Token导航 LogoToken导航

Agent Skill · 研究检索

handoff-memory(切换记忆)

handoff-memory 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。

handoff-memory 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 54,511。

总安装 511周安装 17GitHub Stars 2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:handoff-memory(切换记忆)
来源仓库:https://github.com/17-sss/agent-skills
仓库路径:skills/handoff-memory
安装命令:
npx skills add https://github.com/17-sss/agent-skills --skill handoff-memory
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/17-sss/agent-skills --skill handoff-memory

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,适合在 Codex、Claude、Cursor、Gemini CLI 中快速定位候选结果。适用于需要根据关键词或任务场景进行信息检索和筛选的场景。通过 npx skills add 命令从指定 GitHub 仓库安装使用。建议确认权限范围和维护状态,注意是否涉及联网或文件操作。

handoff-memory 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/handoff-memory

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/17-sss/agent-skills --skill handoff-memory;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

handoff-memory 怎么安装?

在终端执行 npx skills add https://github.com/17-sss/agent-skills --skill handoff-memory。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

handoff-memory 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。

handoff-memory 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。

handoff-memory 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。

handoff-memory 用的人多吗?

公开数据是累计安装 511 次,近一周安装 17 次,下载 140 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。

handoff-memory 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
handoff-memory

切换记忆 · 本页

5112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

Handoff Memory

Overview

Keep shared memory documents close to the work they describe. For a single repository, keep the canonical handoff in the repo. For a multi-repository workspace, keep workspace-wide memory in the workspace root and workstream-specific memory under dedicated workstream folders.

This skill keeps one canonical handoff per scope and adds lightweight operational tooling:

  • resolve the right document path
  • resolve the right resume target without guessing a workstream
  • create or refresh the canonical file
  • sync a lightweight _memory/INDEX.json for workspace workstreams
  • validate that the document is structurally usable
  • check whether the document is stale before trusting it
  • optionally write timestamped session snapshots without making them the primary state

Compatibility and Prerequisites

  • python3 must be available in PATH to run the bundled scripts.
  • git must be available in PATH for freshness checks and repo metadata.
  • The skill still resolves paths and validates documents outside Git, but stale checks are less informative there.
  • All bundled scripts are non-interactive and expose --help. Prefer running them directly instead of recreating their logic by hand.
  • Branch names may still be recorded in repo-local metadata, but resume selection should not use branch as the primary signal.

Default Workflow

Use this as the default flow unless the user already has a stronger convention:

  1. Resolve or initialize the canonical document. Run scripts/create_handoff.py --project-root <path> --scope auto --document handoff --format json. Add --workstream <name> when the task belongs to one specific repo combination inside a larger workspace.
  2. Read the canonical document before changing it. Preserve still-valid context. Remove stale claims that would mislead the next session.
  3. Update the canonical document with the current state. Keep it short, explicit, and action-oriented.
  4. Validate it before ending the session. Run scripts/validate_handoff.py --project-root <path> --scope auto --document handoff. Use --strict only when strict template conformance should fail the check. The default mode only requires the handoff to be resume-usable.
  5. When resuming, check staleness before trusting old notes. Run scripts/resolve_handoff_path.py --project-root <path> --scope auto --document handoff --resume --format json first, then scripts/check_staleness.py --project-root <path> --scope auto --document handoff.

In mixed multi-repo workspaces, do not treat the parent folder as automatically workspace-wide. Resume selection should prefer an explicit handoff or workstream, then repo overlap, then _memory/INDEX.json, then a unique current workstream. If more than one workstream still matches, return an ambiguous result instead of guessing. Use --workspace-wide only when the user explicitly wants status for every child repository.

For resume-style requests such as "continue", "resume", or "pick up where we left off", treat Next Actions as the default execution queue and Resume Prompt as the default execution framing. Do not reopen design direction that the handoff already narrowed unless the user asks to rethink it or the current repo state clearly invalidates it.

Use --snapshot with create_handoff.py only when the user wants a timestamped checkpoint that is worth preserving. Always pair it with --snapshot-kind and a short --snapshot-reason. Do not make snapshots the default shared state.

