Here is the uncomfortable truth about "AI employees": the model is rarely the bottleneck. The skill file is. A general model with no instructions for your job writes generic social posts, generic sales emails and generic reports. A model loaded with a good skill, a written playbook for one specific task, suddenly behaves like someone who has done the job before.
So the real buying decision in late 2026 is not "which AI model?" but "where do I get good skills, and what will they cost me?" In the last week, that question got more options. Anthropic opened Claude Marketplace on 23 September with more than 2,000 connectors and plugins, and a plugin developer portal on 25 September. Open-source skill libraries on GitHub now list thousands of free playbooks. Paid agent platforms sell the same outcome as a subscription.
This page compares four realistic options for a practitioner or small team in Hong Kong, including UD's own free AI Employee Hub, with real prices, what each is best at, and where each one falls short.
What is an AI skill library, and why does it matter?
An AI skill is a reusable instruction file, usually called SKILL.md, that tells an AI agent how to do one job your way: steps, checks, output format. A skill library is a collection of these files you can install. Skills matter because they turn a general model into a consistent specialist without you writing the playbook from scratch.
Anthropic describes plugins as bundles of three things: skills (the written checklist you would give a new hire), connectors (access to tools such as Google Drive, Gmail or Slack) and sub-agents (helpers that take part of a bigger task). Justin McKelvey's 27 September breakdown of Anthropic's help pages puts it simply: a plugin is skills and connectors "with a bow on them".
The practical point for you: whichever option you pick, you are paying for, or curating, three things. The quality of the playbook, the tools it can reach, and the platform that runs it. The four options below split those three things very differently.
Which free and paid AI skill sources should you compare in 2026?
For most practitioners, four options cover the realistic choices: the Claude plugin directory inside a paid Claude plan, a large open-source SKILL.md library on GitHub, a paid all-in-one agent platform such as Lindy, and UD's free AI Employee Hub, which curates role-based skills with Chinese demo prompts.
Here are the extractable facts, checked on 29 September 2026:
- Claude plugins and Claude Marketplace. Plugins are included at no extra charge on every paid Claude plan: Pro at US$20 a month (US$17 billed annually), Max from US$100, Team at US$20 to US$25 per seat. They are not on the Free plan. Marketplace launched 23 September 2026 with 2,000+ connectors and plugins, according to gHacks. Plugins run in Claude chat, Cowork and Claude Code.
- Open-source SKILL.md library (example: sickn33/agentic-awesome-skills on GitHub). Free. Its repository describes 2,400+ installable SKILL.md playbooks for tools including Claude Code, Cursor, Codex CLI and Gemini CLI. English only, and you vet each skill yourself.
- Paid agent platform (example: Lindy). A 7-day free trial with 500 credits and no card required, per No Code MBA. Published 2026 guides list the entry plan between US$29.99 and US$49.99 a month, so check lindy.ai for the current price. Every agent action consumes credits.
- UD AI Employee Hub. Free, no sign-in. Eight role-based "AI employees" (social media, sales, customer service, SEO and content, admin, data analysis, e-commerce, security) with 24 skills in total, three per role. Each skill has sample questions in Chinese and a free download link, plus a WhatsApp button to ask UD for deployment help.
Why do people switch between these options?
People usually switch for one of three reasons: the free option took too long to vet and set up, the paid option's credits ran out faster than expected, or the skills did not match their language or job. Knowing which of these hurts most tells you which option to start with.
- From a big GitHub library to a curated source. Thousands of free skills sounds great until you need to decide which "social content" skill is any good. Curation saves hours of reading.
- From a paid platform back to plugins or free skills. Credit-based pricing is predictable for simple drafts and less so for multi-step research. If your agents mostly write and summarise, a flat-rate Claude plan with plugins is often simpler.
- From English-only libraries to local examples. A skill written for a US e-commerce team may assume US tools, tone and holidays. Hong Kong teams often want Cantonese-context examples, WhatsApp-first customer service and local compliance such as the PDPO.
