I Tested 15 WorkBuddy Skills. These Are the Ones I Kept.

A few people in my group chat kept asking for a WorkBuddy Skill list. WorkBuddy with no Skills installed and WorkBuddy with the right Skills installed are two different tools. Without Skills, it is a regular Saiyan. With them, it is Super Saiyan III.

I went through the Skill Center, tested a batch of popular Skills, and kept 15. The first 13 are general-purpose: research, writing, office work, search, design, reporting. Slot 14 is for students, teachers, and researchers, and it contains two ArXiv Skills. Slot 15 is for complex tasks and development work. The daily free 100 credits plus the free Hunyuan models are enough for normal use.

How to install and call a Skill

In WorkBuddy:

1. Open the left sidebar.

2. Click Experts · Skills · Connectors.

3. Choose the Skills tab at the top.

4. Pick a source: Recommended, SkillHub, or Bundles.

5. Search by name in the search box if you already know what you want.

6. Click the + button on the right side of the Skill card.

To use a Skill in chat, type / and the skill selector opens. Pick the Skill, then write your task.

That is the setup. The rest of this article is the list.

1. Agent Browser: outsource the browser chores

If I had to pick one Skill for every WorkBuddy user, it would be Agent Browser.

Agents can already drive browsers through playwright, but playwright struggles with high-concurrency jobs and long automation flows. Agent Browser ships with stealth, fingerprint control, proxy support, and CAPTCHA handling. It behaves more like a person reading, clicking, scrolling, and filling out a page. That makes it useful for forms, batch data collection, competitor research, ranking tables, and multi-page information gathering.

Use it when the job involves ten tabs, repeated copy-paste, and a spreadsheet at the end.

Prompt example:

Use /agent-browser to open the 10 competitor websites and product pages I provide. For each company, collect product positioning, core features, current pricing, plan differences, last update date, and main selling points. Follow secondary pages when needed and verify the details there. Keep the page source for every data point. Export everything to an Excel file I can keep analyzing. Do not rely on search snippets alone.

2. Deep Research Pro: a multi-source research assistant

Deep Research Pro takes a large question, breaks it into smaller questions, searches across multiple sources in several rounds, and assembles a cited report.

I use it for academic evaluation, industry checks, business analysis, technology selection, and policy research. These tasks usually involve messy sources and mixed reliability. Deep Research Pro can return a report with citations, but you need to set the rules before it starts.

Tell it to provide source material, cross-check independent sources, and write “not confirmed from public sources” when it cannot verify something. That one instruction cuts down a lot of invented detail.

Prompt example:

Use /deep-research-pro to research the current state of XXXX products. Cover major vendors, product forms, pricing models, typical users, deployment scenarios, and limitations. Cross-check multiple independent sources. Put the source and retrieval date next to important numbers. If a piece of information cannot be confirmed from public material, write “not confirmed from public sources.” End with a short set of conclusions.

3. Skill Creator: turn your workflow into a reusable Skill

A Skill is a record of experience. If you want WorkBuddy to work more like you, you have to give it your methods.

Skill Creator takes a conversation, a work session, or a repeated process and turns it into a Skill that the agent can follow later. Suppose every week you pull progress notes from coworkers, clean them up, summarize them, and build a short PPT. Run that process twice with WorkBuddy, then use Skill Creator to freeze the successful steps into a Skill. Next week, call the Skill and skip the setup.

Remove the trial-and-error parts. Keep the steps that worked. Define the input, the output, and the points where the agent must stop and ask you.

Prompt example:

Use /skill-creator to turn our conversation into a Skill. Remove the trial-and-error steps. Keep only the successful workflow. Define the input and output formats. All data must come from the source files I provide. Do not invent anything. If the workflow reaches a point where I need to make a judgment, stop and ask me.

4. Agent Reach: search communities, not just homepages

Agent Reach connects to major social platforms and communities. It helps you find, retrieve, and analyze trends and comments that never appear on a company’s official homepage.

If you need global market signals, product feedback, or community sentiment, this Skill is more useful than a standard web search. Official pages tell you what a company wants you to see. Community threads tell you what users ran into after they paid.

It works well with Agent Browser and Deep Research Pro. Use Agent Reach to find the raw discussion, then use the other Skills to verify and organize it.

Prompt example:

Use /Agent-Reach to search major communities for comments about XXXX from the last 30 days. Group the comments by topic. Separate complaints, praise, workarounds, and feature requests. Give me the evidence behind your trend judgment.

