Mogao Content Studio

English | 简体中文

Format Obsidian notes for WeChat Official Accounts, Zhihu, Xiaohongshu and X. Choose one of six built-in templates, adjust typography and colors, or learn reusable styles from an article URL. The interface is currently in Simplified Chinese.

Desktop only. This branch: 0.1.18 preview build with browser assistant 0.3.8. Requires Obsidian 1.8.7 or later; earlier native validation used 1.13.7. Formatting preserves your source text and runs locally. No AI account, API key or automatic publishing is required.

Features

Six templates: Ink (墨白), Cinnabar (朱砂), Bamboo (青竹), Blueprint (蓝图), Cream Notes (奶油手记) and Black & Gold (黑金刊物). Save adjustments as a new template, or import/export template JSON.

Images are loaded automatically from Vault attachments and public HTTP(S) URLs when you open a note in the studio. Their positions and order are preserved. Rich copying and package export stop if any image is unresolved, with a specific source and retry action. Each note supports up to 40 distinct images, at most 8 MB per image, with three simultaneous downloads.

Installation

1. Download main.js, manifest.json and styles.css from Releases.
2. Create .obsidian/plugins/content-studio/ inside your Vault and place the three files in that folder.
3. Enable Mogao Content Studio in Settings → Community plugins.
4. Open a note and use the newspaper ribbon icon, the 排版当前笔记 command (format current note), or 用墨稿排版 in the note's context menu.

For updates, replace only main.js, manifest.json and styles.css; keep data.json to preserve settings and custom templates. The plugin does not update itself.

Install the browser publishing assistant

The browser assistant is currently an unpacked Chrome/Edge extension, not a store listing. Mogao can open its onboarding page after you click a publish action. To install it directly from GitHub:

1. Open the repository Actions page and open the latest successful Verify plugin or Release plugin run.
2. Download content-studio-package (verification run) or content-studio-0.1.18-package (release run) from Artifacts. Unzip it and locate mogao-browser-extension-0.3.8-preview.zip.
3. Unzip that assistant archive. In Chrome open chrome://extensions (or edge://extensions), enable Developer mode, choose Load unpacked, and select the extracted mogao-browser-extension directory. Select the directory, not the ZIP file.
4. Confirm the extension card shows version 0.3.8. The first toolbar click may ask to open Obsidian; allow it so the assistant can return to Mogao.

To update an unpacked installation, replace the files in the same directory and click Reload on the extension card. Confirm 0.3.8 before publishing again. The assistant only receives a draft after a user-started publish action; it does not enumerate the Vault or read the system clipboard.

Workflow

1. Select a note with 选择笔记, read the active note with 读取当前笔记, or paste Markdown in the studio.
2. Write in the full-size editor; switch to 预览 for preview or 对照 for a side-by-side view on wide panes. Templates and styling live in the 模板与样式 drawer.
3. Wait for the image counter. Use 重新载入图片 to retry failures.
4. For WeChat, Xiaohongshu long-form, Zhihu or X Articles, click 发布到平台 in Mogao. The companion browser extension receives the current draft and images, opens the destination, waits for login if necessary, then uses its platform adapter to synchronize the separate title, structured body and inline images. There is no per-article file selection or import step.
5. If the extension is missing, the browser opens installation onboarding with the bundled preview extension download. After installation, subsequent articles start directly from Mogao. On X and Xiaohongshu, the workbench exposes separate 发布长文 and 发布图文 actions.
6. Review the platform draft and its save status, then publish manually. Full extension acceptance on real platforms remains pending. Title/body copying and file exports remain available as secondary tools. X Articles requires the corresponding account access; ordinary X threads retain copying.

Each export creates a new folder under 墨稿导出/ by default, configurable in plugin settings. Packages include article.html, article.md, title.txt, caption.txt, template.json, manifest.json, and numbered image files. Markdown references are rewritten to the image files. HTML embeds the image bytes. Xiaohongshu exports additional PNG cards; X exports separate thread text files.

