Token导航 LogoToken导航

Agent Skill · 研究检索

context-pack(上下文包)

context-pack 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。

context-pack 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,权限为调用外部服务;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 71,089。

总安装 196周安装 8GitHub Stars 9下载 63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:context-pack(上下文包)
来源仓库:https://github.com/epismoai/skills
仓库路径:skills/context-pack
安装命令:
npx skills add https://github.com/epismoai/skills --skill context-pack
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/epismoai/skills --skill context-pack

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。

context-pack 通过 Epismo 上下文包实现知识持久化,支持跨工具切换和团队协作交接。以主题为单位组织知识块,鼓励渐进式完善而非碎片化存储。统一 CLI 与 MCP 接口连接到同一 Epismo 服务,避免双通道冲突。安装前建议确认 API 密钥配置和网络访问权限。

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/context-pack

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/epismoai/skills --skill context-pack;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

    它需要调用外部服务,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

context-pack 怎么安装?

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

context-pack 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。

context-pack 需要什么权限?

权限级别是调用外部服务。该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

context-pack 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。

context-pack 用的人多吗?

公开数据是累计安装 196 次,近一周安装 8 次,下载 63 次,GitHub 9 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。

context-pack 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
context-pack

上下文包 · 本页

1969调用外部服务未明确有提醒

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

Context Pack

Store and retrieve context through Epismo context packs — so knowledge survives tool switches, session ends, and team handoffs.

The unit of organization is a block inside a pack. The goal is one well-structured pack per topic, grown over time by adding and refining blocks — not many small packs.

For Epismo connection setup, CLI/MCP surface conventions, scope model, share URL resolution, and error handling, see Epismo Basics.
Surface selection: CLI and MCP connect to the same Epismo service. Use CLI if available; fall back to MCP if not. Never use both in the same session.

Commands

CommandNatural language triggers→
packpack this, save this, summarize sessionPACK
newnew pack, start fresh, hand off a task, share project statusNEW
publish [<id>]publish this, make this public, publish a guide, share with communityPUBLISH
`get <id\alias>`get this ID, load this ID, restore from ID, read alias, open aliasGET
find <query>what context do I have, load my context, searchFIND
`update [<id\alias>]`edit this pack, update my last packUPDATE
organizereorganize blocks, split this, clean upORGANIZE
—unclear intentAsk once

Operations (context packs)

OperationCLIMCP
search packepismo pack search --type context [--query <keywords>] [--filter '{...}']epismo_pack_search
get packepismo pack get --id <id> [--full] [--block-id <id>]epismo pack get --alias <alias>epismo_pack_get
create packepismo pack create --input '<json>'epismo_pack_create
update packepismo pack update --id <id> --input '<json>'epismo_pack_update
delete packepismo pack delete --id <id>epismo_pack_delete
like packepismo pack like --id <id> --likedepismo_pack_like
upsert aliasepismo alias upsert --type context --id <id> --alias <name>—
get aliasepismo alias get --alias <name>—
list aliasesepismo alias list --type context—
delete aliasepismo alias delete --alias <name>—

PACK

Goal: save current context as a block inside the right pack. Find an existing pack and add or refine a block; create a new pack only when nothing relevant exists.

Step 1 — Find the right pack

Infer 2–4 topic keywords from the current session. Search private packs (titles only — one fast call).

search pack — type: context, query: <inferred topic>, filter: {visibility: ["private"]}

ResultAction
One clear matchFetch it → add or update a block → upsert
Multiple candidatesShow titles and IDs → ask user to pick
No matchProceed as NEW

Step 2 — Identify the right block

get pack — fetch the pack and read its existing blocks. Decide:

  • Update an existing block — if the new content belongs to a block that already exists (e.g., decisions, next steps), integrate it there. Rewrite the block so it reads as a coherent whole, not as an append.
  • Add a new block — if the new content is a distinct topic not covered by any existing block.

The pack should read well from scratch at any point. Do not leave stale or contradicted information in place.