5. Office Four-Piece Set: PDF, Word, Excel, and PPT under AI control

The iOfficeCLI set covers PDF, DOCX, XLSX, and PPTX. It includes an HTML rendering engine that keeps document layout close to the original.

Once installed, WorkBuddy can read a PDF, clean up an Excel sheet, generate a Word document, and output a PPT from the same source material. If you use a computer for office work, this is the first batch to install.

If you prefer native WPS, the WPS Office bundle is an alternative.

Prompt example:

Read the attached XXXX file. Extract all content. Numbers must follow the original file. If two figures conflict, list both and ask me to evaluate them. Then generate a summary Word document, an Excel sheet for data analysis, and an editable PPT with X pages. The three files must use the same numbers.

6. Smart Charts: dashboards instead of Excel screenshots

Smart Charts turns databases, Excel files, JSON, and other data files into interactive charts. It can export charts as images, but the better use is a small dashboard.

Build an HTML dashboard and send the file to the person who keeps asking for updates. They can filter by time and region without asking you to re-cut the spreadsheet.

Prompt example:

Use /smart-charts to analyze this XXXX detail file and generate an interactive HTML dashboard. Put XXXX at the top, XXXX in the middle, and XXXX at the bottom. Add filters for time and region.

7. Guizang PPT: presentation-style HTML slides

Guizang PPT produces presentation-style slides with strong visuals. It also builds a teleprompter page into the file, so you can put your speaking notes in the same document and use it for rehearsal.

The output is a single HTML file that runs full-screen in a browser. Use it when the deck needs to look sharp on screen and you want to practice the delivery.

Prompt example:

Use /guizang-ppt-skill to turn this document into a PPT. Build a strong visual first page. Use timelines, comparisons, large numbers, and flow diagrams in the middle to reduce reading load. Keep the overall style in a high-end tech-business direction. Generate a single HTML file that can run full-screen in a browser.

8. PPT Generator: an editable PPTX

Guizang PPT gives you a polished HTML deck. PPT Generator gives you a PPTX you can open and edit.

That difference matters. If the file goes to a colleague who needs to change a chart, move a slide, or add a logo, editable PPTX saves a round trip. The Skill also asks you to rebuild the reporting logic before it paginates, so the deck does not simply copy the chapter order of a Word document.

Prompt example:

Use /ppt-generator-skill to generate an editable PPTX from this document. Rebuild the reporting logic first. Do not split pages by Word chapter order. Use charts on data pages. Put the main conclusion in one large number and one sentence. Keep one core judgment per page. All chart data must match the source Excel file.

9. Summarize: a filter for information overload

Summarize looks ordinary. I use it more than most of the flashier Skills.

A single topic can produce a pile of PPTs, documents, and tutorials. You do not have time to read all of them, and most of them do not deserve a close read. Summarize acts as a filter. It produces a one-page summary, keeps the sections you name, and lists conflicts instead of smoothing them over.

Prompt example:

Use /summarize on the files I provide. Generate a one-page summary. Keep the XXXX items. If different materials disagree on the same point, do not decide for me. Put the conflicting statements side by side.

10. Humanizer-ZH: remove the AI tone from an existing draft

Humanizer-ZH is for drafts you already wrote. It is not a from-zero article generator.

It cleans up passages that read too even, too polished, or too empty. It removes common AI phrases and breaks up sentence patterns that sound like a template. Use it on a draft that already has your experience, your data, and your judgment. Do not use it to manufacture a voice you do not have.

Prompt example:

Use /humanizer-zh to revise this article. Keep the real experience, the data, and my personal judgments. Do not add new content. Delete common AI phrases such as “it is worth noting,” “crucial,” “not only… but also,” and “marks.” Break up overly even sentence structures. Make the tone read like someone who has used the product for a long time and is sharing notes. Keep the complaints and the uncertainty.

11. Canvas Design: small design jobs without a designer

Canvas Design handles posters, magazine covers, event graphics, and illustrations. It works on composition, visual hierarchy, white space, and style. If you run a small shop, an e-commerce store, a social account, or a one-person operation, this Skill covers the small design jobs that do not justify a full design workflow.

Prompt example:

Use /canvas-design to design an image from the article I provide. Use a 2.35:1 horizontal format. Use a dark cinematic light-and-shadow background. Keep numbers, text, icons, and tags to a minimum.

