personal-wechat
GitHub通过自托管Wisdom服务操作个人微信，支持登录检查、联系人管理、聊天记录读取与摘要、搜索及消息发送。发送前需用户确认，确保隐私安全。
Trigger Scenarios
Install
npx skills add NeverSight/learn-skills.dev --skill personal-wechat -g -y
SKILL.md
Frontmatter
{
"name": "personal-wechat",
"license": "Apache-2.0",
"metadata": {
"author": "acedatacloud",
"version": "1.0"
},
"connections": [
"personalwechat"
],
"description": "Operate the user's personal WeChat account through their self-hosted Wisdom service (BYOC) — check login status, list contacts\/conversations, read and summarize history, search contacts, refresh the local history DB, and send messages only after explicit confirmation. Use when the user mentions 个人微信, 我的微信, WeChat personal chat, 微信聊天记录, 微信联系人, reading\/summarizing WeChat messages, or sending a WeChat message.",
"when_to_use": "Trigger for the user's personal WeChat account via their own Wisdom server:\ncheck status\/account, list contacts, list recent conversations, read or\nsummarize a chat, query local history, search contacts, or send a message.\nThis acts on the user's real desktop WeChat, so writes are gated behind\nexplicit confirmation.\n",
"allowed_tools": [
"Bash"
]
}
Personal WeChat via Wisdom
Use the user's self-hosted Wisdom service to operate their personal WeChat account. Wisdom runs on a Windows host with WeChat Desktop logged in and exposes an HTTP API.
Credentials are injected by the personalwechat BYOC connector:
PERSONALWECHAT_BASE_URL— Wisdom server base URL, e.g.http://82.156.126.14:8000.PERSONALWECHAT_API_TOKEN— WisdomAPI_TOKEN. Secret — never echo, print, or log it.
The helper sends the token as Authorization: Bearer ...; it never puts the
token in the URL. For queued UI operations such as search and send, the helper
submits the task and waits for /api/tasks/{id} before printing the final
result.
This is the user's real personal WeChat account. Read operations can run directly. Sending messages or files must be dry-run first, then performed only after the user explicitly approves the exact target and content.
CLI
The skill ships a stdlib-only helper:
WX=$SKILL_DIR/scripts/personal_wechat.py
Verify Connection First
Always start with status when the user asks to use WeChat:
python3 $WX status
Expected healthy shape:
{"auth":{"logged_in":true,"wechat_running":true,"page":"logged_in"}}
If logged_in=false, tell the user to open the Wisdom web UI / RDP and scan the
WeChat QR code. If the API returns 401, ask the user to reconnect the Personal
WeChat connector with the current Wisdom API token.
Read Workflows
Current Account
python3 $WX account
Contacts
python3 $WX contacts --limit 50
Recent Conversations
Use the normal conversations endpoint first; it prefers WeChat DB when ready and falls back to Wisdom's app DB.
python3 $WX conversations --limit 20
For decrypted local WeChat history sessions:
python3 $WX conversations --history --limit 20
Messages in a Conversation
First list conversations, then use the id as conversation_id:
python3 $WX messages "34642176898@chatroom" --limit 50 --order asc
Historical Messages
Read from Wisdom's decrypted WeChat local databases:
python3 $WX history --limit 50
python3 $WX history --talker "34642176898@chatroom" --limit 50
python3 $WX history --limit 20 --offset 20
Raw SQL, Read-Only Only
Wisdom permits only SELECT and PRAGMA:
python3 $WX sql MicroMsg.db 'SELECT count(*) AS cnt FROM Session'
Use raw SQL only for diagnostics or targeted metadata queries. Do not dump large message tables unless the user explicitly asks.
Refresh History DB
If history looks stale, refresh the decrypted DB snapshot:
python3 $WX refresh-history
Search
python3 $WX search "Alice"
Search drives the WeChat UI, so it may be slower than local DB history reads.
Sending Messages — GATED
send dry-runs by default. It never sends unless --confirm is present, or
unless an AceDataCloud scheduled task pre-authorized this Skill and you use
--unattended-confirm.
python3 $WX send "Alice" "今晚 8 点开会吗？"
# -> {"dry_run": true, ...}
Show the dry-run output to the user and ask for explicit approval of the exact recipient and text. Only then run:
python3 $WX send "Alice" "今晚 8 点开会吗？" --confirm
Never add --confirm in the first attempt. Never infer consent from vague text.
The user must clearly approve sending this exact message.
Scheduled-task unattended confirmation
When running inside an AceDataCloud scheduled task, the platform may pre-authorize specific Skills for unattended execution. If all of these are true:
AICHAT_UNATTENDED_MODE=trueAICHAT_ACTIVE_SKILLispersonal-wechatoracedatacloud/personal-wechatAICHAT_ACTIVE_SKILLappears inAICHAT_UNATTENDED_ALLOWED_SKILLS
then the user has pre-authorized this Skill for that scheduled task. In that case, use:
python3 $WX send "Alice" "今晚 8 点开会吗？" --unattended-confirm
If the helper returns unattended_confirmation_denied, do not retry with
--confirm; report the dry-run and explain that the task needs this Skill to be
selected in its unattended authorization settings.
Safety Rules
- Never print
PERSONALWECHAT_API_TOKEN. - Treat
PERSONALWECHAT_BASE_URL + API_TOKENas full remote control of the user's WeChat. - For normal chat write/send operations: dry-run first, ask for explicit approval, then re-run with
--confirm. - For scheduled-task unattended writes: use
--unattended-confirmonly when the platform env says this Skill is pre-authorized. - Do not call logout/restart endpoints from the skill unless the user explicitly asks to repair the Wisdom service.
- If Wisdom returns 503 for history, run
python3 $WX refresh-historyonce, then retry the read. - If the server is unreachable, ask the user to check the Windows host / security group / port 8000.
Endpoint Mapping
The helper wraps these Wisdom endpoints:
GET /api/statusGET /api/auth/statusGET /api/accountGET /api/contacts?version=2.0GET /api/conversationsGET /api/conversations/historyGET /api/messagesGET /api/messages/historyPOST /api/messages/history/queryPOST /api/messages/history/refreshPOST /api/searchPOST /api/messages/send(only after--confirmor verified--unattended-confirm)
Version History
- e0220ca Current 2026-07-05 22:55
