wewrite-topic — Topic Selection
Operating Conventions
- CLI: Deterministic operations use the command (must be in PATH; if missing, guide users to run , or in the repository).
- {home}: User state directory = or (check via ). All config/style/history/playbook/output/exemplars are stored in {home}, not in the repository; state paths in references documents follow this convention.
- = Use a file reading tool to fully read the file before proceeding; this is not a comment.
- references/: This skill comes with ; in references documents refers to the directory of this skill.
- Task State: Read and updated via
wewrite run show/update/step
; do not modify state files directly.
Prerequisites
- If exists → extract and . If not → ask the user "What are the main
topics your WeChat Official Account covers?" Use the answer as a temporary replacement for , and prompt that wewrite-style can be used for complete setup.
2.1 Hotspot Crawling
bash
wewrite hotspots --limit 30
Fallback: If the script reports an error → WebSearch "今日热点 {first vertical category of topics}"
2.1b High-Frequency Demand Reference
Search for WeChat Official Account articles from the past week for each of the first 1-2 vertical keywords in
:
bash
wewrite search-articles "{垂类词}" -n 15 -t 2
Outputs a list of WeChat Official Account articles in the same vertical category from the past week (title/summary/publish time/account), which serves as a "validated content demand"
signal for use in section 2.3. Sogou results do not include read counts, so they can only indicate the frequency of similar topics, and cannot claim that the articles are "viral".
Fallback: If the command reports an error / returns empty results (Sogou anti-crawling rate limit) → skip this step, do not block the process, and the topic composition falls back to default.
2.2 Historical Analysis + SEO
读取: {home}/history.yaml(不存在则跳过)
Historical analysis (when stats data is available):
- Count which of articles performs the best (read count/share rate) → weighted when recommending frameworks
- Count which of articles performs the best → reference when selecting enhancement strategies
- Deduct scores for keywords that have been written in the past 7 days (deduplication)
Fallback: If the SEO script reports an error → use LLM for judgment; if history has no stats → skip performance analysis and only perform deduplication.
2.3 Topic Generation
读取: {skill_dir}/references/topic-selection.md
Generate 10 topic ideas, including:
- 5-6 hot topic ideas: Based on the hotspots from 2.1, scored according to the rules in topic-selection.md
- 2-3 high-frequency angle topics (only when 2.1b has output): Based on similar articles from the past week, scored according to the rules in
topic-selection.md, marked as "high-frequency demand"; if 2.1b has no output, this quota is reallocated to hot topics
- 2-3 evergreen topics: Not dependent on hotspots, generate long-tail content (tutorials/methodologies/experience summaries/tool recommendations) from the user's domain, marked as "evergreen". Suitable for users with of 干货型/测评型 (practical content / review type)
Each topic includes a title, score, click-through rate potential, SEO friendliness, and recommended framework.
- Auto mode → select the highest-scoring one
- Interactive mode / standalone activation → display all topics and wait for user selection
Completion
Use
wewrite run update --patch
to write
,
,
("热点抓取" / "高频需求" / "常青"),
, then run
wewrite run step topic completed
.
When activated standalone and the user only wants the topic list, just display the 10 topics; after the user makes a selection, write to the state and prompt
"You can directly say '就写这个' to start writing (wewrite-write)".