12. Work Report Pro: you do the work, then you report it

Work Report Pro turns daily notes, chat logs, progress files, and data into a daily, weekly, or monthly report. It reorganizes the material into completed items, key results, open issues, risks, and next steps.

It forces raw material into a reporting structure; the writing is secondary. You still check it. Over time, the Skill learns your reporting pattern.

Prompt example:

Use /work-report-pro to build a weekly report from the files I organized this week. Organize it into completed items, key results, unresolved issues, risks, and next week’s plan. All numbers must come from the source files. Work in progress cannot be written as completed. Results that cannot be confirmed must be marked “to be confirmed.” Put the points my manager cares about at the top.

13. Find Skills: describe the task, get the Skill

Find Skills searches the SkillHub library and recommends Skills for the task in front of you.

Use it when you know what you want to do but do not know which Skill to install. It is infrastructure, like Skill Creator. Instead of saving a dozen recommendation articles, describe the actual job and let it build a short list.

Prompt example:

My task is XXXX. Use /find-skills to find and recommend the most suitable Skills. Put them in a table. List the strengths, weaknesses, and the reason each one fits this task.

14. ArXiv Paper Tracking and ArXiv Close Reading: two Skills for research work

This slot contains two Skills.

ArXiv Paper Tracking watches a research direction and collects recent papers. It organizes them by publication date, institution, author, research question, core method, and experiment target. It also flags repeated new directions and picks papers for a close read.

Prompt example:

Use /ArXiv paper tracking to search for new papers in the XXXX direction from the last 60 days. Organize them by publication date, institution, author, research question, core method, and experiment target. Flag directions that appear repeatedly. Pick the 5 papers most worth a close read and explain why. Do not sort only by citation count or title keywords.

When I tested it on a broad topic in computational media art, it asked three clarifying questions before it settled on a direction. After I answered, it narrowed the scope to ink aesthetics, data aesthetics, and human-machine co-creation. It also listed how much theory and how many papers existed for each direction, the risk points, and a final recommendation.

ArXiv Close Reading acts as a teaching assistant for a single paper. It builds the paper’s structure first, then points to the sections you should read in the original.

Prompt example:

Use /ArXiv close reading to analyze this paper. Break it down by research question, gaps in existing methods, core innovation, model structure, dataset, experiment design, main results, and limitations. Then take the perspective of a graduate student in the same field. Identify the author’s three key assumptions. Check whether the experiments actually support the conclusions. List the questions the paper leaves open. Tell me the three sections in the original paper most worth reading closely, and why.

15. SuperPowers: run complex tasks with an engineering workflow

SuperPowers is the Skill I recommend when a task needs more structure than a single answer.

It works through Plan, test, audit, and optimize. That workflow costs more tokens, but it reduces hallucination on multi-step work. Use it for development tasks, complex research pipelines, and jobs where each stage needs a check before the next one starts.

Prompt example:

Use /SuperPowers to complete this task. Start with a plan. Define the tests or checks for each stage. Audit the output before moving on. Optimize only after the checks pass. Stop and ask me when a decision affects scope, data, or final delivery.

How I spend the daily 100 credits

The daily login claim is the baseline. I claim it every day.

For normal tasks, I use Hunyuan 3 and Hunyuan 4. They are free, and they handle daily study and office work.

For complex tasks, I use GLM5.3 to plan. Then I switch on GLM5.3-Flash for coding and image recognition. I use DeepSeek-V4-Flash for execution work. Kimi K3 handles PPT jobs. With that split, the free credits usually last.

Install in batches

Do not install all 15 at once and then stare at the Skill list.

Start with the Skills that match the work you already do:

• Browser chores: Agent Browser

• Research: Deep Research Pro

• Community signals: Agent Reach

• Office documents: Office Four-Piece Set

• Data charts: Smart Charts

• Presentations: Guizang PPT or PPT Generator

• Long documents: Summarize

• Draft cleanup: Humanizer-ZH

• Small design jobs: Canvas Design

• Weekly reporting: Work Report Pro

• Skill discovery: Find Skills

• Turning your workflow into a reusable Skill: Skill Creator

• Paper tracking and reading: ArXiv Paper Tracking and ArXiv Close Reading

• Complex multi-step tasks: SuperPowers

Install three or five. Run one task with each. Keep the ones that remove a copy-paste loop, a manual check, or a blank-page pause. Add the rest when the work calls for them.