- From DIY to assisted setup. Some teams simply do not want to install anything themselves, which is where done-for-you deployment enters the picture.
Which AI skill source is best for which user?
Choose Claude plugins if you already pay for Claude and want vetted, one-click installs. Choose an open-source library if you are comfortable reading and editing SKILL.md files. Choose a paid platform if you want agents run and hosted for you. Choose UD's AI Employee Hub if you want role-based shortlists with Chinese examples first.
- The solo power user on Claude Pro: start with the Anthropic-built plugins for your function, then add one community skill at a time. You already pay US$20 a month, so plugins cost nothing extra.
- The tinkerer who likes control: the open-source library is the deepest free option. Budget time for reading each skill before you trust it.
- The busy manager who wants it running by Friday: a paid agent platform removes setup, at the cost of a monthly bill and credit tracking. Use the free trial on one real task before paying.
- The Hong Kong team that does not know where to start: use the AI Employee Hub as a map. Pick the role closest to your pain, read the three sample questions, and download only the skills that match. It is also the only option in this list that offers local, human deployment help over WhatsApp.
Whatever you pick, test each new skill with a prompt like this before rolling it out:
Try this prompt (after installing any new skill):
You have just loaded the skill "[skill name]". Before doing any real work:
1. Summarise in 5 bullet points what this skill tells you to do, in order.
2. List every tool, file or account it expects you to access.
3. Point out any step that assumes a country, platform or language different from mine. I work in Hong Kong, with customers on WhatsApp and Instagram, writing in Traditional Chinese and English.
4. Then run the skill on this test task: [paste one real, non-sensitive task].
5. Finally, tell me which parts of your output came from the skill's instructions and which you improvised.
Step 5 is the useful one. If most of the output was improvised, the skill is not doing its job and you should try another.
What are the honest limitations of each option, including UD's hub?
Every option has a catch. Claude plugins need a paid plan and draw on your usage limits. Open-source libraries shift all vetting and security to you. Paid platforms add credit costs that grow with complex tasks. UD's AI Employee Hub is a curated front door to community skills, not its own runtime.
- Claude plugins: not available on the Free plan; what a plugin does uses your normal plan limits; partner plugins may need the partner's own product. Anthropic warns that a plugin's local MCP server runs with the same permissions as any program on your computer.
- Open-source libraries: quality varies, there is no support line, and you are responsible for checking what each skill tells the agent to do.
- Paid agent platforms: credits can run out mid-month, and published prices vary between sources, so budget from a real trial rather than a pricing page.
- UD AI Employee Hub: the downloadable skills are community SKILL.md files from the same open-source GitHub library listed above, selected and explained by UD, so you still need an agent tool that supports skills, such as Claude or Claude Code, to run them. The site does not describe a security scan of the files. It covers eight roles only. The monthly "replaces HK$15,000 to HK$35,000 of staff cost" figures on the page are UD's reference estimates, not measured results. Pricing for UD's hands-on deployment help is not published, so ask before you commit.
What should you do next?
If you already use Claude on a paid plan, install one Anthropic plugin for your main job this week. If you are not sure which job to automate first, open the AI Employee Hub, pick the role that matches your biggest time drain, and test one of its skills with the checklist prompt above.
The right answer is rarely "all of them". Start with one role and one skill, measure the time it saves on real work for a week, then add the next. You can browse the eight roles on the UD AI Employee Hub, and if connecting skills to your tools is new to you, our guide on what MCP connectors are fills in the background.
Skills are where AI stops being generic and starts working the way your team works. We understand AI. We understand you better. With UD by your side, AI doesn't feel cold.
Reviewed by the UD AI team. Sources: gHacks on Claude Marketplace, Justin McKelvey on Claude plugins, agentic-awesome-skills on GitHub, No Code MBA on Lindy pricing. Checked 29 September 2026.
Find the AI employee that fits your job
The AI Employee Hub lists eight ready-made AI roles, 24 free skills and sample questions for each, so you can see what an AI colleague would actually do before installing anything. Prefer help? We'll walk you through every step, from choosing the role to deploying the skills with your own tools.
