Tencent WorkBuddy Skills System Explained: From Installation to Custom Agents

Tencent WorkBuddy, an AI office agent, has its core capabilities rooted in its Skill system. A tutorial video on Bilibili that took three months to produce breaks down the entire chain—from skill installation and selection to custom creation—in nearly 280 minutes. Based on that tutorial and cross-verified with multiple sources, this article distills a practical, quick-start guide.

What Is WorkBuddy: Not a Chatbot, but a Digital Employee

According to Bilibili creator "AI Office Efficiency Guide," WorkBuddy^Ad is positioned as a "digital employee that works 24/7 and is always on call." It is not a simple conversational AI, but an agent that can plug directly into workflows, automatically decompose tasks, and execute complex processes. Whether it's copywriting, coding, or Excel report processing, it can handle it all.

This view is corroborated by a review from the YouTube channel "TechFlow": the essential difference between WorkBuddy and ordinary chatbots lies in "task execution" rather than "content generation." It is more like a configurable automation assistant, and the skill system is the key to extending its capabilities.

The Three-Column Structure of the Skill Marketplace: Recommended, Skill Hub, and Bundles

Open the "Expert Skill Connector" on the left side of WorkBuddy to enter the skill marketplace. The marketplace is divided into three columns with distinctly different purposes:

• Recommended column: Officially curated premium skills—few in number but guaranteed in quality. Best for newcomers to browse first.
• Skill Hub: An open public skill library with a huge number of entries but uneven quality. When selecting, it's advisable to sort by download count and prioritize skills with more users. According to the creator's demo, among similar coding skills, download counts range from over 4 million to just a few thousand—a massive gap.
• Bundles: Not a single skill but a collection of skills packaged together, suitable for filling gaps in a certain capability area in bulk.

The YouTube channel "AI Tool Watch" also noted in a comparative review that Skill Hub's openness brings richness but also requires users to have some discernment—download count is an effective filtering signal.

The Essence of a Skill: An Assignment Guide for AI

Many people mistakenly believe that installing a skill "adds new functionality" to the AI, but the tutorial emphasizes: a skill is essentially an assignment guide written for the AI to read. The AI already has the ability to process Excel, build web pages, and make PPTs—what it lacks is "following your rules."

A skill consists of three layers: name and description (telling the AI when to use it), version and installation information, and specific operational details (workflow, precautions, output format, boundary conditions). Installing a skill is equivalent to handing the AI all the instructions you'd want to convey in one go—after that, a single sentence can trigger it.

From a technical implementation perspective, skills typically exist as structured files (such as Markdown or JSON with metadata) that follow specifications similar to the "Model Context Protocol" (MCP) or plugin manifests. Such specifications require declaring the skill name, description, trigger conditions, required permissions, and specific execution steps, enabling the agent to extend its capabilities without retraining the model. This differs from traditional software where "installing adds a function": the model's inherent capability boundary hasn't changed—what changes is how it's guided. It's essentially using prompt engineering to solidify domain knowledge, operational norms, and output constraints into reusable templates. Therefore, a skill's quality depends heavily on the writer's clarity of expression: the more precise the description and the clearer the boundaries, the more stable the model's invocation.

Progressive Disclosure: Why a Hundred Skills Won't Cause Lag

Won't too many skills blow up the context? WorkBuddy uses a progressive disclosure mechanism: normally, only each skill's name and description follow the conversation, which is a very small footprint. According to the creator's estimate, the names and descriptions of a hundred skills amount to just a few thousand characters. Only when a trigger word is matched in the conversation does the full content of the corresponding skill get loaded.

This also explains why "whether the description is written accurately" is crucial—if the description is vague, the AI won't know when to invoke which skill, and the skill is as good as not installed.

Progressive Disclosure originates from the field of interaction design, referring to gradually revealing information only when the user needs it, avoiding the cognitive burden of presenting everything at once. In the context of AI agents, this mechanism is used to manage the context window—large language models can only process a limited amount of text at a time, and stuffing every skill's full instructions into the conversation would quickly hit the limit and dilute the model's attention. WorkBuddy's approach: load only lightweight metadata (skill name and description) first, let the model judge whether the current task is relevant, and retrieve the full instructions only upon a match. This "index first, load later" approach is also common in Retrieval-Augmented Generation (RAG) and Function Calling—essentially trading two-stage filtering for context efficiency.

