relay-chat-and-work — 接力干活 + 自动存档对话

Relay your long tasks AND auto-archive your chats. Split work into stages, progress step-by-step, persist to disk, resume across sessions — all while silently backing up each finished conversation (and its files) to your private GitHub repo.

🇬🇧 English

relay-chat-and-work is a single skill for AI agents (Claude Code, Minis, Codex, Gemini CLI — anything that supports Agent Skills) that does two things:

1. Relay long tasks — split work into numbered stages, persist progress to disk, and resume across sessions on any device.
2. Auto-archive chats — when a new conversation opens, silently back up the previous session (full transcript + generated files) to your private GitHub repo, organized by conversation title.

It is client-agnostic: everything is plain Markdown, so you can start a task in one agent and continue it in another without losing context. Comes with a cross-agent adapter layer (scripts/adapters/) and an ISSUES.md project log for tracking open tasks/decisions across sessions.

Install:

# Copy the whole relay-chat-and-work/ folder into your skills directory.
# e.g. Claude Code: ~/.claude/skills/  ·  Codex: ~/.agents/skills/

Do I need a GitHub token?

• No, not always. The core relay-work feature (staging, progress on disk, cross-session resume), ISSUES tracking, and local archiving all work purely on local files — no token required.
• Only the "auto-archive to your private GitHub repo" step needs a token, because it git pushes to GitHub. Set the GITHUB_TOKEN env var to enable that. Without it, the skill still works for everything local, but the remote backup step will error out.

Local ≠ relay. Relay work (picking up a half-finished task) already runs 100% locally — the GitHub archive is only an optional cross-device / cross-client backup of your full conversation history. If you never switch devices or clients, local-only is fine; GitHub adds the power to restore everything from anywhere.

Create a least-privilege GITHUB_TOKEN:

Prefer a fine-grained personal access token — it scopes access to only the repos you pick.

1. GitHub → Settings → Developer settings → Personal access tokens → Fine-grained tokens → Generate new token.
2. Token name: anything, e.g. relay-chat-and-work.
3. Expiration: 90 days or shorter (rotating is safer).
4. Repository access: Only select repositories, then check:
   • your archive target https://github.com/<you>/<you>-chat-and-work
   • and the skill repo itself if your tool needs to push it.
5. Under Permissions, set Contents → Read and write (the only permission needed to push). Leave everything else at its default.
6. Generate token and copy it immediately (shown only once).

If your tool only supports classic tokens: use Tokens (classic) → Generate and check repo (full read/write, needed to push public + private). It's broader than fine-grained, so prefer fine-grained when possible.

Finally, set it as the GITHUB_TOKEN environment variable:

GITHUB_TOKEN=your_token

Quick usage:

• Relay work — tell your agent: "接力干活" / "split this into stages" — it stages, persists progress, and resumes later.
• Auto-archive — nothing to say; it silently archives the previous session on the next new chat. Archiving is local by default; pass --push to archive_session.py to git commit + push to GitHub.

中文

一个给 AI 助手(Claude / Minis / Codex 等支持 skills 的 Agent)用的一体化长任务技能。它把两件事合二为一:

1. 接力干活(Rely Work)—— 长任务拆阶段、落盘、跨会话续接
2. 对话自动存档(Auto-Archive)—— 每次新对话打开,自动把上一轮会话(对话全文 + 产物文件)静默备份到你的私人 GitHub 仓库

它解决什么问题

长任务一次做不完,会话一断上下文就丢;而且时间一长,有价值的对话散落各处、找不到也续不上。

relay-chat-and-work 用进度落盘让长活能接力;用自动存档让每次对话都能永久、分类地保存到你的私人仓库,按对话标题归档,随时可回溯、可续接。

核心机制

私人存档仓库

存储目标: <你的GitHub用户名>-chat-and-work(private)

例如账号是 minioutman,则私人库为 minioutman/minioutman-chat-and-work。每个对话 → 仓库/<对话标题>/:

<对话标题>/
├── conversation.md   ← 完整对话
├── meta.json         ← 会话元信息
└── files/            ← 该会话产生的产物文件

安装

把 relay-chat-and-work/ 整个目录放进你的 skills 目录。对 Minis 用户:设置 → Skills → 导入。

需要 GitHub token 吗?

不一定。 分两种情况:

• 接力干活(拆阶段、进度落盘、跨会话续接)、ISSUES 记录、本地存档 —— 全部只依赖本地文件,不需要 token。
• 只有**「自动存档到私人 GitHub 仓库」**这一步需要 token,因为它要 git push 到 GitHub。设置环境变量 GITHUB_TOKEN 并在调用 archive_session.py 时加 --push 即可开启;不加 --push 只做本地存档,不会报错。

所以:只想接力干活→不用 token;想自动备份对话到私人库→需要 token。

本地≠接力。 接力干活(接着上次没做完的活)本来就在本地运行,和 GitHub 无关;GitHub 存档只是对你完整对话历史的可选跨设备/跨客户端备份。如果你一直只在同一台设备、同一个客户端用,纯本地就够了;接 GitHub 的价值在于——无论换到哪台设备、哪个客户端,一条 git clone 就能把所有对话全找回来。

如何创建最小权限的 GITHUB_TOKEN

直接创建 fine-grained (细粒度) Personal Access Token 最安全——它只授权"那几个指定仓库"。

1. 打开 GitHub → 头像 → Settings → Developer settings → Personal access tokens → Fine-grained tokens,点 Generate new token。
2. Token name:随便填,如 relay-chat-and-work。
3. Expiration:建议选 90 days 或更短(过期后需重新生成,更安全)。
4. Repository access:选 Only select repositories(指定仓库),勾选:
   • 自动存档目标私人库:<你的用户名>-chat-and-work
   • 如果你的工具也要 push 技能公开库,一并勾上 relay-chat-and-work
5. Permissions 里,展开 Contents,改成 Read and write(这是 push 文件所需的唯一权限)。其余全部保持默认(只读/不授权)。
6. 点 Generate token,马上复制并保存这个 token(只会显示一次)。

如果你用的工具只支持老式 token,就建 Classic token: 同上入口选 Tokens (classic) → Generate,勾选 repo(完整仓库读写,含 push 公开+私人库),别的都别勾。注意 classic 的 repo 权限范围偏大,不如 fine-grained 精确。

最后把 token 设置成环境变量 GITHUB_TOKEN 即可,例如在配置里加:

GITHUB_TOKEN=你的token

一个可选的裁剪技巧:如果只想 push 公开库、绝不碰私人库,可用只含 public_repo 权限的 classic token;但请记得这个 skill 的自动存档目标是私人库,所以默认至少要能写那一个私人仓库。

用法

接力干活 —— 对 AI 说:「接力干活」「拆成步骤」「先做xx再做xx」… 自动存档 —— 无需说话,新对话打开自动静默执行;默认写本地,--push 才推送到私人库。

目录结构

relay-chat-and-work/
├── SKILL.md                   # 主文件:触发 + 规则
├── scripts/
│   └── archive_session.py     # 会话存档工具脚本
└── references/
    └── protocol.md            # 进度文件规范

License

MIT