Many thanks to @dontbesilent for the open-source spirit!
Before the new year, I happened to come across a post by the author sharing a workflow for writing scripts. I made two rounds of adjustments based on that post.
The first time, I followed the article's instructions and fed it to Claude Code. It quickly built my own workflow, but honestly, the back-and-forth optimization and tweaking was both exciting and frustrating. At the time, I was on Claude Pro, and it kept asking me to take breaks—I was both hyped and totally drained. That said, I did get familiar with Claude Code really fast during this process. As soon as I learned about Max, I upgraded immediately. I also abandoned that workflow.
The second time, I started from scratch based on my actual needs, following the author's core principles: "systematization over fragmentation" and "data-driven." I talked to Claude Code in my own plain language, building from a basic framework. Initially, I used a material library to generate scripts, but the output was too homogeneous—the library was too small. So I launched a second production path: "remake trending hits." This ensured output while also accumulating material. Once the library grew enough, I started a third path: letting Claude Code generate scripts based on its own usage logs (I thought this was a great idea). But the results were still unsatisfying—the script structure was too rigid. So I added a fourth path: "trending clone," with the core goal of accumulating trending frameworks.
This workflow has been running for about a month now. To be honest, it's fallen far short of my expectations. After self-reflection, the core issue is still my limited time, energy, and focus—I can't fully commit to content creation.
In my view, Douyin is already on the decline, but every bit counts—free traffic is free traffic, so why not take it? I let go of my previous creative assistants and kept only one real human helper to manage my content creation end-to-end. It's helped cut costs, but hasn't boosted efficiency yet.
Finally, thanks again to @dontbesilent for the open-source spirit.