Step 3 — Update and return

update pack:

{
  "id": "<existing-id>",
  "content": "<top-level summary or intro — does not include block content>",
  "blocks": [
    { "op": "update", "id": "<existing-block-id>", "title": "<block-title>", "content": "<updated block content>" },
    { "op": "add", "title": "<new-block-title>", "content": "<new block content>" }
  ]
}
  • Omitted fields are left unchanged. Omitting "blocks" means no block changes.
  • "op": "update" updates an existing block by ID. "op": "add" adds a new block. "op": "remove" deletes a block by ID (only "id" needed).
  • "content" at the top level is a brief intro or summary. Block content lives in "blocks[]", not in the top-level "content".
context  <context-id>  <context-title>
block    <block-id>    <block-title>

NEW

Goal: create a fresh pack with an initial block structure. Use when no relevant pack exists, or when the content is clearly a separate topic.

Step 1 — Design the block structure

Before writing, decide how the content divides into blocks. Each block should be focused enough to update independently and broad enough to stay relevant over time.

Design for selective fetch. Because blocks can be loaded individually with --block-id, a well-split pack lets future sessions load only what they need without pulling the whole pack. Aim for blocks where any one block answers a distinct question on its own.

Use caseStarting blocks
Session / tool handoffContext, Decisions, Next Steps
Source summary (YouTube, blog, paper)Source, Summary, Key Takeaways, Reuse Notes
Project memory / working contextPurpose, Constraints, Decisions, References
Prompt or instruction packWhen To Use This, Inputs, Instructions, Output Expectations
Project statusSummary, Progress, Blockers
Team rules or planPurpose, Rules, Plan
Task handoffCurrent State, What Remains, Context for Recipient

Good block boundaries: topic changes, audience changes, or update frequency changes (e.g., "Decisions" is stable; "Next Steps" changes every session — keep them separate).

For templates, see Content Templates.

Step 2 — Write

Content rules:

  • Each block is self-contained — no "see above" references
  • Prefer scannable structure: bullets, short paragraphs, and small tables only when they help
  • Include IDs, URLs, and precise names so the reader can act without searching
  • Title: infer from session topic. Do not ask unless genuinely ambiguous.

Step 3 — Publish

Default private. Use PUBLISH to go public.

create pack:

{
  "title": "<title>",
  "content": "<top-level summary or intro>",
  "type": "context",
  "visibility": "private",
  "blocks": [
    { "title": "<block-title>", "content": "<block content>" },
    { "title": "<block-title>", "content": "<block content>" }
  ]
}
  • Blocks are passed as a separate "blocks[]" array — not embedded in the top-level "content" string.
  • "content" at the top level is a brief intro only. All substantive content belongs in blocks.
Who needs itvisibilityAccess target
Just me"private"omit targets
My team"private"targets.projectIds=["pj_xxx"]
context  <context-id>  <context-title>
block    <block-id>    <block-title>

PUBLISH

Goal: make a pack publicly discoverable. Always requires explicit user confirmation before writing.

Two paths:

SituationPath
Existing private pack worth sharingPromote — change visibility to public
New content written for a broad audienceCreate new public pack

Promote: private → public

  1. get pack — fetch the pack. Run the Public Review Gate. Flag issues before proceeding.
  2. Confirm with user: "Publish '{title}' ({id}) as public under category {category}?"
  3. update pack:
{
  "id": "<id>",
  "visibility": "public",
  "category": "<category>"
}

Create new public

Write for an external reader using the public guide template in Content Templates. Content must be self-contained and pass the Public Review Gate.

Choose category carefully — it determines how people find this pack. See Category Reference.

Confirm with user, then create pack:

{
  "title": "<title>",
  "content": "<content>",
  "type": "context",
  "visibility": "public",
  "category": "<category>"
}

Return ID, title, and share URL. For share URL resolution, see Epismo Basics — Resolving Share URLs.

context  <context-id>  <context-title>
share    https://epismo.ai/hub/contexts/<id>

GET

Goal: fetch a pack by ID, alias, or explicit read/open target.

