lark-codex-bridge 0.0.1

package/CHANGELOG.md

@@ -0,0 +1,9 @@
+# Changelog
+
+## 0.0.1 (2026-05-21)
+
+Initial public package setup for `lark-codex-bridge`.
+
+- Forked from `feishu-claude-code-bridge`.
+- Replaced the local agent backend with Codex CLI.
+- Added Feishu/Lark bridge runtime, Codex session resume, interactive cards, and runtime configuration.

package/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 Lark Channel Bridge contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -0,0 +1,228 @@
+# lark-codex-bridge
+
+A lightweight bot that bridges Feishu / Lark messenger with your local Codex CLI. Run one command, scan a QR code to bind a Lark app, and talk to Codex from chat — read screenshots, edit code, anything you'd do at the terminal.
+
+> Note: this project is forked from [feishu-claude-code-bridge](https://github.com/zarazhangrui/feishu-claude-code-bridge). It keeps the original Feishu/Lark bridge, session management, interactive cards, and setup wizard, while replacing the local agent backend from Claude Code with Codex CLI and renaming the package to `lark-codex-bridge`.
+
+[中文 README](./README.zh.md)
+
+## What it does
+
+- Forwards Feishu / Lark messages (DM directly, or `@bot` in a group) to your local `codex` CLI, running in a working directory you control.
+- **Streaming card**: Codex's text and tool calls update on a single Lark card in real time — no waiting for the final reply.
+- **Per-chat sessions**: each chat keeps its own Codex thread, so conversations resume where they left off.
+- **Preempt + batch**: a new message interrupts the running run; rapid-fire messages get coalesced into one request.
+- **Multiple workspaces**: `/ws` switches between named project directories, with sessions tracked per workspace.
+- **Images and files**: send them to the bot directly — Codex reads the locally downloaded paths.
+- **Interactive cards**: `/help`, `/ws list`, `/status` return cards with buttons you can click.
+
+## Prerequisites
+
+- Node.js **>= 20**
+- `codex` CLI installed and logged in — see https://developers.openai.com/codex
+- A Lark / Feishu **PersonalAgent** app (the QR-code wizard on first launch can create one for you).
+
+## Install
+
+```bash
+npm i -g lark-codex-bridge
+# or
+pnpm add -g lark-codex-bridge
+```
+
+## First run
+
+```bash
+lark-codex-bridge start
+```
+
+The first run detects there's no app configured and **opens a QR-code wizard**:
+
+1. A QR code renders in your terminal.
+2. Scan it with the Feishu / Lark app.
+3. Pick or create a PersonalAgent app.
+4. Credentials are written to `~/.lark-codex/config.json`.
+
+### Granting scopes and event subscriptions
+
+The wizard creates the app shell, but you still need to confirm a few things on the Lark Developer Console:
+
+**Permission scopes:**
+- `im:message`
+- `im:message:send_as_bot`
+- `im:resource`
+
+**Event subscriptions (over long-lived WebSocket):**
+- `im.message.receive_v1`
+- `card.action.trigger`
+- `im.message.reaction.created_v1` / `deleted_v1` (optional)
+- `im.chat.member.bot.added_v1` (optional)
+
+After enabling those, run `lark-codex-bridge start` again. Once you see `✓ Connected`, find the bot in Feishu / Lark and start chatting.
+
+## Commands
+
+### Host CLI
+
+```
+lark-codex-bridge start [-c <config>]   Start the bot
+lark-codex-bridge ps                    List all running start processes on this machine
+lark-codex-bridge stop <id|#>           Stop a start process (SIGTERM, SIGKILL after 2s)
+lark-codex-bridge --help                List all commands
+```
+
+> When the same app is started multiple times, Lark's open platform routes events to one of the live WebSocket connections at random. `start` detects existing processes for the same app and (in a TTY) prompts: `[c]ontinue / [k]ill old / [a]bort`. In non-TTY mode it warns and continues.
+
+`status` / `doctor` / `handover` / `workspace` / `service` are placeholders, planned for later releases.