For day-to-day agent behavior, follow references/agent-usage-best-practices.md.

Path Rules

  • Prefer an explicit --handoff-path when the project already defines a canonical location.
  • If --handoff-path points inside _memory/workstreams/<name>/..., the scripts infer workspace scope and workstream name automatically.
  • Explicit --handoff-path or --workstream should override auto-detection.
  • In repo scope, reuse an existing shared handoff file in a recognized location such as docs/HANDOFF.md, memories/HANDOFF.md, or HANDOFF.md.
  • In workspace scope, keep cross-repo memory under _memory/.
  • If no repo handoff exists, create docs/HANDOFF.md.
  • If no workspace handoff exists, create _memory/HANDOFF.md.
  • For workspace-wide memory, use these defaults:

- _memory/HANDOFF.md for workspace-wide summary and active workstream index - _memory/WORKSPACE.md for durable workspace structure - _memory/DECISIONS.md for cross-repo architecture or policy choices - _memory/PATTERNS.md for repeatable conventions

  • For workstream-specific memory, use these defaults:

- _memory/workstreams/<name>/HANDOFF.md - _memory/workstreams/<name>/WORKSTREAM.md - _memory/workstreams/<name>/DECISIONS.md - _memory/workstreams/<name>/PATTERNS.md

  • Most sessions should only update the canonical handoff for the active scope. Touch companion workspace or workstream documents only when durable shared context changed.
  • In workspace scope, maintain a lightweight _memory/INDEX.json with last_active_workstream plus per-workstream canonical_path, last_updated, status, and repositories.
  • Optional snapshots live under docs/handoffs/, _memory/handoffs/, or _memory/workstreams/<name>/handoffs/.
  • Canonical files keep canonical names such as HANDOFF.md and WORKSTREAM.md. Only snapshots get timestamped names, using YYYYMMDD_HHMMSS[-n]-<kind>-<label>.md.
  • Avoid .codex, .claude, .windsurf, or .agents as the default shared mutable handoff location.

Scope Guidance

Repo Scope

Use repo scope when the task belongs to one repository.

  • Canonical file: docs/HANDOFF.md
  • Session snapshots: docs/handoffs/*.md
  • Keep implementation-level detail here

Workspace Scope

Use workspace scope when the agent session starts from a parent folder that coordinates multiple repositories and the task spans more than one of them at the workspace-wide level.

  • Canonical file: _memory/HANDOFF.md
  • Session snapshots: _memory/handoffs/*.md
  • Keep workspace-wide coordination and durable shared context here
  • Leave repo-specific implementation detail in each repo's own docs/HANDOFF.md

Workstream Scope

Use workstream scope when the same workspace hosts multiple independent repo combinations and one handoff per workspace would blur them together.

  • Canonical file: _memory/workstreams/<name>/HANDOFF.md
  • Durable overview: _memory/workstreams/<name>/WORKSTREAM.md
  • Session snapshots: _memory/workstreams/<name>/handoffs/*.md
  • Keep only the repos relevant to that initiative in the workstream handoff
  • Prefer workstream names that describe the initiative, not just the repo list

Content Rules

  • Keep the document scan-friendly. The next session should understand it in under a minute.
  • Start with a strong TL;DR.
  • Prefer exact workspace-relative paths, commands, dates, and repo names over vague prose.
  • Avoid machine-specific absolute paths such as /Users/... or other paths outside the current project root unless they are truly required. The validator warns on foreign absolute paths for portability.
  • Record what is true now, not a transcript of the full chat.
  • Note what changed recently, what is risky, and what should happen next.
  • Include a quick reference section with the few files, commands, dashboards, or docs that matter most.
  • Keep a short resume checklist so the next session can verify the notes before acting on them.
  • Note when a claim is unverified.
  • Do not store secrets, tokens, private keys, raw credentials, or long confidential logs.

Document Model

The canonical handoff is the shared source of truth. Optional session snapshots are secondary artifacts for traceability.

