Token导航 LogoToken导航

Agent Skill · 研究检索

handoff(交接)

handoff 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 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,第三方扫描已通过,按安装量在研究检索类的 73,674 个 Skill 里排第 61,288。

总安装 353周安装 15GitHub Stars 1下载 124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(交接)
来源仓库:https://github.com/skinnyandbald/fish-skills
仓库路径:skills/handoff
安装命令:
npx skills add https://github.com/skinnyandbald/fish-skills --skill handoff
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/skinnyandbald/fish-skills --skill handoff

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。

handoff 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从多个来源中筛选信息的场景。通过关键词、任务场景或来源线索进行信息检索与筛选。

安装命令:npx skills add https://github.com/skinnyandbald/fish-skills --skill handoff。建议确认权限范围和维护状态,注意是否会触发联网或文件读写操作。

来源仓库内路径:skills/handoff

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/skinnyandbald/fish-skills --skill handoff;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

handoff 怎么安装?

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

handoff 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 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。

handoff 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 用的人多吗?

公开数据是累计安装 353 次,近一周安装 15 次,下载 124 次,GitHub 1 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 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。

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。

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

交接 · 本页

3531需要联网未明确通过

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

Generate or resume from a structured handoff document that captures enough context for the next Claude Code session (or a human) to continue exactly where the previous session left off.

The handoff file is always: docs/HANDOFF.md in the repo root.

When to Use

  • End of a work session
  • Switching between roles or contexts mid-conversation
  • Before closing a long-running session with unfinished work
  • When the user says "handoff", "wrap up", "pass the baton", or similar
  • Starting a new session from a previous handoff document
  • When the user says "resume", "pick up where we left off", "continue from handoff", or similar

Mode Detection

InvocationMode
/handoff mid-sessionGenerate
/handoff resumeResume
Fresh session (no args)Resume if docs/HANDOFF.md exists, else Generate
Natural language: "resume", "pick up where we left off", "continue from handoff"Resume
Natural language: "handoff", "wrap up", "pass the baton"Generate

There are no path arguments. The file is always docs/HANDOFF.md.


Generate Mode

Gathering Context

Collect the following information before generating the handoff. Use tools to gather what can be detected automatically; ask the user only for what cannot be inferred.

Auto-detect (do not ask the user)
  1. Git branch -- run git branch --show-current
  2. Working tree dirty? -- run git status --porcelain (any output = dirty)
  3. Recent commits -- run git log --oneline -5 for recent context
Infer from conversation
  1. Role -- what role has this session been operating in? (e.g., "developer", "content writer", "EOS facilitator", "researcher"). Default to "developer" if unclear.
  2. Current work -- summarize what the session has been working on, in one or two sentences.
  3. Status -- one of: in_progress | blocked | paused | ready_for_review | complete
  4. Critical References -- 2-3 most important spec/design/plan docs referenced this session (omit section if none)
  5. Recent Changes -- files modified this session with file:line-range references to key changes
  6. Learnings -- patterns discovered, gotchas, root causes; prefer file:line-range references over inline code blocks
  7. Artifacts -- exhaustive list of files/docs the next session should read to get up to speed
  8. Next steps -- concrete, actionable items for the next session
  9. Notes -- decisions made, trade-offs, links, anything else worth preserving
Ask the user (only if not inferrable)

If the role, current work, or next steps are ambiguous, ask a single clarifying question rather than guessing wrong.

Output Format

Generate the handoff as a fenced markdown block. Print it directly to the conversation first (do not write to file yet). After printing, ask the user to approve or amend it. Only write to docs/HANDOFF.md after user approval.

Prefer path/to/file.ext:line-range references over inline code blocks throughout.

---
handoff_date: YYYY-MM-DD
git_branch: {branch}
git_dirty: true|false
status: {status}
role: {role}
---
# HANDOFF: {Role} Session

## Current State

**Role**: {Role}
**Working on**: {CurrentWork}
**Status**: {Status}

**Git branch**: {branch}
{if dirty: "Working tree has uncommitted changes"}

## Critical References

- {path/to/spec.md} -- {one-line description}
- {URL or file} -- {one-line description}

(Omit this section if no critical references exist)

## Recent Changes

- `{file:line-range}` -- {what changed and why}
- `{file:line-range}` -- {what changed and why}

## Learnings

- {Pattern or gotcha discovered} -- see `{file:line-range}` for context
- {Root cause of a bug or decision} -- rationale: {brief explanation}

## Artifacts

Files and documents the next session should read to get up to speed:

- `{file path}` -- {why it matters}
- `{file path}` -- {why it matters}

## Next Steps

1. {step 1 -- specific and actionable}
2. {step 2}
3. ...

