WorkBuddy, Tencent's free AI, automates your WeChat account: 5 daily topics, drafts in your voice, pushes a ready draft. You pick a topic and tap publish—no code.
Tags:
每天早上，你只干两件事：挑个选题，点个发布。
以前写一篇公众号文章要多久？找选题、翻素材、写稿、排版，一下午就没了。写完还得调配图，还得把 AI 腔一句一句挡掉。一天下来，真正写的那点东西反而是最少的。现在这些活全交给了 WorkBuddy，你早上起来只干两件事：从它给的五个选题里挑一个，然后打开公众号后台点一下发布，中间那些乱七八糟的活，它全包了。
阻碍你日更的，从来不是写作能力，是那些找选题、翻素材、排版、去 AI 腔的体力活。这些活全部包给 WorkBuddy，你只管审稿。
这篇就把我整套流程拆开给你看，最后附一份工程书。新手复制粘贴丢给 WorkBuddy，它自己就能跑起来。你看完会觉得：这事我也能干。
先说一句 WorkBuddy 是什么：它是腾讯推的桌面 AI 工作台，免费下载。很多人以为本地工具就是个聊天框，其实不是。这是腾讯自家做的一款完整的 Agent 工作台。
WorkBuddy 自带浏览器操作、定时自动化任务、全文检索的资料库、凭据管理、MCP。腾讯把一整套能力都塞进这个桌面客户端，刚好把每天发一篇文章需要的活全挂上了。
整条流水线：从搜选题到推草稿箱，WorkBuddy 一步步自己跑。
01 TOPIC SCOUT：把找选题这件事交出去
找选题是最磨人的。你得满网翻，看别人在写什么，还得想自己能写出什么角度。有时候一个小时过去了，空白文档上还是一个字没有。WorkBuddy 的自动化任务每天早上定时跑，它把你领域 24 小时内的热点捞回来，筛掉重复的，给你五个选题，每个选题带一句话的角度，带读者痛点，带一个跟别人不一样的切入点。
早上起来，五个选题已经摆在那里，挑一个就行
为什么是五个，不是十个？给十个等于没筛。早上刚起床，脑子还没开机，五个里挑一个是一眼就能定的事。你也可以不挑，直接报个自己想写的题目给它。
这步跑在 WorkBuddy 的 Automation 里，侧边栏新建一个每日任务，设好时间就行。它走的是腾讯自家的通道，国内网络下跑得稳，不用折腾代理。
02 MATERIAL：素材平时随手存，写的时候自动调
你平时看到好文章、数据报告、实战案例，随手丢进 WorkBuddy 的资料库，不用建文件夹，不用打标签，不用分类，拖进去就行。写哪篇的时候，它全文检索整个库，把跟这个选题相关的材料自动调出来。你半年前存的一篇报告，现在刚好用得上，它自己就翻出来了。你连存在哪里都不用记。
资料随手丢进去，写的时候全文检索自动调
这个习惯养成了，你的资料库就是你自己的脑子备份，越存越多，写出来的东西越是你自己的，而不是网上那些人云亦云的套话。很多人存料存得很痛苦，建了一堆文件夹，打了一堆标签，最后自己都找不到存在哪。这里不用。看到一篇好文章，复制链接丢进去；看到一张数据图，截图丢进去。丢完就忘，写的时候它自己会把该用的那几条摆到你面前。
03 REAL VOICE：出稿不是 AI 腔，是你自己的话
最让人烦的就是 AI 腔。那些套话开场、那些升华总结、那些副词堆成一堆的长句，一看就是机器写的，读者不傻，他们一眼就能认出来。WorkBuddy 出完初稿后自动去 AI 味，把破折号干掉，把长短句打乱。更关键的是，它读过你过去的文章，学你的语气：你怎么起头，你善用什么词，你怎么收尾，它都照着来。
你不用改 AI 腔，因为从头到尾就是你的声音。不是把一篇机器文改成人话，是从一开始就没有机器味。这一步是两层：下面一层先把 AI 腔清干净，上面一层再把你自己的语感贴上去。两层走完，出来的稿子读着就像你自己写的。不是 AI 比人强，是 AI 把你已经写过的东西学了回去。
04 FORMAT & IMAGE：排版和配图自动干
你固定的那套样式，调子、颜色、间距、名片卡，WorkBuddy 记住了就每天照着拼，你不用一个一个标签去调。配图用 OpenAI 最新的图像模型出，封面每天新生成，不复用旧图。教程步骤、流程指引这种场景，它还能出动态 GIF，比一张死图生动得多，你上面看到的那张流程图就是动的。
动态 GIF 用 512x512 尺寸出，微信支持 GIF。步骤演示、前后对比、白模到精修这些场景，能用 GIF 就用 GIF，读者停留更久。
05 PUSH：推到你自己的草稿箱
WorkBuddy 拿你的 AppID 和 Secret 获取 access_token，把封面上传成永久素材，再把排好版的 HTML 调草稿接口推上去，整个过程不用你碰一下。它推完会告诉你一个 media_id，草稿已经在你箱子里了，你打开公众号后台，审一眼排版和配图，点发布。完事。
它永远不替你点发布。审稿和发布是你的活，不是它的。这一条写进了工程书的红线里。为什么最后这一点必须人来？全自动发稿的机器人活不长，机器会把日期写错，会把数据记错，会把一句话的语气拿捣。你审这一眼，就是把你的名字和账号拦在最后一道门前面。
所以这套流程不是无人值守，是你把体力活全交出去，把判断留给自己。每天五分钟，你只干两件事：挑题目，点发布。
06 ENGINEERING BOOK：工程书（复制即用）
下面是完整的工程书，把它复制下来，整段粘贴给 WorkBuddy，把 FILL_ 开头的几个常量填上你自己的（领域、推送时间、公众号 AppID 和 Secret、一篇你过去的文章当语气样本），它就能跑起来。
这是一份纯文本。不写代码，不建仓库，不搭服务器，代码是 WorkBuddy 读懂了自己当场生成的，不是我分发的。你看不懂里面的标记也没关系，AI 读得懂。
为什么把流程写成工程书，而不是写成一套软件？因为软件要装、要配、要维护，新手一起头就卡死。工程书不一样，它就是一段话，谁都能复制，谁都能跑。你不需要懂技术，只需要把它交给一个能跑工程书的 AI 工作台。
::ILANG::v5.0::ENGINEERING_BOOK
