The AI tool I most want to recommend to everyone in 2026 is Claude Skill.
It's truly amazing! For example:
Interpret a paper in one sentence, generate AI illustrations, and post to a WeChat Official Account.
Generate a knowledge canvas in one sentence
Generate Xiaohongshu copy, covers, and illustrations in one sentence
However, this workflow has obstacles for many friends.
Claude accounts are frequently banned, and finding reliable third-party proxies isn't easy.
Even if you can use it, it costs hundreds of dollars a month.
Is there a free solution?!
After research, I found a true "living Buddha" in the tech world.
No account registration required, no credit card binding, and no messy restrictions.
The most critical point is: it's free.
Ready to use out of the box and extremely friendly to beginners, no wonder it's exploding in popularity overseas.
You can see the sentiment from its name: OpenCode.
How to Install OpenCode
Official website:
Supports Mac OS, Windows, and Linux.
It has both a command-line interface (CLI) and a desktop client.
I recommend using the CLI; the client is said to have quite a few bugs.
Taking Mac as an example, open your terminal (or Warp), enter the following command, and press Enter:
curl -fsSL
https://opencode.ai/install | bash
Once installed, type opencode and press Enter to see the interface.
If you've used Claude Code, you might find it familiar.
If not, don't panic. You can type /model to see a large number of free models, such as GLM-4.7.
If you have a subscription to GPT or Gemini, you can type /connect.
Select OpenAI or Google, open your browser to log in and authorize, and you can use Codex and Gemini models.
They provide over 75 model providers in total.
This includes well-known overseas models and API proxies like Mistral, OpenRouter, Groq, etc.
You can even add models installed locally via Ollama, which is very comprehensive.
I'll use GLM-4.7 for the demonstration. Besides being free, there are two other reasons:
1. Excellent code generation capabilities
On SWE-Bench, which solves real GitHub issues in open-source Python projects, GLM scored 73.8%, putting it in the top tier.
On HumanEval for Python programming problems, GLM-4.7 scored 94.2%, which is still excellent.
2. Tool calling and instruction following capabilities
IFEval mainly tests instruction following, with a score of 88%.
GSM8k tests multi-step reasoning, with a score of 98%.
It feels like GLM-4.7 has been heavily optimized for AI programming scenarios, greatly enhancing Agentic Coding capabilities.
Especially in code generation and instruction following, which are crucial for creating and calling Skills.
Additionally, friends often praise GLM-4.7's frontend aesthetic, which is a nice bonus.
OpenCode provides this model for free; just use it as much as you want.
Install the Strongest Agent Framework: Oh-My-Opencode
OpenCode is not only open-source and provides free models, but its good reputation also comes from its active developer ecosystem.
The most highly regarded is Oh-My-Opencode (referred to as OMO).
With this installed, the OpenCode programming experience really takes off.
It's said to be an AI Agent programming framework built by the author after spending $24,000 on tokens.
How to install?
You just need to tell OpenCode: "Help me install the oh-my-opencode plugin."
GLM-4.7 will reason, call search tools, and find the installation instructions.
OMO is an Agent framework designed to better integrate and orchestrate models.
It will ask if you have Claude, ChatGPT, or Gemini subscriptions for future integration.
You can answer which ones you have, or continue the installation if you have none.
It installs very quickly.
You can add subscriptions later by typing opencode auth login.
Why must you install the OMO plugin?
First, it has built-in Agents that follow AI programming best practices.
"Sisyphus" is its master Agent, usually played by a high-IQ model (like Opus 4.5 High).
It then directs its sub-agents to work, such as:
- Oracle: Architect and debugging master (GPT 5.2 Medium)
- Frontend UI/UX Engineer: Frontend and design expert (Gemini 3 Pro)
- Librarian: Browses docs, checks open-source implementations, and explores codebases (Claude Sonnet 4.5)
- Explore: High-speed codebase scanning (Grok Code)
Second, it has built-in common MCPs, such as:
- Exa (Web search)
- Context7 (Official documentation search)
- Grep.app (GitHub code search)
Finally, and most importantly, it achieves compatibility with Claude Code.
It supports the same technical solutions like Commands, Agents, Skills, MCPs, and Hooks.
This allows us to easily play with Skills in OpenCode.
There are many more cool things and effects; it's incredible!
Installing and Using Skills
Installing a Skill is very simple.
