English | 中文
一套給 AI Agents 採用的「網站內容營運作業系統」。它把每日發文、內容品質閘門、正式環境驗證、Google Sheets 營運追蹤,以及每週內容集群治理串成可稽核、可回補、可安全重試的完整流程。
這個 repo 不是單一寫作 prompt,也不綁定特定 CMS、LLM 或部署平台。核心價值是把 AI 產文放進工程化的控制面:模型只負責提出草稿;schema、validator、build、live verification 與 tracking ledger 決定內容能否真正上線。
- 每日從有狀態的主題佇列選題並生成一篇內容
- 避免重複選題、搜尋意圖互吃與無限制擴張 core pages
- 針對價格、法規、健康、金融、規格、研究與導購內容執行來源/時效/責任邊界檢查
- 阻止內部治理欄位、prompt 策略或審核備註洩漏到前台
- 同步文章 registry、搜尋索引、HTML、schema、sitemap、feed 與 production build
- 發布後檢查正式 URL,而不是只相信 commit 或 CI 成功
- 以 idempotent upsert 同步 Google Sheets,並支援全量回填
- 每週產出 cluster map、cannibalization、內鏈、孤兒頁、內容缺口與 freshness 報告
Topic Queue
↓
Structured Generation
↓
Editorial + Fact + Contract Gates
↓
Derived Artifacts + Production Build
↓
Commit / Deploy
↓
Live Verification
↓
Google Sheets Ledger
↓
Weekly Cluster Governance ──→ next queue decisions
建議狀態機:
planned → drafting → generated → validated → committed
→ deployed → live_verified → tracked
任何寫入階段均可轉為 needs_review / blocked / failed
Clone 後,把整個目錄放到 agent 可讀取的 skills 目錄,或在任務中直接提供此 repo 路徑:
git clone https://github.com/mkhsu2002/ai-content-operations-skill.git
cp -R ai-content-operations-skill ~/.codex/skills/對 agent 說:
Use $ai-content-operations-skill to audit this website repository and design a safe daily publishing, Google Sheets tracking, and weekly cluster-governance workflow.
依 導入指南 盤點 repo,先啟用:
- content record validation
- cluster audit report
- Google Sheets dry run
- fixed-topic generation dry run
- production build 與 live verifier smoke test
不要在第一天就打開排程寫入。只有同一筆測試內容通過 local、CI、deployed page、live verification 與 Sheet upsert,才啟用每日自動發布。
assets/content-record.schema.json:可攜式文章紀錄 schemaassets/google-sheets-headers.csv:建議的追蹤欄位assets/github-actions-daily.yml:每日流程骨架assets/github-actions-weekly.yml:每週治理骨架
這些是模板,不應直接硬套。請先調整 content path、build command、deploy trigger、canonical 規則、secret 名稱、時區與 provider contract。
python3 scripts/validate_content_record.py path/to/content-record.json
python3 scripts/audit_content_clusters.py path/to/articles.json --threshold 0.62 --max-core 12audit_content_clusters.py 的相似度只是一個 review signal;不可依分數自動刪文、redirect、canonical 或 noindex。
Repo 或 CMS 建議維持內容 source of truth,Google Sheets 作為營運 ledger。同步必須:
- 以
article_id為主鍵,舊資料遷移時才 fallback 到 canonical URL。 - 驗證固定 header contract。
- batch update 既有列、append 新列。
- 提供
--dry-run與--allbackfill。 - 預設拒絕本機寫入正式 Sheet,除非有明確開關。
- 不列印 service-account JSON 或 token。
若要建立新的 Google Sheet,請遵循組織的 Drive 父層與 ACL 政策;skill 本身不保存任何 spreadsheet ID 或憑證。
模型輸出一律視為未信任草稿。以下情況應 hard fail:
- contract、stable ID、canonical、圖片或文章檔缺漏
- title/canonical 重複,或搜尋意圖無法與既有內容區隔
- 高風險斷言缺外部來源、資料時間或限制說明
- 內部治理欄位或策略文字出現在前台內容
- registry、頁面、canonical、OG、JSON-LD、sitemap、feed 或 search index 不一致
- production build 失敗
- 正式頁持續回傳錯誤、舊內容或錯誤 metadata
完整規則請見 品質閘門。
每日新文預設為 supporting。每週針對整站執行:
core / supporting / conversion / regional / update角色盤點- intent similarity 與 cannibalization review
- core page 數量上限
- parent、adjacent-risk、next-step/conversion 內鏈檢查
- orphan、broken link、stale source 與內容缺口盤點
合併、刪除、redirect、canonical 與 noindex 都可能破壞累積價值,必須人工審核。詳見 每週治理。
SKILL.md AI agent 的核心執行指令
agents/openai.yaml skill UI metadata
references/ 按需讀取的詳細 SOP
assets/ schema、Sheets 與 CI 模板
scripts/ 無外部依賴的驗證/稽核工具
README.md 中文採用指南
README.en.md English guide
LICENSE MIT
- 不要把 API key、token、service-account JSON、固定 Sheet ID 或私有站點資訊提交到 repo。
- 不要把 chain-of-thought、prompt history、一次性事故全文或內部策略草稿塞進公開 skill。
- 用環境變數或 CI secrets 傳入憑證。
- 修改內容/追蹤契約後,同步驗證程式碼、部署版本、Sheet header 與 live behavior。
如果這個專案對你有幫助,歡迎請我喝杯咖啡,支持後續維護與更多開源工具的開發:
- HotGits:GitHub 熱門專案追蹤與 AI 分析工具。
- EC SEO Booster:協助電商網站進行 SEO 優化與內容增強。
- OpenClaw Agent Feeds:為 AI Agents 整理與供應可追蹤的資訊來源。
- AI Auto BD Starter Kit:AI 自動化商務開發的入門實作套件。
如果你的團隊需要把這套內容營運機制進一步整合到既有 CMS、資料庫、Google Sheets、內部審核流程、AI 模型、部署環境或企業治理規範,可前往:
如果你對 AI Agents、自動化系統,以及 AI 將如何改變工作、企業與未來文明感興趣,歡迎追蹤: