fengyun-publish

fengyun-publish is a Python-based automation pipeline that runs the full publishing workflow for a solo-operated WeChat Official Account called "Research Agent Cloud." The pipeline takes a topic from initial selection all the way to a finished draft sitting in the WeChat drafts inbox, ready for the author to approve and publish with one tap. No human steps are required in between.

The process has 19 stages, each building on the previous. It starts by gathering trending topics from several sources, including RSS feeds from other public accounts and academic paper lists, then picks the most relevant one using a scoring system. An AI writing assistant then drafts the full article based on the selected topic and the account's established writing style.

Once a draft exists, it goes through a three-track review. One track scores the article numerically on dimensions like predicted engagement. A second track asks whether the article fits the voice and sensibility of the account. A third track checks whether the opening reads like something the author would genuinely publish. All three tracks vote independently, and a set of rules decides whether the draft proceeds, needs revision, or is stopped. If revisions are needed, the system loops up to three times before making an automatic final decision.

After passing review, the pipeline handles layout using a specific typographic style with warm ivory backgrounds and clay-orange accents, generates cover images and in-article illustrations via an image generation service, and pushes the assembled draft to WeChat via its API. A gating script checks that each previous step actually ran and did not produce a placeholder result before allowing the publish call to proceed.

The setup requires Windows, Python, Docker for RSS feed services, and an Anthropic API key. Configuration is done through a single environment file. The repository includes 48 scripts in a tools folder, documentation for technical decisions, and research reports from earlier development phases. Corpus data from other authors and private drafts are not included in the public release.