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.
Get API Key → · More Skills →
Onboarding
⚠️ MANDATORY: When the user just installed this skill, show this message IMMEDIATELY. Translate to the user's language:
✅ AI WeChat Article Writer installed!
Tell me a topic or paste a finished draft — I'll either write it end-to-end or just format and publish it to your WeChat draft box.
Try it now: "帮我写一篇关于 AI 编程的公众号文章"
What it does: full article pipeline when you need writing, or built-in formatting + direct send when the article is already ready.
Setup (one-time):
cd toolkit && npm install && npm run build && cd .. && pip install -r requirements.txt && mkdir -p ~/.youmind/config && cp shared/config.example.yaml ~/.youmind/config.yaml- Get a YouMind API Key → fill
youmind.api_key in ~/.youmind/config.yaml
- Bind your WeChat Official Account in YouMind Connector Settings (paste AppID + AppSecret; YouMind encrypts + proxies, no IP whitelist needed)
preview and themes work without WeChat connection. Need help? Just ask.
For first-run setup and client onboarding details, see references/operations.md.
Usage
Provide a topic, brand/client name, or a finished article:
帮我写一篇关于 AI 编程趋势的公众号文章 — full pipeline from topic给 demo 客户写一篇推文，主题是远程办公 — client-scoped run把这篇 Markdown 排版成公众号样式并发布到草稿箱 — skip writing, publish only把这篇现成文章直接发到公众号 — keep the article as source of truth, do a light WeChat fit pass, then publish用交互模式… — pause at topic / framework / image / theme decisions
Setup
Prerequisites: Node.js ≥ 18, Python ≥ 3.9, a verified WeChat Official Account bound in YouMind Connector Settings.
Run the one-time setup from references/setup.md, then verify with:
node toolkit/dist/cli.js validate
For screenshots and connector binding walkthroughs, see README.md §安装. For onboarding and post-setup operations, see references/operations.md.
Skill Directory
Read files on demand — do NOT load everything upfront. Five always-relevant entries below; full inventory in references/skill-directory.md.
| Path | When to read |
|---|
references/platform-dna.md | Before any content work |
references/content-generation-playbook.md | Generating from a topic/idea |
references/content-adaptation-playbook.md | Adapting existing article |
references/pipeline.md | Running the writing pipeline |
references/writing-guide.md | Step 4 (writing + de-AI) |
references/skill-directory.md | Full file inventory (Tier 2/3 + clients + toolkit) |
Draft Location Rule
Write drafts to ~/.youmind/articles/wechat/<client>/<slug>.md; only fall back to output/ if ~/.youmind/ is unavailable.
For the full path contract, execution modes, and result-links behavior, see references/runtime-rules.md.
Content Modes
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.
Intent routing:
Quality gates before publish: (1) de-AI pass per writing-guide.md, (2) playbook Step 6 self-critique, (3) Step 7/8 conformance report, (4) auto-publish to draft box (mandatory).
Execution Modes
The skill has two modes: auto and interactive.
Read references/runtime-rules.md for exact pause behavior and mode triggers.
Critical Quality Rules
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.
The exact contract lives in references/runtime-rules.md.
Pipeline Overview
Two execution lanes:
- 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
Full per-step detail: references/pipeline.md.
Routing shortcuts:
- 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)
Full explanation, fix, and detection method for each: references/gotchas.md.
References
