Token导航 LogoToken导航

Agent Skill · 研究检索

scout(scout 搜索)

scout 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。

scout 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,权限为需要联网;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 66,724。

总安装 259周安装 11GitHub Stars 374下载 91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:scout(scout 搜索)
来源仓库:https://github.com/huytieu/cog-second-brain
仓库路径:skills/scout
安装命令:
npx skills add https://github.com/huytieu/cog-second-brain --skill scout
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/huytieu/cog-second-brain --skill scout

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。

scout 用于查找、检索和筛选相关信息。适合在 Codex、Claude、Cursor、Gemini CLI 中根据关键词快速定位候选结果。通过 npx skills add 命令从指定仓库安装并使用该技能。安装前需确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

建议结合来源仓库和原始 README 核验具体用法和功能细节。

来源仓库内路径:skills/scout

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/huytieu/cog-second-brain --skill scout;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

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

scout 怎么安装?

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

scout 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。

scout 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

scout 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。

scout 用的人多吗?

公开数据是累计安装 259 次,近一周安装 11 次,下载 91 次,GitHub 374 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。

scout 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。

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。

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
scout

scout 搜索 · 本页

259374需要联网未明确有提醒

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

COG Scout Skill

Purpose

Lightweight URL/tool triage that sits between "ignore" and /url-dump. Evaluates whether a URL or tool is worth saving or skipping — checking existing vault coverage, assessing relevance to the user's profile and interests, and recommending a clear next action.

When to Invoke

  • User wants to evaluate a URL or tool before committing to a full save
  • User says "scout this", "evaluate this", "should I save this?", "is this relevant?"
  • User shares one or more URLs and wants a quick relevance assessment
  • User mentions a tool/service name and wants to know if it's worth investigating

Agent Mode Awareness

Check agent_mode in 00-inbox/MY-PROFILE.md frontmatter:

  • If agent_mode: team — delegate vault scanning and web fetching to parallel sub-agents (one for vault search, one for content fetch/analysis). Combine results for recommendation.
  • If agent_mode: solo (default) — handle all scanning and analysis directly in the conversation. No delegation.

Pre-Flight Check

Before executing, check for user profile:

  1. Look for 00-inbox/MY-PROFILE.md and 00-inbox/MY-INTERESTS.md in the vault
  2. If NOT found: Welcome to COG! Scout works best with a profile for relevance matching. Would you like to run /onboarding first, or should I evaluate with general criteria?
  3. If found:

- Read MY-PROFILE.md for active projects and role - Read MY-INTERESTS.md for topic areas - Read 00-inbox/MY-INTEGRATIONS.md for active integrations (check if web-fetch and web-search are available)

Boundary with /url-dump

Scout evaluates ("should I save this?"). URL-dump saves ("save this now").

  • Scout checks existing coverage, assesses relevance, and recommends an action
  • If the recommendation is Save, scout hands off to /url-dump with pre-filled category
  • Users who already know they want to save should use /url-dump directly

Process Flow

1. Accept Input

Accept one or more of:

  • URL(s): Direct links to evaluate
  • Tool/service name(s): Will search for the tool first
  • Mixed: Combination of URLs and names

Prompt (if no input provided):

What URL(s) or tool(s) would you like me to evaluate?
(You can paste URLs, tool names, or a mix)

Batch mode: Multiple URLs/names in one invocation are processed together with a summary table at the end.

2. Vault Coverage Check

For each URL or tool name, search the entire vault for existing coverage.

Search strategy:

  • Extract domain from URL (e.g., github.com/owner/repo → search for repo name)
  • Search for tool/service name across the whole vault (grep for domain, repo name, tool name)
  • Match against URL strings in frontmatter (url: fields) and inline links

If found:

🔍 Existing coverage found for [name]:
- [file path] — saved [date], category: [category]
- [file path] — mentioned in [context]

Want me to check if an update is needed, or skip this one?