Draft edits are held in the open studio until exported; export before closing or reloading. Export manifests include the source note's relative path, so treat them as part of your private writing material.

Learn styles from a link

链接学模板 reads a public article and up to three stylesheets, extracts constrained typography, colors, headings and quotation styles, and previews them against your draft. Save the result with a name. Template JSON retains provenance and extraction notes, without copying the reference article's body.

Style learning skips scripts, embedded image data and other media while reading; it never requests the reference article's image URLs. Preview images become placeholders without changing the draft or its publishing images. Both URL and pasted-HTML inputs allow up to 20 MB of source data and retain at most 3 MB of reduced text and styles.

From 0.1.16, sampling follows nested text spans and their paragraph/container styles. It retains the dominant body font, letter spacing, indentation, blank-line spacing, heading foreground/background pairs and the main article shell. Reference templates start from neutral styling instead of inheriting built-in decorations. Switch between 样式示例 (headings/body/emphasis/quotes) and 当前稿件 before saving. Relearn a URL to use the new sampler; existing saved templates are preserved. Extraction is not a visual similarity score: image-based ornaments, conditional CSS, multi-column layouts and complex graphics still need review.

For a login page, verification challenge or failed download, use 粘贴 HTML with article HTML you can access. This is static style extraction: scripts, remote fonts, animation and complex nested designs are not reproduced pixel for pixel. No language model is called.

Images and destination compatibility

PNG, JPEG, GIF, WebP and AVIF attachments are supported, including Obsidian ![[image.png]] embeds and Markdown image references. SVG, PDFs, audio/video and other plugins' dynamic blocks are not embedded as images.

Rich copying writes text/html plus text/plain, with native selection copying as a compatibility fallback. Clipboard HTML includes embedded image data, which platform editors can reject. An HTML image tag is different from a clipboard image file: the per-image and per-card buttons write image/png without competing text formats. GIF image copying is static; source images over 40 megapixels must be resized first. Copy controls are excluded from article HTML and cards.

Ordinary Xiaohongshu photo posts have separate image and caption areas; long-form articles have a body editor. Paste images into the appropriate surface. Browser synchronization sends this article and its media only after the user clicks publish in Mogao. Login and final publication remain user-controlled. Local clipboard validation does not imply the destination accepted the images. DOCX and numbered image exports remain available.

DOCX retains basic paragraph and inline formatting, lists, tables, links and embedded images. It simplifies template decorations and maps light text to readable ink on a white page. WebP/AVIF/GIF become PNG (GIF is static); oversized images are bounded to 2000 px on their longest edge. Platform exports include separate title.txt and body.txt. Content package caption.txt contains the body only.

Long paragraphs are paginated for Xiaohongshu cards. An oversized indivisible table or image reports an adjustment requirement. Destination limits on text, image count, and X long-form posting depend on the platform and account.

Network and privacy disclosure

• Reading/selecting a note in the studio, editing in image URLs, retrying images, or completing a copy/export with unresolved images downloads the referenced remote images. The image host receives its URL path/query and normal request metadata. Local attachments stay local.
• Link-based template learning requests the supplied article URL and up to three stylesheets. Requests do not include browser cookies, platform credentials or the entire note body.
• When the system DNS returns a TUN proxy's 198.18.0.0/15 fake addresses, the downloader queries Cloudflare DNS-over-HTTPS (cloudflare-dns.com) for the target hostname only. It does not send the article body or URL query to that DNS service.
• The downloader validates public addresses and redirects and caps response size and time. It does not execute fetched scripts.
• No telemetry, model service, unsolicited synchronization, automatic updates or final publication. A user-started browser delivery polls only its newly opened destination until ready or cancelled.
• Vault enumeration occurs only when the user opens the note picker: it lists Markdown note paths for selection, without bulk-reading their bodies or transmitting the list. Opening the plugin or formatting the current note does not enumerate the Vault.
• Clipboard access is write-only, triggered by the copy buttons. The plugin never reads the clipboard, monitors it, or uploads clipboard contents.
• Settings and custom templates are stored in the plugin's data.json; exported drafts are stored in the Vault.