[TYPE:engineering_book]
[ID:WB-DAILY-AUTOPUBLISH]
[VERSION:1.0]
[DATE:2026-11-07]
[LANG:en]
::STATE{@USER, role:decision_maker,
does:pick_one_topic_each_morning_and_tap_publish,
skill_level:zero_code}
::STATE{@AGENT, role:WorkBuddy_desktop,
has:automation_scheduler,
has:knowledge_base_fulltext_search,
has:web_search,
has:image_generation,
has:wechat_draft_api,
does:schedule_harvest_draft_format_push,
never:publishes_without_user_tapping_send,
never:leaves_AI_tone_in_final_draft}
::STATE{@CREDIT,
author:ai_chuhai,
spec:https://ilang.ai,
license:MIT}
::OBJECTIVE{
id:g1|owner:user|version:1|status:active
target: every morning a publish-ready draft
sits in the user own WeChat draft box,
the user only picks a topic and taps send
ACCEPT: all RUBRIC pass
NON_GOALS: auto-publishing without review,
managing followers, ad placement
DONE_WHEN: topic list delivered at set time,
chosen topic drafted and pushed to draft box}
::BOUNDARY{
never:auto_publish_skip_user_review|scope:permanent}
::BOUNDARY{
never:fabricate_numbers_or_personal_stories|scope:permanent}
::BOUNDARY{
never:leave_21_banned_AI_words_in_draft|scope:permanent}
::BOUNDARY{
never:store_wechat_secret_outside_local|scope:permanent}
::STATE{@INPUT,
FILL_NICHE:your content niche
default none example cross-border AI tools,
FILL_TOPIC_TIME:when the topic list arrives
default 08:30 local,
FILL_APPID:your WeChat official account AppID
default none,
FILL_SECRET:your WeChat AppSecret
default none kept local only,
FILL_VOICE:a sample of your own past article
default none, used to learn your tone}
::MODULE{SETUP|title:one time preparation}
[STEP] 1 ask user for FILL_NICHE and FILL_APPID
and FILL_SECRET, store secret in local env only
[STEP] 2 ask user to paste one past article
as FILL_VOICE to learn their tone
if none, use plain register
[STEP] 3 build a knowledge base folder:
tell user to drop good articles, data reports,
case files into it over time, no need to sort,
the agent full-text searches it when writing
[STEP] 4 verify WeChat draft API works:
fetch access_token with FILL_APPID and secret
if fails, show the IP whitelist setup step
::MODULE{TOPIC_SCOUT|title:daily topic list}
[STEP] 1 schedule automation at FILL_TOPIC_TIME
[STEP] 2 search last 24h hot topics in FILL_NICHE
across web search and trend sources
[STEP] 3 read the published log to skip repeats
[STEP] 4 output exactly 5 candidate topics,
each with one-line angle, reader pain point,
and a differentiated entry point, as a table
[STEP] 5 deliver the list to the user
the user picks one or names their own
[NOTE] exactly 5. more than 5 is no filter.
