Building a Working Agent, Starting from the Implementation Logic
Building a Working Agent, Starting from the Implementation Logic Background: these are my notes on Deep Understanding of AI Agents: Design Principles and Engineering Practice by Li Bojie (Apache 2.0 open source, GitHub: bojieli/ai-agent-book). The book has ten chapters, from theory to engineering practice, with 88 runnable experiments. Rather than summarizing chapter by chapter, this post reorganizes the material around “what problems must you solve to build an Agent”, and for each modul...
Flint: Turning Skills into Personal Assets
Lately, every time OpenAI ships a new model, a familiar argument makes the rounds: skills are useless, and more and more users are abandoning them. Is that true? Yes and no. To be fair, the logic has merit. As models get stronger, plenty of skills really have been abandoned — mainly the so-called “general-purpose” ones like Superpower. Those skills existed to patch over gaps in model capability, and they did help back when no model was aggressively optimized for coding. As the models themselv...
From Token Farming to Skill Management: My pks Tool in Practice
I wrote an article before: https://lichuanyang.top/en/posts/26060/ . It taught people how to build Agent-independent workflows, so you can spin up different Agents anytime and smoothly use the free/trial plans from various providers. After practicing this way, one thing that obviously becomes tedious is skill management. So I wrote another tool to manage skills across different Agents and projects. Given the current Agent ecosystem is still quite chaotic, everyone does their own thing,...
Bringing Notes, WeChat Reading, and Zhihu into Obsidian: My LLM-Wiki Knowledge Hub
A while back, I came across Andrej Karpathy’s llm-wiki concept and felt an instant sense of resonance. I’ve always enjoyed writing things down, but the problem was that everything ended up scattered across different places, and I never had the energy to manage it properly. When I discovered llm-wiki, I realized — all that stuff I’d been writing over the years was finally going to pay off. Why a Knowledge HubFirst, what is llm-wiki? llm-wiki is a concept recently proposed by Andrej Karpathy:...
Free AI Video Generator: How I Built a Zero-Cost Tool for Narrated Multi-Scene Videos
“The solution is not to suppress AI, but to make it a more equitable capability, so that everyone knows how to create more with AI. This is a very important vision for our company — to make world-class AI belong to everyone.” This is something Bruce Yang, the founder of Agnes AI, said in an interview. Many Chinese AI companies today — DeepSeek, Zhipu, and others — are driving down the price of AI. To be fair, the cost of text and code processing has already been pushed remarkably low. But v...
Agnes Free Models: I Rewrote ViMax to Generate AI Videos for $0
A while back, I wrote about “token hunting” — the practice of bouncing between free AI models instead of paying for subscriptions. At the time, free models mostly meant text and maybe some image generation. Then Agnes AI quietly dropped something unexpected: free video generation models. Not text. Not images. Video. Three models — agnes-video-v2.0, agnes-image-2.1-flash, and agnes-2.0-flash — all free, no credit card, no GPU, just an API key. Among free agnes video generation solutions, Agnes...
How to Farm Free Tokens (Part 2): Building an Agent-Independent Skills Management Workflow
Last post ended with this: Agents are just tools. Documentation is the core. After that went up, someone asked: sure, but how do you actually manage all those docs? You can’t just copy-paste them into every project, can you? Nope, you can’t. So I built a tool to solve exactly that — pks (Personal Skills Manager). 408 lines of pure bash, zero dependencies. It does one thing: packages your AI workflow docs as skills, installs them into projects on demand, and locks you into no Agent platform. ...
How to Farm Free Tokens: Building Agent-Independent AI Workflows
What’s the biggest—or perhaps only—pain point of using AI right now? Probably the bill. Claude, Codex, Cursor, Qoder, MiMo, MiniMax—their basic plans all start around $20/month, and you still have to watch your usage. If you want unlimited access, it gets even more expensive. Meanwhile, lots of platforms offer free trials or permanently free models. For instance, OpenCode’s DeepSeek Flash has been free for a while. So how do we actually make the most of these free models? I’ve been using...
Automating Hexo Theme Migration with AI Agent: From Next to Butterfly
The Problem: A Tedious Task No One Wants to DoRecently, I decided to migrate my Hexo blog from the Next theme to Butterfly. Sounds simple, right? If you’ve ever done a Hexo theme migration, you know it’s anything but simple: Config files are 1000+ lines long, requiring line-by-line comparison between old and new theme formats Feature mapping is complex: Next’s leancloud_visitors maps to Butterfly’s busuanzi, Next’s reading_progress maps to Butterfly’s preloader Two sites need syncing: Both C...
How I Recovered My Blog After Vercel Banned 163 Email
A few days ago, I wrote a new article and deployed it to Vercel as usual, but the pipeline threw an error. At first I thought it was a build issue, and after messing around for a while I realized something was wrong — it was the Vercel account itself. The account registered with a 163 email couldn’t log in at all. Later I found out that Vercel had banned the entire 163.com email root domain. What is Vercel?If you’re into independent blogging, you’ve probably heard of or are currently using ...
Recent Posts