# Articles Repo Maintenance > Maintain this articles repository. Use when adding, updating, publishing, indexing, verifying, committing, or pushing articles; when converting source URLs such as WeChat, X/Twitter, Substack, Google/OpenAI/model launch pages, docs, benchmark pages, or product pages into bilingual Markdown articles; and when preserving article images/media, updating README.md/MOC.md, checking bilingual pairs, or returning public GitHub links. - Skill: `quriosity-agent/articles-repo-maintenance` (Agent Skill, multi-file: 18 files) - Install (CLI): `npx skillmds@latest add quriosity-agent/articles-repo-maintenance` - Raw SKILL.md: https://api.skillmd.com/api/skills/quriosity-agent/articles-repo-maintenance/raw - Safety review: pending - Works with: Claude Code, Claude.ai, OpenAI Codex - Category: AI & ML - Author: quriosity-agent (https://skillmd.com/u/quriosity-agent) - Updated: 2026-09-17 - Page: https://skillmd.com/skills/quriosity-agent/articles-repo-maintenance --- # Articles Repo Maintenance ## Core Rules - Work from the articles repository root. - Keep article folders directly under the repo root: `YYYY-MM-DD/YYYY-MM-DD-slug.md`. - Do not create or use a nested `articles/` content directory. - Create bilingual pairs by default: - Chinese: `YYYY-MM-DD/YYYY-MM-DD-slug.md` - English: `YYYY-MM-DD/YYYY-MM-DD-slug-en.md` - Preserve meaningful source media locally under `YYYY-MM-DD/imgs//`. - Embed local relative image links in both language versions. - Update `README.md` as the chronological index and `MOC.md` as the topic map. - Commit and push by default after article work unless the user explicitly says not to. - Return verified GitHub links for Chinese and English articles after pushing. ## Workflow 1. Inspect current state: - Run `git status --short --branch`. - Check `README.md`, `MOC.md`, and recent date folders for the source URL, project name, slug, or topic keywords. - If related coverage already exists, write a distinct follow-up angle instead of duplicating the same article. 2. Resolve source and date: - Prefer the source publish date for the folder and filenames. - Record canonical source URLs in the article metadata/body when useful. - For social posts that point elsewhere, resolve the canonical external page and ground claims there. 3. Preserve media: - Download meaningful screenshots, diagrams, benchmark charts, tweet images, page figures, GIFs, or thumbnails. - Verify file types with `file` when URLs are extensionless or proxied. - Use stable names and embed only local paths. 4. Draft both articles: - Write analytical, operational articles rather than plain summaries. - Include concrete evidence, caveats, and source-specific context. - Use `references/writing-polish.md` for the final structure and humanizing pass. 5. Update indexes: - Add matching rows in `README.md`. - Add matching wikilinks in `MOC.md` without `.md`. 6. Verify before staging: - Check every image link exists. - Check each article has its bilingual pair. - Check `README.md` links point to real `.md` files. - Check `MOC.md` paths point to real article bases. - Run `git diff --check`. 7. Commit and push: - Stage only the current article files, relevant assets, `README.md`, and `MOC.md`. - Do not sweep unrelated dirty files into the commit. - If pushing requires a rebase, use a scoped stash for only the current article files. - Push to the active branch, normally `main`. 8. Verify public links: - Check GitHub blob/raw URLs are reachable before sending them to the user. - Mention any unrelated remaining dirty state if intentionally excluded. ## Source-Specific References Load only the reference needed for the current source: - WeChat / `mp.weixin.qq.com`: `references/wechat-source-articles.md` - X/Twitter article cards: `references/x-article-canonical-blog-cards.md` - X/Twitter longform media: `references/x-article-longform-media.md` - X/Twitter Substack cards: `references/x-article-substack-roadmap.md` - X/Twitter videos or demos: `references/x-video-product-launches.md` - Google Keyword / Google AI launches: `references/google-keyword-source-pages.md` - OpenAI official news/product pages: `references/openai-news-pages.md` - Model launches, model cards, open-weight releases: `references/model-launch-source-pages.md` - Benchmark or leaderboard pages: `references/benchmark-leaderboard-pages.md` - Voice-agent benchmark sources: `references/voice-agent-benchmark-sources.md` - JS-rich product/research pages: `references/js-rich-source-pages.md` - Interactive JS/WebGL apps needing screenshots: `references/js-app-screenshot-cdp.md` - Product landing pages with API docs: `references/product-landing-api-docs.md` - Tencent Docs: `references/tencent-docs-source-articles.md` - Substack / Dwarkesh-style articles: `references/substack-source-articles.md` ## WeChat Shortcut For `mp.weixin.qq.com` sources: 1. Run Baoyu URL-to-Markdown for text/title/date. 2. Use a browser-rendered page and scroll through the full article to trigger lazy images. 3. Extract `document.images`; prefer `data-src` over transparent SVG `src` placeholders. 4. Keep meaningful `mmbiz.qpic.cn` article assets and filter WeChat chrome such as QR codes, avatars, and follow widgets. 5. Download assets with a WeChat referer and preserve GIFs when present. 6. Verify all local media links before committing. ## Writing Polish Use `references/writing-polish.md` before finalizing: - Repair structure before sentence-level polish. - Remove filler, formulaic AI phrasing, unsupported hype, and vague transitions. - Preserve facts, caveats, source dates, and technical claims. - Keep articles analytical and useful: explain why the source matters, what workflow changes, and what teams should verify. ## Completion Reply After a successful article task, reply concisely with: - Chinese GitHub link. - English GitHub link. - Commit hash and message. - Note that `README.md`, `MOC.md`, and local assets were updated when applicable. - Note that public links were verified.