Three Ways to Acquire Skills

1. Pick from the official Recommended column: Quality is filtered, quantity is small, quick to get started—highest priority.
2. Manually search Skill Hub: Many entries but you need to discern for yourself; filtering by download count is safer.
3. Let the AI install for you: In the top-right "Add Skill," select "Find Skills" and describe your needs in natural language—the AI will automatically search, perform a security review, and install. Note: by default it only searches the Recommended column; to search the Skill Hub community repository, you need to install the "Find Skills" skill first.

During installation, the AI performs a security review, scanning whether the skill is trustworthy, whether it accesses the network, and whether it contains dangerous operations (such as file deletion permissions). When encountering sensitive operations, it waits for user confirmation.

The security review before skill installation corresponds to an important risk in the AI agent field—supply chain attacks. Since skills can contain scripts, network requests, and file operation instructions, malicious skills could steal data or execute destructive commands. Common industry protection approaches include: static scanning of dangerous calls in skill packages (such as file deletion or access to sensitive paths), declarative permissions (skills must explicitly state whether they access the network and which directories they read/write), and human confirmation at runtime. WorkBuddy performs a trustworthiness scan during installation and pauses for user authorization when encountering sensitive operations—a practice of "least privilege + explicit consent." When acquiring skills from open repositories like Skill Hub, users should still prioritize sources with high download counts and many reviews to reduce risk.

Two Ways to Invoke: Slash Commands and Natural Language Triggers

Once skills are installed, there are two ways to use them:

• Slash command: Type "/" to bring up the list of installed skills and select directly. Suitable when you know exactly which skill to use.
• Natural language trigger: Include the trigger word from the skill's description in your speech, and the AI automatically loads the corresponding skill. For example, saying "I want to make a PPT"—the system recognizes the "PPT" trigger word and invokes it automatically.

The tutorial demonstrates using a "de-AI-ify" skill to polish an article and using a PPT skill to generate a three-page professional presentation—both reaching usable quality.

Creating and Uploading Custom Skills

Uploading skills: Download a skill package from the internet (usually in ZIP format), drag it in or select the file to upload, and the system automatically detects and installs it. Non-high-risk skills install automatically.

Creating skills: Through the "Skill Creator" skill, describe your needs in natural language to generate one. The tutorial identifies three key elements: what it can do, how it's triggered, and what the output looks like. For example, to create an "Excel to visual dashboard" skill: the trigger is sending an Excel sheet and saying "make a dashboard"; the output requirements are an HTML file that opens offline with a double-click, core metric cards at the top, and three to four interactive charts below.

If the first version isn't satisfactory, you can continue to provide revision feedback—such as adjusting the color scheme, merging trend charts to display by month, or replacing product IDs with product names in the ranking. After several rounds of revision, the skill will increasingly fit your personal habits.

Beginner Pitfall Avoidance: The Official Top Ten Curated Skills

For users just getting started who don't know what to install, the official team recommends ten curated skills covering the most common scenarios: Agent Browser (web clicking, page turning, screenshots), PDF reading/writing, security scanning, audio-to-text transcription, video summarization, web search, Find Skills, and more. The principle is "install what you lack"—if you have no specific needs, just install the first few common ones.

Conclusion

WorkBuddy's skill system isn't complicated. The core logic is: use descriptions to tell the AI when to use a skill, and use guides to tell the AI how to do it. Installing a skill isn't adding functionality—it's setting rules. When selecting, prioritize official recommendations, then look at download counts; when using, invoke via slash commands or natural language triggers; if unsatisfied, create your own and refine it through Skill Creator. Master this approach, and the AI office agent can truly become the digital employee that works for you.

One-line Summary

A practical guide to the WorkBuddy skills system: from selection and installation to custom creation, making AI your digital employee.

Paragraph Summary

Based on a 280-minute WorkBuddy tutorial on Bilibili and cross-referenced with multiple reviews, this article outlines the complete usage path of an AI office agent's skill system. It points out that WorkBuddy is positioned as a "digital employee" that executes tasks rather than a chatbot, with its capability expansion relying on the skill system. The skill marketplace has three columns—official recommendations, the open Skill Hub, and bundles—with newcomers advised to prioritize the recommended column and filter Skill Hub by download count. The article emphasizes that a skill is essentially an assignment guide written for AI, composed of name/description, version information, and operational details; the progressive disclosure mechanism prevents hundreds of skills from blowing up the context. There are three ways to acquire skills, invocation supports slash commands and natural language triggers, users can also create custom skills via Skill Creator using natural language, and the article lists the official top ten curated skills that beginners should install first.