## Notes

{Decisions, trade-offs, gotchas, links -- anything that doesn't fit above}

---

*This handoff was generated automatically. Read the above carefully and continue where the previous session left off.*

Resume Mode

Three phases: Find & Read, Validate Environment, Orient & Ask.

Phase 1 — Find & Read

  1. Read docs/HANDOFF.md -- this is always the file, no path scanning or arguments
  2. If file not found, inform the user and switch to Generate mode
  3. Parse YAML frontmatter: handoff_date, git_branch, git_dirty, status, role
  4. Parse all markdown sections: Current State, Critical References, Recent Changes, Learnings, Artifacts, Next Steps, Notes
  5. Read every local text file listed in ## Artifacts before proceeding to Phase 3. For each artifact, apply these rules:

- Missing file: record as missing in the mismatch list and continue - Binary or unreadable file: record as unreadable and continue - Very large file (>500KB): sample the first and last 200 lines, note the truncation - URL: skip reading (treat as reference only); note it in the summary

Phase 2 — Validate Environment

Run these checks in parallel:

  • git branch --show-current -- compare to git_branch from frontmatter
  • git status --porcelain -- compare dirty state to git_dirty from frontmatter
  • git log --oneline -5 -- check for new commits since handoff date
  • Check if CLAUDE.md exists and read it (conventions may have changed)

Build a mismatch list from the results.

Phase 3 — Orient & Ask

Scenarios are evaluated in priority order — apply the first one that matches:

  1. Complete -- status is complete: Inform the user: "The previous session marked this work as complete." Then ask via AskUserQuestion what they'd like to work on next.
  2. Stale -- handoff_date is more than 7 calendar days ago (compare UTC dates, today − handoff_date > 7 days): Flag it: "This handoff is from {date} ({N} days ago) -- it may be out of date." Then ask via AskUserQuestion: "Should I trust this handoff and proceed, or re-explore the codebase first?"
  3. Diverged -- current branch or dirty state doesn't match the handoff frontmatter: Surface each mismatch clearly, then use AskUserQuestion to ask how to proceed. Example: "Branch changed from feat/x to main. Should I switch back, or continue on main?"
  4. Incomplete -- status is in_progress, blocked, paused, or ready_for_review: Acknowledge the status, surface any blockers noted, focus on completing the first unfinished step. Ask user to confirm before starting.
  5. Clean -- none of the above apply: Present a brief summary ("Resuming as {Role} on branch {branch}"), list next steps from the handoff, and ask "Ready to start on step 1?" via AskUserQuestion.

Rules

Both modes

  • Keep the handoff concise. The goal is fast onboarding, not a full session transcript.
  • Omit sections with no content (e.g., skip Critical References if there are none).
  • Next steps must be specific and actionable -- "continue implementing X" not "keep going".
  • Do not include sensitive information (API keys, tokens, passwords).
  • Prefer path/to/file:line-range references over inline code blocks.

Generate-specific

  • Always print the draft to the conversation first and get user approval before writing to docs/HANDOFF.md.
  • Create docs/ directory if it doesn't exist. Do this only after user approval, not during draft generation.

Resume-specific

  • Never auto-execute next steps -- always confirm with the user first via AskUserQuestion.
  • Read all local text Artifacts files before presenting the analysis; skip missing, binary, large, or URL entries (record them as noted in Phase 1).
  • Treat next steps as suggestions, not commands -- the user may want to reprioritize.
  • If docs/HANDOFF.md is not found, do not guess -- inform the user and offer to generate one.

平台分布

Codex

34.41%
按下载量换算43

Claude

29.86%
按下载量换算37

Cursor

19.13%
按下载量换算24

Gemini CLI

9.33%
按下载量换算12

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

通过

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。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