Content Pipeline SEO
wolfganghuang0083/content-pipeline
Pipeline SEO completo: SERP → redação → fact-check → publicação
Ver no GitHubSkill de produção de conteúdo SEO repetível e controlado para Claude Code. Automatiza todo o fluxo: seleção de tópico → análise SERP → redação com FAQ → verificação de fatos adversária independente → geração de imagem → publicação. Usa FACTS.md como única fonte de verdade para evitar fabricações.
Instalação
npx convez-ai wolfganghuang0083/content-pipeline1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.
Métricas
Casos de uso
- Produção em larga escala de artigos longos verificados para blogs/sites
- Automação de SEO com gate de qualidade (revisão ou auto-publicação segura)
- Integração com WordPress/Rank Math com controle de metadados e CTA
Veredito

Ferramenta sólida e bem documentada para equipes que produzem conteúdo SEO em volume. O mecanismo de fact-check adversário independente e a única fonte de verdade (FACTS.md) diferenciam a segurança. Requer Python/Node e configuração cuidadosa de secrets, mas o isolamento da camada CMS torna reutilizável além de WordPress.
README
A tradução para o português deste README ainda está na fila.
content-pipeline
📖 手把手 fork + 使用教學(GitHub Pages):https://wolfganghuang0083.github.io/content-ops-toolkit/
A brand-agnostic, repeatable SEO content production pipeline for Claude Code, packaged as a Claude skill. It turns one keyword into a published, fact-checked, on-brand long-form article:
topic selection → SERP analysis → drafting (+ FAQ + an independent adversarial fact-check) → cover image → draft → publish
Every step reads a single source of truth — your FACTS.md — and an independent
adversarial fact-check acts as a hard safety gate before anything goes live. There
are two operating modes: review mode (default; you approve the draft) and
auto-publish mode (only when three guardrail conditions all pass).
The publish/metrics scripts ship with a WordPress + Rank Math reference implementation, but all site-specific values live in CONFIG blocks / argparse flags / environment variables, and the CMS layer is isolated so you can retarget it.
What's in the box
| File | What it does |
|---|---|
SKILL.md | The pipeline SOP Claude follows (6 steps, guardrails, the two modes). |
templates/FACTS.template.md | Fill-in-the-blanks single source of truth. Copy to FACTS.md. |
scripts/serp_analysis.js | SERP competitive analysis workflow → structured JSON + recommended angle. |
scripts/write_pillar.js | Writes the pillar, appends an 8-Q FAQ, runs the independent fact-check. |
scripts/publish_pillar.py | Markdown → CMS (draft/publish), CTA block, FAQ block, SEO meta. |
scripts/backfill_metrics.py | Backfills impressions/clicks/conversions into your content map (optional). |
templates/COLLAB_RULES.template.md | Optional rules for splitting "mass drafting" vs "QA + publish" across tools. |
Install
Clone into your Claude skills directory:
git clone https://github.com/<your-org>/content-pipeline.git \
~/.claude/skills/content-pipeline
Then create your facts file from the template and fill in the blanks:
cp ~/.claude/skills/content-pipeline/templates/FACTS.template.md \
~/.claude/skills/content-pipeline/FACTS.md
# edit FACTS.md — replace every <PLACEHOLDER>; never put secrets here
Requirements
- Claude Code (the skill + Workflow runner).
- Python 3.8+ for the publish script (stdlib only).
- For
backfill_metrics.pyonly:pip install google-auth google-api-python-client. - Node is only needed if you want to
node --checkthe workflow JS locally.
Secrets are read from environment variables ONLY — never written to any file:
export WP_USER='your-cms-user'
export WP_APP_PW='xxxx xxxx xxxx xxxx' # application password / API token
Site config can come from flags (--base, --site, --cta-link) or env
(CMS_BASE_URL, SITE_URL, CTA_LINK).
Quick start
- Build / open your FACTS.md (positioning, naming rules, anti-fabrication rules, published-article link targets, technical notes).
- SERP analysis for a keyword:
Workflow({ scriptPath: '~/.claude/skills/content-pipeline/scripts/serp_analysis.js', args: { keyword: '<keyword>', outPath: '/tmp/serp.json', factsPath: '~/.claude/skills/content-pipeline/FACTS.md' } }) - Write the pillar (auto-adds FAQ + the independent fact-check at the end):
Read theWorkflow({ scriptPath: '~/.claude/skills/content-pipeline/scripts/write_pillar.js', args: { keyword: '<keyword>', serpPath: '/tmp/serp.json', outPath: '/tmp/pillar.md', factsPath: '~/.claude/skills/content-pipeline/FACTS.md' } })## Independent adversarial fact-check resultsection. If it says "Needs N fixes", fix or report back — do not publish. - Cover image with your cover-image skill (1080×1080 or your brand template).
- Create the draft:
Confirm the verify line showsWP_USER=... WP_APP_PW=... \ python3 ~/.claude/skills/content-pipeline/scripts/publish_pillar.py \ --src /tmp/pillar.md --slug my-slug \ --base https://example.com/wp-json --site https://example.com \ --rank-title '...' --rank-desc '...' --focus '<keyword>' \ --cta-link https://example.com/contact/ --status draftfaq_block=True cta=True stray_md=0. - Publish after review (
--status publish, or update an existing draft with--update-post-id <id>).
How it fits a tool chain
The pipeline is one link in a chain. It does not decide what to write — that's the job of the upstream skill — and it can hand off cluster interlinking downstream.
content-map-builder content-pipeline (this) (downstream)
decides WHAT to write → turns one topic into a → cover-image skill
(clusters, priority, published, fact-checked, internal-linking skill
GSC-ranked backlog) on-brand article metrics backfill
- Upstream — content-map-builder: owns the content map (which keywords, their strategic role, cluster, priority and status). This pipeline pulls one "to-write" topic from that map and writes status + post id back when done.
- This — content-pipeline: topic → SERP → draft → fact-check → cover → draft → publish. The map is the single hand-off point.
- Downstream: a cover-image skill (step ④), an internal-linking skill to weave
published cluster articles together (step ⑥ follow-up), and
backfill_metrics.pyto close the measurement loop back into the map.
Retargeting a different CMS
publish_pillar.py isolates the WordPress REST + Rank Math calls in its api()
helper and a few block builders (to_blocks, the FAQ block, the meta call). The
Markdown parsing above that layer is CMS-agnostic. To target another CMS, swap that
layer for your CMS's API and keep the rest. All URLs/accounts are CONFIG/flags/env,
so you never edit the file per client.
繁體中文摘要
這是一條品牌無關、可重複用的 SEO 內容產線(Claude skill):選題 → SERP 分析 → 撰稿(含 FAQ 與獨立對抗式事實查核)→ 配圖 → 草稿 → 發佈。以 FACTS.md 為唯一事實來源,內建防編造護欄與「審稿模式(預設)/自動發佈模式」雙閘門。安裝:git clone 到 ~/.claude/skills/content-pipeline/,再把 templates/FACTS.template.md 複製成 FACTS.md 填空。密碼只走環境變數(WP_USER/WP_APP_PW),絕不寫進檔案。CMS 以 WordPress 為參考實作、可替換;站台 URL/帳號等放 CONFIG/argparse/環境變數。工具鏈:上游 content-map-builder 決定「寫什麼」→ 本 skill 把一個主題變成已查核、合品牌的上站文章 → 下游接封面圖 skill 與內鏈 skill,並用 backfill_metrics.py 把成效回填內容地圖。
License
MIT