Write and publish WeChat Official Account articles end-to-end with AI, or take an existing
draft straight into the skill's built-in formatting + direct-send capability. Supports full
pipeline from topic, plus ready-article direct send when the article is already written.
Use when user says "写公众号文章" / "微信推文" / "发布到草稿箱" / "微信排版" / "WeChat publish".
Do NOT trigger for: generic blogs, newsletters, PPT, short-video scripts, non-WeChat SEO work.
Write and publish WeChat Official Account articles end-to-end with AI, or take an existing
draft straight into the skill's built-in formatting + direct-send capability. Supports full
pipeline from topic, plus ready-article direct send when the article is already written.
Use when user says "写公众号文章" / "微信推文" / "发布到草稿箱" / "微信排版" / "WeChat publish".
Do NOT trigger for: generic blogs, newsletters, PPT, short-video scripts, non-WeChat SEO work.
triggers
["公众号","微信公众号","写公众号","微信推文","草稿箱","微信排版","选题","封面图","配图","把这篇文章发到草稿箱","微信发文","WeChat article","WeChat publish","publish to WeChat"]
AI WeChat Article Writer — From Topic to Draft Box in One Prompt
Write professional WeChat Official Account articles with AI that doesn't sound like AI. Full mode handles topic mining → research → writing → themed formatting → cover generation → draft-box publishing. For ready articles, the skill also has a built-in formatting + direct-send capability that skips the writing pipeline and sends the article straight to the draft box.
Before writing, read references/platform-dna.md (index → writing-guide.md, style-template.md, wechat-constraints.md). Use the specialized playbooks it points to only when the draft needs them.
Non-negotiable. Violating any one means the article has failed:
Read references/writing-guide.md BEFORE writing — pre-writing framework + de-AI protocol are mandatory.
Zero AI-sounding text — run the full 4-level de-AI pass from writing-guide.md.
H1 title: mobile-first, concise, and front-loaded — the converter extracts H1 as the WeChat title, so subject + payoff must appear early.
Digest: ≤54 汉字 (120 UTF-8 byte limit).
Word count: 1,500–2,500 字 — sweet spot for completion is 1,500–2,000.
Specificity over abstraction — every claim grounded in concrete detail.
Depth over polish — run the Depth Checklist; if thesis is in top-3 Google results, rewrite don't polish.
Obey client style.yaml blacklist (words AND topics). No exceptions.
Playbook overrides writing-guide when clients/{client}/playbook.md exists.
Ask about image scope before Step 6 — use AskUserQuestion if available, plain text otherwise.
Always publish to drafts — Step 7 is mandatory and automatic; do NOT ask.
Ready article means ready article — if the user already supplied the article and only wants formatting/publish, do NOT reopen topic ideation, KB mining, or full rewriting unless they explicitly ask for it.
Result Links Rule
Always end draft, publish, list, and stats-review actions with Result links.
Full pipeline: Load config → Mine KB → Trending topics → Dedup + SEO → Topic select → Framework select → Write → SEO + de-AI → Generate images → Publish to drafts → Archive → Report with result links
Built-in formatting + direct-send capability: Validate the supplied article → do the minimum WeChat adaptation → publish to drafts → report with result links
User gave a specific topic → Skip Steps 2–3, go 1.5 → 3.5
User gave raw Markdown or a finished article for formatting/publish only → run the minimal checks from content-adaptation-playbook.md, then jump to Step 7
Resilience: Never Stop on a Single-Step Failure
Every pipeline step has a fallback. If a step AND its fallback both fail, skip that step, note it in the final output, and continue. Never halt the whole pipeline on a single-step failure.
See references/resilience.md for the full per-step fallback chain and environmental failure recovery.
Operations
For post-publish commands (polish, rewrite, change theme, stats review), client onboarding, learn-from-edits, custom themes, and first-run setup, read references/operations.md.
If the request is about improving this skill itself, refactoring its structure, or checking for documentation drift, read references/skill-maintenance.md.
Gotchas — Common Failure Patterns
Six named anti-patterns observed across real runs. Call them out by name instead of re-deriving the problem.
The AI Essay — correct, comprehensive, boring; no person behind it
The Generic Hot Take — summary of other summaries; no unique angle
The Word-Count Pad — verbose instead of deep
The Pretty But Empty Article — styling without substance
The Blacklist Miss — forbidden words slip past the final scan
The Broken Pipeline Halt — one step fails, whole run aborts (never do this)
