Token导航 LogoToken导航

Agent Skill · 研究检索

session-checkpoint(会话检查点)

session-checkpoint 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。

session-checkpoint 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 57,766。

总安装 424周安装 17GitHub Stars 2下载 137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:session-checkpoint(会话检查点)
来源仓库:https://github.com/guardzcom/skills
仓库路径:skills/session-checkpoint
安装命令:
npx skills add https://github.com/guardzcom/skills --skill session-checkpoint
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/guardzcom/skills --skill session-checkpoint

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。

session-checkpoint 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、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。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/session-checkpoint

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/guardzcom/skills --skill session-checkpoint;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

session-checkpoint 怎么安装?

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

session-checkpoint 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。

session-checkpoint 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。

session-checkpoint 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。

session-checkpoint 用的人多吗?

公开数据是累计安装 424 次,近一周安装 17 次,下载 137 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。

session-checkpoint 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
session-checkpoint

会话检查点 · 本页

4242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

Session Checkpoint

CRITICAL: This skill runs near context window limits. Minimize tool calls — no subagents, no file exploration. Use only conversation context. Write files in one shot.

Commands

  • Save: save checkpoint {name}, save checkpoint (auto-picks name from context)
  • Resume: continue, continue {name}, resume (auto-select if one checkpoint, list if many)
  • Cleanup: clear checkpoint {name}, clear all checkpoints

Save Procedure

1. Determine Name

If user gave a name, use it.

If no name given:

  • Pick a short descriptive name from the current work (2-4 words, e.g., "my-feature") and use it directly — don't ask for confirmation. The user sees the name in the output.

Sanitize final name: lowercase, replace spaces// with -, strip leading -.

2. Locate Memory Directory

Find the auto memory directory path from the system prompt line: You have a persistent auto memory directory at <path>

3. Gather State

Run in a single bash call:

git branch --show-current && git status --short | head -10 && date '+%Y-%m-%d %H:%M'

From conversation context, determine:

  • What was accomplished this session
  • What failed or was abandoned (and why)
  • Where work left off — don't infer next steps, but DO capture all actionable context the next session needs: pending decisions, proposed options, open questions
  • Which files were modified

4. Find Active Plan

If the system prompt or conversation already references a plan file (~/.claude/plans/*.md), note its path and determine step N of M. Otherwise skip. Do NOT search for or glob plan files.

5. Write Checkpoint File

If checkpoint-{name}.md already exists, use AskUserQuestion to confirm — show the existing checkpoint's Saved date and Left Off summary, with options: "Overwrite" / "Different name" (user provides new name via "Other"). If they choose a different name, continue from this step with the new name.

Write <memory_dir>/checkpoint-{name}.md:

# Checkpoint: {name}

- **Branch:** `{branch}`
- **Saved:** {YYYY-MM-DD HH:MM}
- **Plan:** `~/.claude/plans/{file}` (step N of M)

## Left Off
{1-3 lines describing where work stopped}

## Done This Session
- {accomplishments}

## Failed Approaches
- {what + why — or "None"}

## Modified Files
- {list}

Omit the Plan: line if no plan file exists.

6. Update MEMORY.md Index

Always use Edit (not Write) for MEMORY.md changes — parallel sessions would clobber each other with Write.

Read MEMORY.md. Find the ## Active Checkpoints section.

  • If checkpoint name already in index: update its line.
  • If not: add a new bullet.
  • If no ## Active Checkpoints section exists: insert it after the first # heading. If MEMORY.md doesn't exist, create it with # Project Memory as the heading.

Bullet format:

- **{name}** ({branch}, {Mon DD}) — {one-line summary}

Keep the ` Resume any: continue or continue {name} ` line after the bullets.

Do NOT touch any other sections in MEMORY.md.

7. Output

Keep output minimal — this often runs near end of context window.

Checkpoint "{name}" saved. Resume anytime: `continue {name}`

That's it. Don't ask follow-up questions. Don't suggest /compact or /clear — the user manages their own context window.


Resume Procedure

1. Determine Name

If user gave a name, use it. If no name given:

  • List checkpoint-*.md files in memory directory
  • If exactly one: use it automatically
  • If multiple: use AskUserQuestion with checkpoint names as options (include branch and date in each option's description)
  • If none: report "No saved checkpoints found." and stop

2. Read Checkpoint File

Read <memory_dir>/checkpoint-{name}.md. If file missing: remove the orphaned bullet from MEMORY.md's ## Active Checkpoints, report "Checkpoint file missing (cleaned up stale entry)", and stop.

3. Check Branch

git branch --show-current

If current branch differs from checkpoint's branch, warn:

Warning: You're on `{current}` but checkpoint was saved on `{saved}`. Switch with: git checkout {saved}

Continue regardless — user decides.

4. Load Plan

If checkpoint references a plan file, read it.

5. Present and Begin

Output the checkpoint summary and last context. If Failed Approaches is non-empty, include them:

⚠ Previously failed: {approach} — {why}

6. Auto-Clear

The checkpoint is now consumed — context is live. Delete the checkpoint file and remove its bullet from MEMORY.md's ## Active Checkpoints (same as Cleanup steps 1-2). If no bullets remain, remove the entire section.

Do NOT mention the cleanup to the user. They see only the summary from step 5, then wait for their direction.


Cleanup Procedure

1. Delete File

Remove the checkpoint file via Bash:

rm <memory_dir>/checkpoint-{name}.md

2. Update MEMORY.md

Remove the bullet for {name} from ## Active Checkpoints. If no bullets remain, remove the entire section (header + resume line).

3. Output

Cleared checkpoint "{name}"

clear all checkpoints

Glob checkpoint-*.md in memory directory. Delete all matches. Remove the entire ## Active Checkpoints section (header + bullets + resume line) from MEMORY.md. Output:

Cleared {N} checkpoint(s)

平台分布

Codex

36.06%
按下载量换算49

Claude

31.73%
按下载量换算43

Cursor

19.27%
按下载量换算26

Gemini CLI

9.73%
按下载量换算13

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