3. Content Fetch & Analysis

If URL provided and web-fetch is active:

  • Fetch the URL content using WebFetch
  • Extract: title, description, content type, author, date

If tool name provided (no URL):

  • Use WebSearch to find the tool's primary page
  • Fetch and analyze the top result

Content type detection:

  • Tool/Service: Software, SaaS, API, library, framework
  • Article/Blog: Long-form content, tutorial, opinion piece
  • Repository: GitHub/GitLab repo (extract stars, last commit, language)
  • Research: Paper, study, academic content
  • News: Industry news, announcement
  • Reference: Documentation, spec, standard

4. Relevance Assessment

Score relevance against user context:

Profile Match (from MY-PROFILE.md):

  • Does it relate to an active project? Which one?
  • Does it align with the user's role?
  • Does it fit the user's tech stack?

Interest Match (from MY-INTERESTS.md):

  • Does it match any declared interest topics?
  • How directly relevant is it?

Quality Signals:

  • For repos: stars, recent activity, maintainer health
  • For tools: pricing model, maturity, adoption
  • For articles: author credibility, publication quality, recency
  • For all: uniqueness vs. what's already in the vault

5. Recommendation

Based on analysis, recommend one of two actions:

Save — Worth adding to the knowledge base
✅ SAVE — [Title/Name]
Category: [suggested category for url-dump]
Relevance: [High/Medium] — [why it matters]
Projects: [affected project(s) if any]

Shall I hand off to /url-dump to save it?
Skip — Not relevant or not worth the time
⏭️ SKIP — [Title/Name]
Reason: [clear explanation — wrong stack, low quality, already covered, irrelevant to interests]

6. Batch Summary (for multiple items)

When processing multiple URLs/tools, end with a summary table:

## Scout Summary

| # | Item | Verdict | Reason |
|---|------|---------|--------|
| 1 | [Name 1] | ✅ Save | [brief reason] |
| 2 | [Name 2] | ⏭️ Skip | [brief reason] |

**Actions:**
- [X] items ready to save via /url-dump

7. Execute Follow-up Actions

Based on user confirmation:

  • Save items: Hand off to /url-dump with pre-filled category suggestion
  • Skip items: No action needed

Fallback Behavior

ScenarioBehavior
web-fetch unavailableEvaluate based on URL structure, domain reputation, and vault search only. Note that content wasn't fetched.
web-search unavailableFor tool-name inputs (no URL), ask the user for a direct URL instead. For URL inputs, proceed normally — web-search is not needed.
No user profileEvaluate with general quality/relevance criteria, skip personalized relevance scoring
URL is paywalledNote limitation, evaluate based on available preview and metadata
Tool not found via searchAsk user for more context or a direct URL

Uncertainty Handling

  • High confidence: Clear relevance match or clear irrelevance — give direct recommendation
  • Medium confidence: Partial match — present pros/cons, let user decide
  • Low confidence: Can't determine relevance — explain what's unclear, ask user for context

Integration with Other Skills

Downstream

  • Save → hands off to /url-dump with pre-filled category

Upstream

  • /daily-brief may surface new tools/services → user can run /scout to evaluate
  • /auto-research may discover tools during research → scout can triage them

Success Metrics

  • Quick triage (< 1 minute for single URL in solo mode)
  • Clear, actionable recommendations
  • Accurate vault coverage detection (no duplicate saves)
  • Relevance scoring matches user expectations
  • Smooth handoff to /url-dump when saving

Philosophy

Scout embodies COG's "evaluate before you accumulate" principle:

  • Not everything deserves a bookmark — be selective
  • Existing coverage should be surfaced before creating duplicates
  • Binary save/skip keeps decisions fast and avoids half-measures
  • Clear recommendations reduce decision fatigue

平台分布

Codex

34.09%
按下载量换算31

Claude

30.72%
按下载量换算28

Cursor

21.21%
按下载量换算19

Gemini CLI

10.15%
按下载量换算9

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

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