+
+### Slash commands inside Feishu / Lark
+
+| Command | Effect |
+|---|---|
+| `/new`, `/reset` | Clear the current chat's session |
+| `/cd <path>` | Switch working directory (resets session) |
+| `/ws list` | List named workspaces (card + buttons) |
+| `/ws save <name>` | Save current cwd as a named workspace |
+| `/ws use <name>` | Switch to a named workspace |
+| `/ws remove <name>` | Delete a named workspace |
+| `/status` | Current cwd / session / agent (card + buttons) |
+| `/config` | Adjust preferences (reply style, tool-call display, ...) |
+| `/stop` | Stop the run in progress (also the `⏹` button on the card) |
+| `/timeout [N\|off\|default]` | Idle-watchdog (minutes) for the current session. `/config` sets the global default. See FAQ below. |
+| `/ps` | List all `start` processes on this host, marking the one replying |
+| `/exit <id\|#>` | Stop a `start` process (your own → graceful; another's → SIGTERM) |
+| `/reconnect` | Force a WebSocket reconnect (use when the bot stops responding after a network blip) |
+| `/doctor [description]` | Feed recent logs and your description back to Codex for self-diagnosis |
+| `/help` | Help card |
+| Any other `/xxx` | Forwarded verbatim to Codex |
+
+**Reply policy**: in a DM, the bot replies to anything. In a **group (including topic groups), the bot only replies when `@`-mentioned** (default since 0.1.22); unmentioned messages are ignored. `@all` is never answered. Cloud-doc comments must mention the bot. To restore the older "always answer in groups" behaviour: `/config` → "Require @bot in groups" → No.
+
+## Data directories
+
+| Path | Content |
+|---|---|
+| `~/.lark-codex/config.json` | App credentials (App ID / Secret), mode 600 |
+| `~/.lark-codex/sessions.json` | Codex thread id + cwd per chat / topic (+ optional `/timeout` override) |
+| `~/.lark-codex/workspaces.json` | Named-workspace map |
+| `~/.lark-codex/processes.json` | Process registry for live `start` instances (used by `ps`/`stop`); dead PIDs are auto-pruned |
+| `~/.lark-codex/media/<chatId>/` | Downloaded images / files, cleaned up after 24h |
+| `~/.lark-codex/logs/YYYY-MM-DD.log` | Structured run logs (JSONL), rotated daily; older than 7 days are pruned at startup (`LARK_CODEX_LOG_DAYS` env var overrides). `/doctor` reads these. |
+
+> Upgrading from before 0.1.11? Run `lark-codex-bridge migrate` once — it moves anything under `~/.config/lark-codex-bridge/` and `~/.cache/lark-codex-bridge/` to the new location and upgrades `config.json` to the new schema.
+
+## Codex Runtime Config
+
+Send `/config` in Feishu / Lark to tune the Codex subprocess. Changes apply to the next run:
+
+- `codexBinary`: Codex CLI command or absolute path; default `codex`
+- `model`: passed to `codex exec --model`; leave blank to use Codex defaults
+- `profile` / `profileV2`: passed as `--profile` and `--profile-v2`
+- `sandbox`: `workspace-write` (default), `read-only`, or `danger-full-access`
+- `skipGitRepoCheck`: enabled by default so `/cd` can point at non-Git folders
+- `search`: enables Web Search by launching `codex --search exec ...`
+
+You can also edit `~/.lark-codex/config.json` under `preferences.agent`:
+
+```json
+{
+  "preferences": {
+    "agent": {
+      "provider": "codex",
+      "codexBinary": "codex",
+      "sandbox": "workspace-write",
+      "skipGitRepoCheck": true,
+      "search": false
+    }
+  }
+}
+```
+
+## Access control (optional)
+
+Out of the box the bot is **open**: anyone who can find it can DM it, any group member can `@`-mention it to trigger a run, and commands like `/account` or `/cd` are usable by all. **That's fine for personal use** — but for a shared team setup, or anywhere you don't want strangers calling `/cd /`, you can tighten three allowlists by sending `/config` inside Feishu.