  • Canonical handoff:

- repo: docs/HANDOFF.md - workspace-wide: _memory/HANDOFF.md - workstream: _memory/workstreams/<name>/HANDOFF.md

  • Snapshot archive:

- repo: docs/handoffs/*.md - workspace-wide: _memory/handoffs/*.md - workstream: _memory/workstreams/<name>/handoffs/*.md

  • Workspace index:

- _memory/INDEX.json with lightweight workstream metadata for resume-time targeting

This model keeps the current state easy to find while still allowing point-in-time captures when they are useful.

Resume Workflow

When asked to continue work from a prior session:

  1. Resolve the resume target, not just the generic canonical handoff path. Run scripts/resolve_handoff_path.py --project-root <path> --scope auto --document handoff --resume --format json. The selection priority is: explicit --handoff-path, explicit --workstream, current repo overlap, index last_active_workstream, unique current or in-progress workstream, then ambiguous instead of guessed restore.
  2. Read the canonical handoff before planning or editing code.
  3. Treat the first unfinished item in Next Actions as the default plan, and treat Resume Prompt as the default continuation frame.
  4. Run the staleness check if the document might be old.
  5. If working in workspace scope, read companion memory files only when they are directly relevant.
  6. If the task is only one initiative inside a larger workspace, resolve the workstream document instead of the workspace-wide handoff.
  7. In a parent folder with unrelated repositories, narrow validation to the selected workstream or the repositories named in the selected handoff before considering a workspace-wide scan.
  8. Compare the document against the current repo, workstream, or workspace state and call out drift.
  9. Execute the first unfinished next action. Explore only as needed to complete it, not to replace it with fresh design work.
  10. Refresh and validate the canonical handoff again before ending the session if anything material changed.

Scripts

scripts/resolve_handoff_path.py

Resolve the canonical repo-local, workspace-wide, or workstream-specific memory path. Use --scope repo|workspace|auto, --document handoff|workspace|workstream|decisions|patterns, --workstream, or --handoff-path to honor an explicit override. Use --resume to apply resume target selection and return ambiguous instead of guessing when multiple workstreams still match. Use --ensure to create the file when it does not exist.

scripts/create_handoff.py

Initialize or refresh the canonical file and sync metadata such as project root and update timestamp. In workspace scope, this also refreshes _memory/INDEX.json. Use --snapshot --snapshot-kind <kind> --snapshot-reason <text> to write a timestamped archive copy before the canonical file is refreshed.

scripts/validate_handoff.py

Check that the document is resume-usable by default, warn on portability issues such as foreign absolute paths, and stay short enough to be useful. Use --strict when strict template conformance should fail on missing sections, placeholders, or empty required sections.

scripts/check_staleness.py

Compare the handoff timestamp against recent repo activity. The script prefers Last Updated metadata over file mtime, uses the selected workstream's declared repos when available, and does not fall back to scanning every child repo unless --workspace-wide is explicit. If resume targeting is ambiguous, it returns an ambiguous result instead of guessing.

References

references/handoff-template.md

Use this to understand the expected handoff structure and section intent.

references/workspace-memory-guide.md

Use this when the task spans multiple repositories and you need to decide whether the canonical workspace handoff is enough or whether the task needs a dedicated workstream.

references/agent-usage-best-practices.md

Use this when you want the recommended start-of-session, end-of-session, and scope-selection behavior for agents using this skill in normal work.

references/snapshot-strategy.md

Use this when you need explicit guidance on when to create a snapshot, when not to, and which snapshot kind to use.

references/agent-integrations.md

Use this when the user asks where to install the skill for Codex or another agent. Keep install notes separate from the handoff workflow itself.

Notes

  • Installation scope and data location are separate concerns.
  • The skill may be installed globally or per-project, but the shared mutable documents should stay inside the repository or workspace root they describe.
  • Agent-specific files such as AGENTS.md, CLAUDE.md, .codex/*, or .windsurf/rules/* may reference the shared handoff, but should not become the primary mutable store.

平台分布

Codex

36.87%
按下载量换算52

Claude

32.35%
按下载量换算45

Cursor

17.33%
按下载量换算24

Gemini CLI

9.48%
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