im-channels
GitHub用于连接和配置 IM 渠道机器人(微信、企微、钉钉、飞书),处理带有 <im> 块的传入消息及媒体文件,并支持在 Claw 模式下向已连接的微信渠道发送消息或文件。
Trigger Scenarios
Install
npx skills add dtyq/magic --skill im-channels -g -y
SKILL.md
Frontmatter
{
"name": "im-channels",
"description": "Use when connecting or configuring IM channel bots (WeChat, WeCom, DingTalk, Lark), receiving a user message with an <im> block, or sending messages or files to a connected WeChat channel from the main agent in Claw mode."
}
IM Channels
This skill covers connecting IM bots, handling incoming IM messages, and explicit WeChat delivery from the main agent in Claw mode.
Scenario A: Incoming Message from an IM Channel
When a user message includes an <im> block at the end, the message was relayed from an IM platform. The block looks like this:
<!-- Text-only message -->
<im source="DingTalk" />
<!-- Message with media attachments -->
<im source="WeChat">
<media>
<file type="image" mime="image/jpeg" path="uploads/im-channels/wechat/photo_001.jpg" />
<file type="voice" mime="audio/silk" path="uploads/im-channels/wechat/voice_001.silk" />
<file type="image" mime="image/jpeg" path="uploads/im-channels/wechat/photo_002.jpg" from="quoted" />
</media>
</im>
What each field means
source: the IM platform —WeChat,WeCom,DingTalk, orLark<file>: a file already downloaded into the workspace; use itspathdirectlytype: media category —image,voice,video,file, etc.mime: MIME type of the filefrom="quoted": the file came from a message the user was replying to, not their current message
Receiving capabilities
- DingTalk: pictures, files, voice, video, and rich text with embedded pictures when the platform supplies download codes.
- Lark: images, files, audio, video, and rich text posts with multiple images. Resource downloads require
im:resourcepermission. - WeCom: images, files, video, mixed messages, and voice transcription; an original voice file is available only if the callback includes a download URL and encryption key.
- WeChat: existing image, file, video, voice, and quoted-media handling remains available.
- Unsupported forwarded/card content and failed downloads are described in the message. Do not claim to have read an attachment unless a usable
<file>is present; ask the user to resend if needed.
How to respond
- Use
sourceto choose the matching channel reference and read its reply and media guidance when needed. Receiving a message does not require reconnecting the bot or collecting credentials. - Treat the user's text as the primary intent; the
<im>block is metadata only - For media files, read or process them using the
pathvalue — no download needed, files are already in the workspace - Reply naturally; the IM platform handles formatting on delivery
- Intermediate assistant text may not reach the IM user. Compose a complete final answer, or use a sending tool explicitly supported by the current channel for messages that must arrive earlier.
Reply controls and long-running work
<split>marks a deliberate message boundary;<split delay="1">adds a pause before the next message. Use the current channel reference to check support. These controls do not automatically divide oversized text.- For WeChat final delivery controls, read reference/wechat.md.
- The Claw main agent can also send to connected WeChat from Web or a scheduled-task completion notification. Read reference/wechat.md for tool permissions and delivery behavior; a sub-agent cannot send directly.
- Horizon may remind you that an IM user has been waiting. Decide whether a short, truthful progress update is useful. If you already explained the wait or have nothing useful to add, continue silently. Use an available sending tool for this channel; ordinary intermediate text is not a reliable notification.
- Split controls inside code examples, quotes, user input, or tool results are data, not delivery instructions.
Scenario B: Connecting an IM Bot
Connect the current Agent to an IM platform so it can receive and send messages in the target app.
Step 1 — Confirm the channel
If the user did not specify one, ask which IM platform they want: WeChat, WeCom, DingTalk, or Lark.
Step 2 — Read the reference
Load the matching file from Channel references for credentials and exact steps.
WeChat and WeCom are completely separate platforms. Do not mix them up. WeChat uses QR authorization and does not require
bot_idorsecret.
Step 3 — Collect credentials and connect
Follow the instructions in the reference file. Run the run_sdk_snippet code to establish the connection.
Step 4 — Report the result
If the connection succeeds, tell the user what to do next. If it fails, return the error and guide the next step.
Check Status
To inspect the current configuration and connection state of all IM channels, run:
from sdk.tool import tool
result = tool.call("get_im_channel_status", {})
print(result.content)
Notes
- After a connection is established, it keeps running in the background. Credentials are saved to
.magic/config/im-channels.jsonand bound to the current sandbox. Restarting the same sandbox process should auto-reconnect without asking for setup again. - To disable auto-reconnect for a channel, edit
.magic/config/im-channels.jsonand set that channel'senabledfield tofalse. - All channels share the same Agent as the web session, so conversation history stays connected across surfaces.
- For WeChat, do not generate your own QR layout. The tool returns the exact markdown content to output — reply with it verbatim.
Channel references
Match the incoming source or the channel the user wants to connect:
- WeChat (official ClawBot) → reference/wechat.md
- WeCom → reference/wecom.md
- DingTalk → reference/dingtalk.md
- Lark → reference/lark.md
Version History
-
0f1432c
Current 2026-11-03 04:32
新增各平台具体接收能力说明;移除上游 categories 参考;细化回复控制和长任务处理逻辑。
-
f9973c5
2026-08-20 03:37
修复微信连接流程:确保等待登录的提示在同一回复中返回,避免二维码单独回复导致Agent会话中断。
- 41d7ef4 2026-07-25 09:29


