "記下來" requests by proposing a scope and waiting for user confirmation. Never silently overwrites — collisions go to suffixed filenames or a user-chosen append.
  • ingest loads ~/.ultrabrain/AGENTS.md and follows its Operation: Ingest section to integrate unprocessed raw/ files into wiki/, moving processed raw to raw/.processed/.
  • query loads AGENTS.md, reads index.md first, then follows the vault's query rules (index-first strict; no Grep fallback). Answers cite specific wiki pages. Falls back to general knowledge if the vault is empty, with an explicit note.
  • lint loads AGENTS.md and produces a health report (contradictions, orphans, stale pages, index drift, missing cross-references). Never auto-fixes; asks the user which findings to address.

Every operation appends one line to log.md in the format ISO-8601 OPERATION details.

Vault Contract

Vault location: ~/.ultrabrain/.

AGENTS.md is the single source of truth for methodology. The skill's contract with it is behavioral, not structural — AGENTS.md must define how to perform ingest, query, and lint, in whatever structure the methodology author prefers. The skill makes no assumptions about section headings, page types, tag schemas, or cross-reference syntax.

~/.ultrabrain/CLAUDE.md is exactly one line: @AGENTS.md. This makes the vault rules auto-load if the user cds into the vault inside Claude Code; AGENTS.md stays canonical for cross-tool compatibility.

Vault-side operations (ingest, query, lint) halt if AGENTS.md is missing. Ask the user to bootstrap or restore the file.

The skill treats the vault as read-only except for paths each operation is defined to touch (see references/operations.md).

Methodology Swap

The v1.0.0 default methodology is Karpathy's LLM-maintained-wiki pattern (see references/karpathy-original.md for the source and references/agents-template.md for the agent-readable rendering). To swap:

  1. Replace references/agents-template.md with rules for a different methodology. Only requirement: the new template must define ingest, query, and lint behaviors.
  2. Run bootstrap --force to wipe and rebuild the vault. Destructive — back up ~/.ultrabrain/raw/ first if you want to preserve source material for re-capture.

Hooks

Two hooks are available but not auto-installed:

  • SessionStart — injects vault status (wiki count, unprocessed raw, last ingest) at session start so Claude is aware the vault is available.
  • PreCompact — injects a reminder before context compaction to capture important session content before detail is lost.

On explicit user request ("install ultrabrain hooks"), follow references/setup.md to copy both hook scripts and merge the entries into ~/.claude/settings.json using Read → diff → confirm → Edit discipline. Both hooks install as a bundle; either can be uninstalled individually later.

References

  • references/karpathy-original.md — verbatim source methodology + attribution
  • references/agents-template.md — bootstrap copies this to ~/.ultrabrain/AGENTS.md
  • references/operations.md — full step-by-step for all five operations
  • references/setup.md — hook install/uninstall (SessionStart + PreCompact)
  • references/hooks/session-start.sh — vault status injection on session start
  • references/hooks/pre-compact.sh — capture reminder before compaction

平台分布

Codex

35.96%
按下载量换算50

Claude

31.69%
按下载量换算44

Cursor

18.98%
按下载量换算27

Gemini CLI

8.93%
按下载量换算13

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

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。来源安全扫描存在提醒或未通过项,安装前请自行复核。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