::MODULE{MATERIAL|title:gather source material}
[STEP] 1 take the chosen topic
[STEP] 2 full-text search the knowledge base
for material related to the topic
[STEP] 3 also web search for fresh data and cases
[STEP] 4 output a structured brief:
3 quotable points, 2 data points, 2 cases
each with its source noted
[STEP] 5 if a number is vague and search found
the real figure, fill it; if not found,
mark it for the user to fill
::MODULE{DRAFT|title:write with real voice}
[NOTE] two layers: floor cleans AI tone,
voice layer applies the user own register
[STEP] 1 write a draft from topic plus material:
title with a specific number hook,
open with a concrete scene not a cliche,
2000 plus characters, 3 to 5 sections
[STEP] 2 floor clean: delete 21 AI fingerprint
words, kill em-dashes, vary sentence length,
lead paragraphs with the judgment,
paragraphs max 3 lines
[STEP] 3 voice layer from FILL_VOICE sample:
match the user sentence rhythm, their
vocabulary, their way of making a point.
do not insert slang the sample does not have
[STEP] 4 for a money or how-to topic:
make the reader feel they can do it too.
plain words, concrete steps, clear math,
bring the distance close not push it far
[CHECK] zero AI fingerprint words
[CHECK] zero em-dashes
[CHECK] every number is real or marked
::MODULE{IMAGE|title:cover and figures}
[STEP] 1 generate a cover image with an image
model, fresh each day, never reuse old cover
[STEP] 2 generate 2 to 3 in-article figures
[STEP] 3 for tutorial steps or flows,
generate an animated GIF instead of a still
[STEP] 4 upload all images to WeChat CDN
and get mmbiz URLs
::MODULE{FORMAT|title:wechat ready html}
[STEP] 1 convert draft to WeChat HTML:
no markdown, bold for section titles,
key lines bold on their own line,
paragraph spacing, no pre tag
[STEP] 2 match the user fixed style if they have
one: paste a past styled article and copy it
[STEP] 3 append images at matching positions
[STEP] 4 add the user profile card and footer
::MODULE{PUSH|title:to the user own draft box}
[STEP] 1 fetch access_token with FILL_APPID
and FILL_SECRET
[STEP] 2 upload cover as permanent material
get thumb_media_id
[STEP] 3 call draft add API with the HTML,
title, author, digest, thumb_media_id
[STEP] 4 output the media_id and notify user:
draft is in your box, open WeChat admin,
review, then tap publish
[NOTE] the agent never taps publish.
review and publish are the user job.
::MODULE{AUTOMATE|title:daily schedule}
[STEP] 1 TOPIC_SCOUT runs at FILL_TOPIC_TIME
[STEP] 2 wait for user to pick a topic
[STEP] 3 after pick, run MATERIAL, DRAFT,
IMAGE, FORMAT, PUSH in order
[STEP] 4 the user wakes up, picks a topic,
reviews the finished draft, taps publish
[NOTE] the only two human actions:
pick a topic, tap publish. the rest is auto.
::MODULE{DELIVER|title:report}
[LINE] topic chosen
[LINE] draft media_id
[LINE] AI tone check result
[LINE] what to do: review and tap publish
::RUBRIC{R1: topic list has exactly 5 items
|check:human}
::RUBRIC{R2: zero AI fingerprint words in draft
|check:cmd}
::RUBRIC{R3: zero em-dashes
|check:cmd}
::RUBRIC{R4: cover is fresh not reused
|check:human}
::RUBRIC{R5: draft in WeChat box with media_id
|check:status}
::RUBRIC{R6: agent never published on its own
|check:human}
::RUBRIC{R7: every number real or marked
|check:human}
::FALLBACK{
no_knowledge_base_yet
then web search only and note the gap|
wechat_token_fails
then show IP whitelist setup and halt push|
automation_not_supported
then give manual trigger phrase
每天帮我出今天的选题和稿子}
::PRIORITY{
user_decision > BOUNDARY > OBJECTIVE
> RUBRIC > MODULE
}
::ILANG::v5.0::ENGINEERING_BOOK
丢给它，回一句“我要你每天帮我出今天的选题和稿子”，它就走起来了。你要做的，是每天早上那五分钟。
你觉得每天发一篇，难在写，还是难在那些乱七八糟的活？
关于「跨境风向标」：纳斯达克上市公司见知教育（股票代码 JZ）的公众号，专注 AI 出海、联盟营销、独立站、Google Ads 实战。创始人静水流深，创立了AI出海新手实战训练营，运营「丝绸之路·AI出海」社群，提供谷歌套利+亚马逊联盟线上课程和线下课程。
没有评论:
发表评论