Route intent before resolving

Prefer alias-first unless the input is obviously a search query.

Input patternRouteWhy
get..., use..., load..., open..., read..., restore...GETExplicit retrieval
find..., search..., what context do I have, show my packsFINDExplicit discovery
Bare UUIDGETIDs are unambiguous
Short ambiguous phraseGETTry alias first, then search if it misses
Obvious search query, question, or long descriptive phraseFINDDiscovery intent is clearer than alias intent

Resolve the input

  1. UUID → get pack --id.
  2. @handle/name or one compact token like weekly-handoff → get pack --alias; if it misses, run FIND with the same text.
  3. Short phrase like auth refactor handoff → try get pack --alias first; if it misses, run FIND.
  4. Question, explicit search wording, or long descriptive text → FIND first.
  5. get/read/open/use <target> always counts as retrieval intent.

get pack — by id, or by alias. Default returns outline only; pass --full for all blocks, or --block-id to load a single block.

Read the most relevant block first. Continue from there without re-reading history.


FIND

Goal: discover the right pack when the ID is not known.

search pack — scan titles only (no full content). Use this for natural-language discovery requests and multi-keyword topic phrases. Search in this order:

  1. type: context, query: <topic>, filter: {visibility: ["private"]}
  2. type: context, query: <topic>, filter: {like: "liked"}
  3. type: context, query: <topic>, filter: {visibility: ["public"]}

Present the title list; if the match is clear, get pack immediately.

For filter keys and search recipes, see Search & Discovery.


UPDATE

Goal: edit an existing pack directly.

ID or alias known

Always fetch before writing.

get pack → inspect current blocks → update pack with id and block operations using op fields.

For update <alias> based on this conversation:

  • merge new context into the right existing blocks
  • rewrite blocks coherently instead of appending raw notes
  • add a new block only for a distinct new topic

ID unknown

  1. search pack — type: context, query: <inferred topic>, filter: {visibility: ["private"]}
  2. One clear match → fetch and update without asking.
  3. Multiple candidates → show titles and IDs, ask the user to pick.
  4. No match → fall back to NEW.

"Last pack" shortcut: search private context packs and take the first result from the default recent-first outline.


ORGANIZE

Goal: improve the block structure of an existing pack.

The most common need is not managing many packs — it's making the blocks inside one pack cleaner and more useful.

Actions

SituationAction
A block has grown too large or covers multiple topicsSplit — divide into two focused blocks, rewrite each
Two blocks overlap or one has become redundantMerge — combine into one coherent block, remove the other
A block is no longer relevantRemove — delete the block from the pack
The whole pack is no longer neededDelete — delete the pack (needs approval)

All changes are applied via update pack using op fields. Use "op": "add" for new blocks, "op": "update" for existing ones, "op": "remove" to delete a block by ID. delete pack requires explicit user approval.


Write Safety

  1. Default private — pack, new, and update always write private. Use PUBLISH to go public.
  2. No silent writes — if scope is unclear, ask once before writing.
  3. Approval required — public publication, deletion, and overwriting another owner's pack require explicit user confirmation before writing.

For the full approval matrix, see Visibility & Sharing — Approval Boundary.


Operation Output

After every operation, return in this order:

  1. Context and block — always return both ID and title at each level that was affected
  2. What changed — one sentence
  3. Next action — single most useful next step (omit if none)
context  <context-id>  <context-title>
block    <block-id>    <block-title>    # when a specific block was created or updated

Skip analysis and evidence unless the user asked for it.

Source of Truth

Repository: https://github.com/epismoai/skills

平台分布

Codex

33.01%
按下载量换算21

Claude

31.47%
按下载量换算20

Cursor

18.85%
按下载量换算12

Gemini CLI

9.17%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

调用外部服务

该 Skill 可能调用第三方服务、云服务或外部模型 API,使用前需要确认账号、额度、数据发送范围和服务条款。

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