For example, the Skills written by the Obsidian CEO have been very popular lately.
You just need to open OpenCode and enter:
"Install the skills from here: https://github.com/kepano/obsidian-skills"
The GLM-4.7 model will download and install them automatically.
Usage is also simple. Just say: "Draw a canvas to interpret 'The Pyramid Principle'."
Because it mentions a canvas, the model will reason that it needs to use the json-canvas Skill.
It generates an Obsidian-format canvas file. Here is the result when opened:
You can also type a forward slash "/" to find installed Skills and call them precisely.
Creating Skills
I suggest installing Anthropic's official meta-Skill first.
That is, skill-creator.
https://github.com/anthropics/skills/tree/main/skills/skill-creator
Use this Skill to generate other Skills, haha!
The simplest Skill can just be a prompt.
For example, I'll create a frontend beautification Skill:
Call skill-creator to create a Claude skill for beautifying frontend design using the following prompt: [Prompt Content]
Prompt content:
https://xiangyangqiaomu.feishu.cn/wiki/K9bMweixHijuf4kOji7cbQVknhZ
In a moment, GLM-4.7 has written a Skill named Designer.
Note: You need to exit and restart OpenCode to find newly installed or written Skills.
(Exit by pressing Ctrl + C)
Let's do a test. First, write a single-page Todolist.
Call the Skill, and it gives 3 design options.
Here is the result of choosing Option B:
See, isn't designing a Skill simple?
Skills can design and package your workflows, prompts, etc.
Complex Skills might include scripts, MCPs, API calls, reference docs, etc.
They are perfect for reusing experience and automating processes.
Imagine the leverage if you turned the workflows of your team's top members into Skills!
Sharing Skills
A written Skill is just a collection of files and folders.
You can share it by zipping it and sending it to others.
If you're lazy, you can tell OpenCode: "Zip and download the [Name] Skill."
It will automatically compress it and save it to your computer's Downloads folder.
You can also publish it to GitHub. Just tell OpenCode:
"Publish the [Name] skill to GitHub using the GitHub CLI."
(Assuming you've configured the GitHub CLI, which isn't complicated.)
Wait a moment, and you'll have a GitHub URL.
My general design beautification Skill:
Recommended Skills
Anthropic Official Skills:
Awesome Skills list provided by ComposioHQ:
https://github.com/ComposioHQ/awesome-claude-skills/tree/master
However, those are a bit generic. Here are a few I've used and liked.
Superpowers
A selection of Skills with 16k stars, covering everything from brainstorming and PRDs to development and testing. Very good reputation.
Planning-with-files
Written by author Ahmad, inspired by Manus's Agent methods after Manus was acquired for $2 billion. Great for multi-step tasks and guiding other Skills. AI also needs to be organized; it writes TodoLists and checks them off to avoid forgetting or going off track.
X-article-publisher-skill
An X article publishing Skill written by Professor Wang Shuyi. It calls the Chrome browser to automatically paste text and images into the editor and can even position images based on the text.
This Skill is worth studying to adapt into various auto-publishing tools.
NotebookLM skill
Also controls Chrome to automatically upload PDFs and YouTube links to NotebookLM and programmatically ask questions. Very helpful for learning. Most impressively, it can remember login info.
Final Thoughts
The combination of OpenCode + OMO + Skills is truly powerful.
You can get started in less than half an hour from installation to use.
The free built-in models are mostly sufficient.
Most importantly, it lowers the barrier to AI programming to an unprecedented level.
You don't need to understand complex configurations or mess with environments.
Install a plugin in one sentence, call a Skill in one sentence, and complete a task in one sentence.
Now, the competition among AI tools is no longer about whose model is stronger, but who can make these capabilities easier for ordinary people to use.
OpenCode has achieved this.
It might not be the strongest, but it might be the friendliest.
If you've been hesitating at the door of AI programming due to accounts, costs, or technical barriers:
The door is now open for you.
Come in and try it; you'll find the AI world is more exciting than you imagined.
![[Нотатка] Керівники звільняють підлеглих, які не справляються з роботою](/cdn-cgi/image/width=1920,quality=90,format=auto,metadata=none/https%3A%2F%2Fcms-assets.youmind.com%2Fmedia%2F1784827522698_408j7z_HN3Kb76awAAvvjF.jpg)