+
+### Common scenarios
+
+**Just me**
+
+In the `/config` form:
+- **Allowed users**: your own `open_id`
+- Leave the other two blank
+
+Messages from anyone else are silently dropped — no denial reply, since that would just confirm the bot exists to outsiders.
+
+**A small team**
+
+- **Allowed users**: comma-separated `open_id`s of team members
+- Other two blank
+
+**Bot only responds in specific work groups**
+
+DMs are unaffected; only listed groups trigger responses:
+- **Allowed chats**: comma-separated `chat_id`s of the groups
+- DMs are **always** exempt from this list — so you can always DM the bot to change config later.
+
+**Anyone can chat with the bot, but only I can change settings**
+
+- **Admins**: your own `open_id`
+- Other two blank
+
+Others running `/account`, `/config`, `/exit`, `/reconnect`, `/doctor`, `/cd`, or `/ws` get a `❌ 此命令仅管理员可用` reply. Normal conversation (asking the bot to do things) is unaffected.
+
+**Lock everything down**
+
+Fill all three. The `/config` form catches common mistakes — e.g. if your admin list doesn't include yourself, or your chat allowlist doesn't include the chat you're submitting from, the submit is rejected with a message explaining why, so you can't accidentally lock yourself out.
+
+### Finding `open_id` and `chat_id`
+
+Easiest path: have the target user send the bot a message (or `@`-mention it in the target group), then in your terminal:
+
+```bash
+grep '"event":"enter"' ~/.lark-codex/logs/$(date +%Y-%m-%d).log | tail -5
+```
+
+Every line carries `chatId` (group or DM id) and `senderId` (the user's `open_id`). Copy them from there.
+
+The Feishu open-platform "Get user info" API also works but needs the `contact:user` scope, which is overkill if you just need a couple of IDs.
+
+### Worth knowing
+
+- Changes take effect on the **next message** — no restart needed.
+- An empty field means **unrestricted**, not "nobody allowed".
+- To revert a restricted list back to fully open, clear that field in `/config` and submit.
+- DMs are deliberately exempt from the chat allowlist — meaning if you ever accidentally restrict the bot out of every group, **DM the bot and send `/config`** to recover.
+
+### Advanced: editing the config file directly
+
+The `/config` form writes to `~/.lark-codex/config.json` under `preferences.access`:
+
+```json
+{
+  "preferences": {
+    "access": {
+      "allowedUsers": ["ou_xxxxxxxxxxxxx"],
+      "allowedChats": ["oc_xxxxxxxxxxxxx"],
+      "admins":       ["ou_xxxxxxxxxxxxx"]
+    }
+  }
+}
+```
+
+After a manual edit, **restart the bridge** or send **`/reconnect`** from any allowed chat to pick up the changes. The form is usually faster; direct edits make sense mostly for deployment scripts where you want to pre-seed access policy.
+
+## FAQ
+
+**The bot stays silent / Codex never replies.** Usually the `codex` CLI itself is not logged in, or the session points to a cwd that no longer exists. Send `/status` to inspect; `/new` to start a fresh session.
+
+**Codex subprocess looks frozen (card stuck on the last frame).** Since 0.1.20 there's an idle watchdog: if Codex emits nothing for N minutes the process is killed and the card is annotated `⏱ N min no response, auto-terminated`. Disabled by default. Enable with `/config` (global, in minutes), or `/timeout 10` to set it on the current session; `/timeout off` disables for the session; `/timeout default` clears the session override.
+
+**Codex says it can't see the image I sent.** Confirm your local Codex CLI supports `codex exec --image`, then check the downloaded attachment path in the bridge logs.
+
+## Maintainer Release
+
+This project publishes through GitHub Actions + npm Trusted Publishing. See [docs/release.md](./docs/release.md).
