跪拜 Guibai
← All articles
Claude · ChatGPT · Gemini

No More Copy-Paste: How Two Siyuan Plugins Turn Self-Media Publishing Into a 10-Minute Morning Routine

By 一缕82年的清风 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Operating across multiple Chinese content platforms without triggering anti-bot systems previously demanded manual copying, formatting, and image hosting; this pattern of pairing local RAG for content generation with official APIs and real browser sessions for distribution is a workable alternative to fragile scraping. It matters to anyone publishing to those networks or building similar multi-platform distribution tooling, though the specific plugins are SiYuan-centric and rely on trusting a browser extension with your logged-in session.

Summary

The workflow, built by an architect known as 一缕82年的清风, replaces web-chat AI with a local-first approach. AI Creation Workshop (siyuan-ai-studio) reads the entire Siyuan document, retrieves personal notes and measured data through RAG, and streams a rewrite in a conversational tone, with one-click replacement of selected paragraphs. Metrix (siyuan-metrix) then handles distribution: for WeChat Official Accounts it uses the official AppID/AppSecret API to generate covers, render code blocks as terminal windows, and push drafts directly; for Zhihu, Juejin, CSDN, Toutiao, Bilibili, and Feishu it wakes a logged-in Chrome or Edge via a companion extension to fill labeled draft boxes, with no passwords or cookies involved.

The motivation is concrete: headless scrapers get caught by risk control (CAPTCHAs, IP bans, UI breakage), and generic Web AI produces articles full of clichés because it lacks the author's private field notes and tested data. The solution pairs a local knowledge base as the "central editorial office" with official APIs where they exist and real browser sessions elsewhere, plus a GitHub+CDN image host to fix cross-platform image breakage. Setup is a three-step install from the Siyuan marketplace, and the author claims both plugins are free to start—Metrix's community edition allows one platform per publish with three free publishes daily, while a paid pro tier unlocks concurrent multi-platform posting and cloud image sync.

Takeaways
— Web chat models produce generic 'AI-flavored' copy because they lack access to your personal notes, bug fixes, and test data; retrieving from a local knowledge base via RAG grounds the output.
— Headless scrapers using Playwright or similar get detected by WebDriver fingerprinting, triggering CAPTCHAs, rate limits, or bans, and break when the platform changes UI.
— Metrix uses the official WeChat API (AppID/AppSecret) to push drafts, auto-generate covers, and convert images to the mmbiz host, avoiding the usual browser-based flow.
— For six other platforms (Zhihu, Juejin, CSDN, Toutiao, Bilibili, Feishu), a browser companion extension reuses your already logged-in Chrome or Edge session, with no password or cookie handling.
— Setup requires installing two plugins from the Siyuan marketplace, configuring an LLM API key, exporting the extension to chrome://extensions/, and filling in WeChat AppID/AppSecret plus a GitHub+CDN image host.
— Metrix's free community edition supports one platform per publish and three publishes per day; a lifetime pro license (bought via Aidfa) unlocks concurrent multi-platform posting and cloud image auto-sync.
— A local proxy bypass in Metrix prevents the WeChat 40164 whitelist error when the machine has a network proxy enabled.
Conclusions

The article's strongest claim is that the real bottleneck in automation is not the writing itself but the integration between a personal memory store and the publishing channels; the author solves it by making the local note editor the center of gravity.

Using a real browser session instead of a scraper sidesteps anti-bot detection but trades that security for a dependency on the user's current login state, which could break if the extension changes or if the platform alters its DOM.

The emphasis on 'removing AI flavor' via personal tone and real data suggests that generic LLM output is a shared pain point; the RAG layer here is what converts generic text into something the author is willing to publish.

The free-tier limits (one platform per publish, three publishes daily) are deliberately sized for an individual maintaining a single main account, which frames this as a solo-creator tool rather than an enterprise solution.

Concepts & terms
RAG (Retrieval-Augmented Generation)
A technique where an AI model retrieves relevant snippets from a local or external database before generating a response, grounding its output in specific documents or data rather than relying only on its training.
Headless scraper
A browser automation script (often via Playwright or Puppeteer) that performs interactions without a visible UI. Platforms detect these via WebDriver fingerprints and often block or throttle them.
Official API vs. browser session
An official API is a sanctioned interface provided by a platform (e.g., WeChat's AppID/AppSecret), while a browser session reuses a user's existing logged-in state. The latter is less reliable but avoids needing a platform-approved API.
mmbiz image host
WeChat's proprietary image hosting used in official account articles; images must be converted to this format for inline display, which the Metrix plugin handles automatically.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