aws-wechat-article-topics

Topic Selection and Headlines

WeChat Official Account Topic Selection & Viral Headline AI Assistant — Hot topic tracking, topic research, headline creation, summary writing, and series scheduling all in one.

Suite Description · This skill belongs to the aws-wechat-article-* all-in-one suite (9 slugs in total, entry point aws-wechat-article-main). Cross-skill relative references depend on the same skills/ directory, so it is recommended to clawhub install the full suite together. Source code:https://github.com/aiworkskills/wechat-article-skills

Capabilities Disclosure

This skill is mainly Agent-driven (conversational topic research, headline generation); the script layer is only used to update this article's metadata.

Credentials: None
Network: The Agent may use web_search / web_fetch(a host built-in capability, not initiated by this skill's script layer).This skill does not call any interface requiring credentials— checking published articles requires WeChat credentials, which is publish 's job
File reads：.aws-article/config.yaml, this article's article.yaml、.aws-article/products/{产品名}/*.md(required reading when the topic involves the user's business)
File writes: under this article's directory topic-card.md、research.md、article.yaml; in series mode {series_root}/{系列slug}/plan.md
shell: may only call {python} {baseDir}/../aws-wechat-article-publish/scripts/article_init.py(purely local YAML writing, no network)

Routing

To produce a finished article and publish it to the WeChat Official Account, or when "what to publish today" requires full orchestration → aws-wechat-article-main. Standalone triggering of this skill is limited tomodifications of existing headlines/summaries。

Configuration Check ⛔

Before any operation, first follow First-time Guidance · Detection Order and proceed only after it passes (or the user explicitly confirms in writing "no check this time").

Four Input Modes

Mode	Trigger Condition	Example
A. Clear topic	The user gave a specific topic	"Write an article about AI Agents"
B. Has a direction	Gave a field but no specific title	"What's good to write about in AI lately"
C. No direction	Only said to pick topics	"What to write this week" "Help me find a few topics"
D. Series planning	Mentions series/column/serialization	"Make an AI introductory series" "Write a 10-article column"

Workflow

选题进度：
- [ ] 第1步：⛔ 配置检查
- [ ] 第2步：⛔ 全局账号三键
- [ ] 第3步：⛔ 归类 A/B/C/D（先于任何联网调研）
- [ ] 第4步：调研
- [ ] 第5步：生成选题
- [ ] 第6步：生成标题与大纲
- [ ] 第7步：⛔ 展示并等用户选
- [ ] 第8步：落盘（topic-card.md + research.md + article.yaml）

For which steps can be skipped in the all-in-one flow and what needs to be supplemented when used standalone, see branches.md "One"。

Step 2: Global Account Three Keys ⛔

Before calling web_search, doing research, or confirming direction with the user, open .aws-article/config.yaml, and check whether article_category、target_reader、default_author is non-empty after trim. If any item is missing:ask the user item by item, after obtaining a clear reply for the current roundwrite back to that file。

Prohibitedfrom article.yaml or other drafts without authorization. Consistent with main "2) Global account constraints" .

Step 3: Classify A/B/C/D ⛔

Before online researchfirst align with the user: whether it isalready hasa specific topic,only hasa general field,has no idea at all, or wants to doa series/column。

The user's current message has already made it clear (directly giving a topic, or explicitly saying "help me find a few topics this week") → briefly confirmand that is enough; no need to repeatedly interrogate.

Prohibited: starting online research when the user is still in the vague state of "only saying to find topics, without mentioning field/preferences"—what is searched out has no direction to converge on and only wastes a round.

Step 4: Research

First read .aws-article/config.yaml: topic boundaries,topic_direction、update_frequency, and account positioning are based on it. If the current table of contents already has article.yaml(for example, the topic has already been decided), also read title / digest, to avoid conflict with subsequent steps.

When the topic involves the user's own business, the business materials library must be read; for the standard, see branches.md "III"。

Use web_search to search + web_fetch to read in depth, to provide data support for topic selection:

Mode	Research goal
A	How competitor articles are written, data support, unique angles
B	Recent hot topics in this direction, what readers care about
C	`config.yaml` account positioning,`topic_direction` and recent hot topics
D	Knowledge system breakdown, competitor series analysis, reader learning path

For the search strategy and search term templates of each mode, see references/research-strategy.md。

Step 5: Generate topics

Use config.yaml of update_frequency to control the quantity: about 3–5 for weekly updates, slightly more for daily updates, slightly fewer for monthly updates; when there is no configuration, follow the conversation agreement.

Mode	Generation rule
A	Around the user's topic, give 3-4 different entry angles
B/C	Filter 3-5 topics, mark the type (🔥hot / 🌲evergreen / 📚series)
D	Plan the series main line + break out the topic for each article (see branches.md "IV"）

Step 6: Generate titles and outlines

Produce one for each topictopic card: title candidates (3-5, mixed styles), entry angle, outline preview, workload assessment, summary candidates. For the output template, see references/output-format.md。

Title styleFor the loading priority and multi-candidate convergence rules, see branches.md "II"; for the 5 built-in styles (suspense/practical/numeric/rhetorical/story), see references/title-presets.md。

Summary candidatesmust be able to stand independently: it will go into article.yaml of digest, and be displayed in the backend article list.Do not write it as a restatement of the opening paragraph of the body—the first > lead in the body is another piece of text; if the two are identical word for word, readers are essentially reading the same paragraph twice (review will block this item).

Step 7: Display and wait for the user to choose ⛔

Must stop and wait for user action. After displaying all topic selection cards, prompt:

请选择：
- 输入编号（如 1）→ 选定该选题
- 「调整 + 意见」→ 按意见修改后重新展示
- 「重新选」→ 换一批选题
- 「组合 1+3」→ 融合多个选题
- 系列模式：「先写第 N 篇」→ 按该篇进入写稿

Prohibited: continuing to write the draft without waiting for the user to choose; assuming the user will choose a certain one; skipping the display and going directly to the next step.

Step 8: Save to disk

After user confirmation:

Article directory: main has already created the directory and it contains article.yaml → Do not change the directory name, write directly into it. Otherwise create {drafts_root}/{YYYYMMDD}-{标题slug}/（drafts_root / series_root subject to config.yaml ).
Topic selection card → topic-card.md
Research summary → research.md
Series mode: series planning → {series_root}/{系列slug}/plan.md
article.yaml: if the same directory does not have it, or the finalized topic information needs updating,this step must create or complete（publish_completed: false; write it in when the title and summary are consistent with the user's selection):
```
{python} {baseDir}/../aws-wechat-article-publish/scripts/article_init.py <本篇目录> --title "…" --digest "…"
```

→ hand over to aws-wechat-article-writing. It must already have .aws-article/config.yaml, and this article's directory is recommended to already have article.yaml。

Branches and detailed rules

One-stop vs standalone use, preset candidate pool convergence, business material library, series planning, who handles the recommendation links at the end of the article → references/branches.md。

Process files

File	Description
`.aws-article/config.yaml`	Global account and topic/writing style constraints; the topic selection is based on this throughout
`article.yaml`	This article's metadata and `publish_completed`; creating a new article for this piece must `false`
`topic-card.md`	Topic selection card (title, summary, angle, outline)
`research.md`	Research summary (search findings, competitor analysis, data points)