+
+## License
+
+[MIT](./LICENSE)

package/README.zh.md

@@ -0,0 +1,228 @@
+# lark-codex-bridge
+
+把飞书 / Lark 消息和本地 Codex CLI 打通的轻量 bot，用一条命令起服务，扫码绑应用，在飞书里和 Codex 对话、让它读图 / 改代码。
+
+> 说明：本项目 fork 自 [feishu-claude-code-bridge](https://github.com/zarazhangrui/feishu-claude-code-bridge)，在其飞书桥接、会话管理、交互卡片、配置向导等能力基础上，将本地 Agent 后端从 Claude Code 改为 Codex CLI，并更名为 `lark-codex-bridge`。
+
+[English README](./README.md)
+
+## 能干什么
+
+- 在飞书（私聊直接发；群里 `@bot`）把消息转给本地的 `codex` CLI，Codex 在你指定的工作目录里工作
+- **流式卡片**：Codex 的文本和工具调用实时出现在同一张卡片上，不用傻等
+- **会话延续**：每个 chat 独立 session，对话能接着上次说
+- **抢占 + 批处理**：中途发新消息会打断旧任务；快速连发几条会合并成一次请求
+- **多工作空间**：`/ws` 切换不同项目，session 自己重置
+- **图片 / 文件**：直接发给 bot，Codex 会读本地下载的文件路径
+- **卡片按钮**：`/help` `/ws list` `/status` 返回交互卡片，点按钮直接操作
+
+## 前置条件
+
+- Node.js **≥ 20**
+- `codex` CLI 已安装并登录：https://developers.openai.com/codex
+- 一个飞书 / Lark PersonalAgent 应用（首次启动的扫码向导能帮你创建）
+
+## 安装
+
+```bash
+npm i -g lark-codex-bridge
+# 或
+pnpm add -g lark-codex-bridge
+```
+
+## 首次启动
+
+```bash
+lark-codex-bridge start
+```
+
+第一次跑会检测到没配置应用，**自动进入扫码向导**：
+
+1. 终端渲染一个二维码
+2. 用飞书 App 扫码
+3. 选择 / 创建 PersonalAgent 应用
+4. 成功后凭据写入 `~/.lark-codex/config.json`
+
+### 开放平台补齐 scope 和事件订阅
+
+向导只负责创建应用，平台侧还需要手动确认：
+
+**权限 scope**：
+- `im:message`
+- `im:message:send_as_bot`
+- `im:resource`
+
+**事件订阅（使用长连接接收）**：
+- `im.message.receive_v1`
+- `card.action.trigger`
+- `im.message.reaction.created_v1` / `deleted_v1`（可选）
+- `im.chat.member.bot.added_v1`（可选）
+
+启用以后再次 `lark-codex-bridge start`，看到 `✓ 已连接` 就可以在飞书里找 bot 对话了。
+
+## 命令速查
+
+### 宿主 CLI
+
+```
+lark-codex-bridge start [-c <config>]   启动 bot
+lark-codex-bridge ps                    列出本机所有正在跑的 start 进程
+lark-codex-bridge stop <id|#>           终止指定 start 进程（SIGTERM，2s 后 SIGKILL）
+lark-codex-bridge --help                列所有命令
+```
+
+> 多开同一个 app 时，开放平台会把事件随机推到其中一个长连接。`start` 启动前会检测同 app 已有的进程，TTY 下提示 `[c]ontinue / [k]ill old / [a]bort` 三选；非 TTY 只 warn 并继续。
+
+其它命令（`status` / `doctor` / `handover` / `workspace` / `service`）是占位，后续版本补。
+
+### 在飞书里用的斜杠命令
+
+| 命令 | 作用 |
+|---|---|
+| `/new` `/reset` | 清空当前 chat 的会话 |
+| `/cd <path>` | 切换工作目录（会重置 session） |
+| `/ws list` | 列所有命名工作空间（卡片 + 按钮） |
+| `/ws save <name>` | 把当前 cwd 存为命名工作空间 |