See Data and network details.

Development

npm ci
npm run verify
npm run verify:extension
npm run package

npm run verify includes the official Obsidian ESLint recommended rules (zero warnings required), CSS lint, type checks, and 122 regression tests. npm run verify:extension adds 79 browser delivery tests, including real Tiptap/Draft.js models and offline WeChat contracts; these are not live platform acceptance. CSS is also checked against a conservative Chrome 120 compatibility profile. The development browser prototype was removed from the plugin source in 0.1.2; run the workbench inside Obsidian to test actual Vault behavior.

Releases are built from version tags by GitHub Actions. Only the three supported plugin files are attached to a Release; optional ZIP/checksum packages are available as Actions artifacts or via npm run package. The workflow generates GitHub build-provenance attestations for all three release files. After downloading them, verify provenance with:

gh attestation verify main.js --repo AI-Scarlett/obsidian-content-studio
gh attestation verify styles.css --repo AI-Scarlett/obsidian-content-studio
gh attestation verify manifest.json --repo AI-Scarlett/obsidian-content-studio

See Review response for the source, CSS, release and behavior checks.

Source: src/main.ts (Obsidian host), src/ui/studio.ts (workbench), src/core/ (rendering, images, templates and export). CI builds and tests the source. Verification scope · Community submission guide · Report an issue.

License

MIT © 2026 AI-Scarlett. See LICENSE and third-party notices.

Browser publishing assistant (preview)

Start in Mogao with 发布到平台. The extension automatically opens the selected platform, waits for login, and synchronizes the current article with inline image uploads and separate title. Missing extensions lead to installation onboarding with a bundled download. Installation, permissions and validation limits. Full real-platform extension acceptance remains pending.

Click the browser assistant toolbar icon to open Obsidian and its Mogao workbench. Existing workbench edits are preserved; a newly opened workbench reads the active note. The first launch may require the browser’s external-app confirmation. The last-used vault must have Mogao enabled. On a pending publish handoff page, the icon continues that handoff instead.

The 0.1.18 preview bundles browser assistant 0.3.8, with platform-specific media and draft adapters. The Zhihu multiple-image fix requires assistant 0.3.8: replace the files in your existing assistant directory and reload the extension once. See the release notes and browser assistant documentation for current validation boundaries.

普通图文发布

在小红书或 X 平台下，直接点击底部 发布长文 或 发布图文。发布图文会打开对应的普通图文编辑器，按笔记顺序上传原图。无需导出文件、选择文件夹或逐张粘贴图片。平台最终发布仍由你操作。

• 小红书：自动填写独立标题和正文，文末 #话题 会预填到可编辑的话题栏，随后通过平台话题候选列表插入原生话题标签。至少 1 张、最多 18 张图片；标题不超过 20 字，正文含话题不超过 1,000 字。话题选择失败会明确提示，已上传图片保留。
• X：普通帖将标题放在首行，正文和话题在后，最多 4 张图片作为帖子附件。普通帖使用纯文本，图片不嵌在段落之间。墨稿不拦截 X 字数，也不截断或拆帖；由 X 判断当前账号的发布限制。
• 支持 PNG、JPEG、WebP。X 普通帖每张不超过 5 MB；小红书本工具每张不超过 8 MB。图片数量、格式或大小超限会停止发送；小红书仍检查标题和正文限制。
• 这两个入口已做模型与上传控件回归测试；本版真实账号多图上传尚未完成验收。见发布说明。

Template engine in 0.1.18

Built-in themes now have distinct paragraph typography and spacing. Reference sampling uses Juice to resolve CSS selectors and variables, and transfers supported nested heading boxes and number slots without retaining the original article text or media. Existing templates are preserved; resample a source to acquire the new structure. See research and limitations, including the GitHub implementations studied.