Token导航 LogoToken导航

Agent Skill · 研究检索

snapshot-writer(快照写入器)

snapshot-writer 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。

snapshot-writer 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,权限为权限未标注;许可证为 MIT,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 46,860。

总安装 759周安装 31GitHub Stars 422下载 243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
MIT

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:snapshot-writer(快照写入器)
来源仓库:https://github.com/willoscar/research-units-pipeline-skills
仓库路径:skills/snapshot-writer
安装命令:
npx skills add https://github.com/willoscar/research-units-pipeline-skills --skill snapshot-writer
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。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/willoscar/research-units-pipeline-skills --skill snapshot-writer

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。

snapshot-writer 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。安装前建议确认权限范围和维护状态,避免触发联网或文件读写操作。

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。

来源仓库内路径:skills/snapshot-writer

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/willoscar/research-units-pipeline-skills --ski…;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

    权限没有明确标注,第一次使用时只给一个无关紧要的任务,观察它到底读了哪些文件和网络。

常见问题

6 条

snapshot-writer 怎么安装?

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

snapshot-writer 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。

snapshot-writer 需要什么权限?

权限级别没有明确标注。第一次使用时建议只给无关紧要的任务,观察它实际读写了哪些内容。

snapshot-writer 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

snapshot-writer 用的人多吗?

公开数据是累计安装 759 次,近一周安装 31 次,下载 243 次,GitHub 422 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。

snapshot-writer 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。
  • 3 个目录同时收录,说明这个 Skill 流传比较广。
  • 许可证明确为 MIT,商用和分发边界好判断。
  • 第三方扫描已通过,常见的高风险行为没有命中。

先想清楚的地方

  • 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。
  • 需要许可证清楚、并且有第三方扫描结果的团队项目。

不太适合

  • 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
snapshot-writer

快照写入器 · 本页

759422权限未标注MIT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

Snapshot Writer (1-page, bullet-first)

Goal: produce a compact, reader-facing snapshot that answers:

  • what is the topic boundary?
  • what are the key themes?
  • what should a reader read first?

This is intentionally not a full survey: prefer tight bullets + concrete pointers over narrative.

Role cards (use explicitly)

Snapshot Editor (scout)

Mission: deliver a one-page, high-signal snapshot that a reader can act on immediately.

Do:

  • Keep every bullet content-bearing: claim -> why it matters -> pointer(s).
  • Prefer contrasts and evaluation anchors over topic lists.
  • Treat paper pointers as the product (auditable, minimal).

Avoid:

  • Outline narration ("This snapshot/section...") and slide navigation ("Next, we...").
  • Generic survey boilerplate and disclaimer spam.
  • Turning the snapshot into a mini-survey with long paragraphs.

Pointer Curator (bibliography hygiene)

Mission: ensure every pointer is concrete and traceable to papers/core_set.csv.

Do:

  • Use a stable pointer format: P#### - Title (arXiv:... / doi:... / url:...).
  • Mix canonical anchors + recent strong baselines + benchmark/protocol papers.

Avoid:

  • Dumping every paper; the snapshot is a reading path, not a catalog.

Role prompt: Snapshot Author (bullet-first; paper-like)

You are writing a one-page literature snapshot.

Your job is to be useful fast:
- define the topic boundary
- surface the key themes as claims (not headings)
- give an actionable reading path (paper pointers)

Style:
- bullets-first, compact, calm
- no narration ("In this snapshot...") and no slide navigation ("Next, we...")

Constraints:
- do not invent papers
- pointers must come from papers/core_set.csv (or the same workspace candidate pool)
- if evidence is abstract-only, state it once as a single bullet, then move on

Inputs

Required:

  • outline/outline.yml
  • papers/core_set.csv

Optional (if available):

  • queries.md (time window / exclusions context)
  • papers/papers_dedup.jsonl (if core_set is very small)

Outputs

  • output/SNAPSHOT.md

Writing contract (paper-like, not generator-like)

  • Keep it to about 1 page (roughly <= 700-900 words).
  • Bullets-first: use short paragraphs only when unavoidable (<= 3 lines each).
  • No outline narration: avoid This section/subsection..., In this snapshot..., Next, we....
  • Don’t spam disclaimers: if evidence is abstract-only, say it once in a short “Evidence policy” line.
  • Every claim bullet should attach at least 1 concrete pointer (paper_id + title; include arxiv_id/doi/url when present).

Recommended structure (stable, minimal headings)

  1. Title + scope (2-3 bullets)
  2. Evidence policy (1 bullet)
  3. Taxonomy (4-6 bullets; groupings only)
  4. Key themes (6-10 bullets; each bullet = 1 claim + 1-2 pointers)
  5. What to read first (6-12 bullets; canonical + recent; each bullet has pointers)
  6. Open problems / risks (4-8 bullets)

Workflow

  1. Read outline/outline.yml and extract:

- the intended chapter structure (H2) - the 6-10 most “write-worthy” bullets per chapter

  1. Read papers/core_set.csv and build a small “pointer palette”

- Prefer: canonical anchors + recent strong baselines + evaluation/benchmark papers. - Avoid: dumping every paper; pick “must-read” sets. - If papers/core_set.csv is very small, also scan papers/papers_dedup.jsonl and cherry-pick a few missing anchors (keep pointers auditable).

  1. Write output/SNAPSHOT.md

- Start each section with a content claim (why it matters), not a navigation sentence. - Make at least 2 cross-paper contrasts (A vs B) to avoid a flat list. - Use consistent pointer formatting, e.g.: - P0012 - <Title> (arXiv:xxxx.xxxxx) or P0012 - <Title> (doi:...)

Definition of Done

  • output/SNAPSHOT.md exists and reads like a human-written snapshot (no template narration).
  • Includes >= 15 distinct paper pointers (or all papers if core_set < 15).
  • Includes >= 2 explicit contrasts and >= 1 evaluation/benchmark bullet (if present in core set).

Troubleshooting

Issue: snapshot feels empty / generic

Fix:

  • Increase papers/core_set.csv size (rerun retrieval/dedupe with broader queries.md).
  • Tighten the outline: fewer headings, stronger H2 names, and bullets that encode “what to compare”.

Issue: snapshot reads like an outline narrator

Fix:

  • Delete all “This section...” openers and replace with: Claim -> why it matters -> pointers.

平台分布

Claude Code

31.33%
按下载量换算76

Gemini CLI

24.91%
按下载量换算61

Cursor

16.36%
按下载量换算40

Codex

12.07%
按下载量换算29

OpenCode

8.24%
按下载量换算20

Antigravity

3.78%
按下载量换算9

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

权限和风险

权限未标注

当前来源未能明确判断权限范围,默认进入异常复核队列。

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。

继续浏览同类 Skills