+| `/ws use <name>` | 切换到命名工作空间 |
+| `/ws remove <name>` | 删除命名工作空间 |
+| `/status` | 当前 cwd / session / agent（卡片 + 按钮） |
+| `/config` | 调整偏好（消息回复方式、工具调用显示等） |
+| `/stop` | 终止当前正在跑的 run（也可点卡片底部 ⏹ 终止 按钮） |
+| `/timeout [N\|off\|default]` | 当前 session 的 idle 探活（分钟）；`/config` 改全局默认。详见下方"常见问题 — Codex 子进程假死" |
+| `/ps` | 列出本机所有 start 进程，标识当前回复的是哪个 |
+| `/exit <id\|#>` | 终止指定 start 进程（自己 = graceful 退出；他人 = SIGTERM） |
+| `/reconnect` | 强制重连 WebSocket（网络抖动后 bot 没反应时用） |
+| `/doctor [描述]` | 把最近运行日志和你的描述喂给 Codex，自助诊断卡住 / 异常的原因 |
+| `/help` | 帮助卡片 |
+| 其它 `/xxx` | 原样交给 Codex |
+
+**消息策略**：私聊 = 不需要 @，任何消息都回；**群（含话题群）= 默认要 @bot 才回**（0.1.22 起的新默认），不 @ 时 bot 完全沉默；@全员永远不响应；云文档评论必须 @bot。要恢复"群里也不强制 @"的老行为：`/config` → "群里需要 @ bot" → 选"否"。
+
+## 数据目录
+
+| 路径 | 内容 |
+|---|---|
+| `~/.lark-codex/config.json` | 应用凭据（App ID / Secret），权限 600 |
+| `~/.lark-codex/sessions.json` | 每个 chat / 话题 的 Codex thread id + cwd（+ 可选的 `/timeout` 覆盖） |
+| `~/.lark-codex/workspaces.json` | 工作空间映射 |
+| `~/.lark-codex/processes.json` | 当前在跑的 start 进程注册中心（`ps`/`stop` 用），死进程会被自动清理 |
+| `~/.lark-codex/media/<chatId>/` | 下载的图片 / 文件，24h 自动清理 |
+| `~/.lark-codex/logs/YYYY-MM-DD.log` | 结构化运行日志（JSON line），按天滚动；启动时清理超过 7 天的老文件（`LARK_CODEX_LOG_DAYS` 环境变量可改）；`/doctor` 命令读它做诊断 |
+
+> 升级自 0.1.11 之前的版本？跑一次 `lark-codex-bridge migrate` —— 自动把 `~/.config/lark-codex-bridge/` 和 `~/.cache/lark-codex-bridge/` 下的内容搬到新位置，并把 `config.json` 升级到新结构。
+
+## Codex 运行配置
+
+在飞书里发 `/config` 可以调整 Codex 子进程参数，改完从下一次 run 开始生效：
+
+- `codexBinary`：Codex CLI 命令或绝对路径，默认 `codex`
+- `model`：传给 `codex exec --model`；留空则使用 Codex 默认配置
+- `profile` / `profileV2`：分别对应 `--profile` 和 `--profile-v2`
+- `sandbox`：`workspace-write`（默认）、`read-only` 或 `danger-full-access`
+- `skipGitRepoCheck`：默认开启，允许 `/cd` 到非 Git 目录
+- `search`：开启后以顶层 `codex --search exec ...` 方式启用 Web Search
+
+也可以直接写 `~/.lark-codex/config.json` 的 `preferences.agent`：
+
+```json
+{
+  "preferences": {
+    "agent": {
+      "provider": "codex",
+      "codexBinary": "codex",
+      "sandbox": "workspace-write",
+      "skipGitRepoCheck": true,
+      "search": false
+    }
+  }
+}
+```
+
+## 访问控制（可选）
+
+默认 bot 是"开放"的：任何能找到它的人都能私聊它，群里 @bot 就触发响应。**个人自己用 / 给朋友用，这就够了**——但如果想给团队用、或者怕在大群里被滥用，可以在飞书里发 `/config`，调下面三栏中的一栏或几栏。
+
+### 几种典型用法
+
+**只让我自己用**
+
+`/config` 表单里：
+- "用户白名单"：填你自己的 `open_id`
+- 其它两栏留空
+
+之后非你发的消息会被 bot 静默丢弃——bot 不会回"你没权限"之类的话，免得暴露它存在。
+
+**只让一小群同事用**
+
+- "用户白名单"：填同事们的 `open_id`，英文逗号分隔
+- 其它两栏留空
+
+**bot 只在指定工作群里干活**
+
+私聊不受影响；群里只有名单上的群才触发响应：
+- "群白名单"：填想让 bot 工作的群 `chat_id`，英文逗号分隔
+- 私聊**永远**不受此约束——意味着你随时能 DM bot 调配置
+
+**谁都能跟 bot 聊，但只有我能改设置**
+
+- "管理员"：填你自己的 `open_id`
+- 其它两栏留空
+
+下次别人发 `/account` `/config` `/exit` `/reconnect` `/doctor` `/cd` `/ws` 这些敏感命令，会收到 `❌ 此命令仅管理员可用`。普通对话（让 bot 帮忙做事）不受影响。
+
+**完全收紧**
+
+三栏全填。`/config` 表单会拦下常见误配——比如管理员名单里没把你自己加进去、群白名单里没包含当前会话，提交时会被拒绝并提示原因，不会让你不小心把自己锁在外面。
+
+### 怎么找 `open_id` 和 `chat_id`
+
+最快的办法：让目标用户给 bot 发一条任意消息（群的话就 @bot 一下），然后在终端：
+
+```bash
+grep '"event":"enter"' ~/.lark-codex/logs/$(date +%Y-%m-%d).log | tail -5
+```
+
+每一行都带 `chatId`（= 群或私聊 ID）和 `senderId`（= 用户 `open_id`），照着复制就行。
+
+也可以查飞书开放平台的"获取用户信息"API，但要先给你的应用加 `contact:user` scope，没必要为了几个 ID 折腾。
+
+### 几点提醒
+
+- 改完 `/config` **下一条消息**就生效，不用重启
+- 把任何一栏设成**空字符串** = 不限制（不是"一个都不允许"）
+- 想从某种受限状态回到"完全开放"，把对应栏目清空再提交即可
+- 私聊不受"群白名单"约束——这是设计上故意的：万一你不小心把所有群都锁死了，**回到 bot 的私聊里发 `/config` 就能解锁**
+
+### 高级：直接改配置文件
+
+不太想登飞书也可以，`/config` 表单背后写的是 `~/.lark-codex/config.json` 的 `preferences.access`：
+
+```json
+{
+  "preferences": {
+    "access": {
+      "allowedUsers": ["ou_xxxxxxxxxxxxx"],
+      "allowedChats": ["oc_xxxxxxxxxxxxx"],
+      "admins":       ["ou_xxxxxxxxxxxxx"]
+    }
+  }
+}
+```
+
+手改完之后**重启 bridge** 或者**找一个被允许的会话发 `/reconnect`** 让新配置生效。日常调整还是用 `/config` 表单更省事，直接改文件主要用在"部署脚本里预填"之类的场景。
+
+## 常见问题
+
+**Codex 挂住不回复**：通常是 `codex` CLI 本身没登录，或者 session 指向了不存在的 cwd。发 `/status` 看当前状态；`/new` 重开会话往往就好。
+
+**Codex 子进程假死（卡片停在最后一帧不动）**：支持 idle 探活：Codex 一段时间没输出就被 SIGTERM kill，卡片末尾会标 "⏱ N 分钟无响应，已自动终止"。默认关闭。开启方式：`/config` 设全局值（分钟），或 `/timeout 10` 只对当前 session 生效；`/timeout off` 关掉某个 session 的探活；`/timeout default` 清掉 session 覆盖回退到全局。
+
+**图片发过去 Codex 说看不到**：确认当前 Codex CLI 支持 `codex exec --image`，并查看日志里的附件下载路径。
+
+## 维护者发布
+
+本项目通过 GitHub Actions + npm Trusted Publishing 自动发布，流程见 [docs/release.md](./docs/release.md)。
+
+## 许可
+
+[MIT](./LICENSE)

package/bin/lark-codex-bridge.mjs

@@ -0,0 +1,2 @@
+#!/usr/bin/env node
+import '../dist/cli.js';