WorkBuddy Practical: 05 - Loading Practical Skills in WorkBuddy

What is a Skill

WorkBuddy itself is responsible for understanding tasks and organizing execution; a Skill is a set of reusable instructions, scripts, reference materials, and resources that tell the Agent how to execute a certain class of tasks, what tools to call, and what formats to deliver.

Anthropic officially launched Agent Skills in October 2025 and released it as an open standard in December 2025.

A standard Skill structure looks like this:

my-skill/
├── SKILL.md
├── scripts/
│   └── check.py
├── references/
│   └── guide.md
└── assets/
    └── template.pptx

Among these, only SKILL.md is mandatory.

---
name: tech-article-writing
description: Used for writing articles about AI products, model evaluations, and the tech industry.
---

Upon receiving a writing task:
1. Confirm the core angle of the article first.
2. Find first-hand materials.
3. Cross-validate core facts.
4. Complete the first draft according to the user's writing style.
5. Check for prohibited sentence structures and AI-flavored expressions.

It can also include:

references/style.md

How Skills Work Programmatically

The most crucial design of a Skill is progressive disclosure, rather than loading everything at once.

Suppose your Agent has 100 Skills installed. It will not load the full content of all 100 Skills into the context window on startup. Doing so would waste tokens and distract the model with irrelevant instructions.

The standard execution flow is divided into three layers:

First Layer: On Agent startup, it reads only the names and descriptions of all Skills. Example:

pptx: Handle PowerPoint creation, editing, and reading tasks
pdf: Handle PDF extraction, merging, editing, and form-filling tasks
tech-article-writing: Write articles for AI and the tech industry

Second Layer: When the user asks:
```
Help me write an official account article about WorkBuddy.
```
The Agent determines from descriptions that tech-article-writing is relevant, and only then loads the full SKILL.md.
Third Layer: During execution, if the Agent needs to mimic your writing style, it reads:
```
references/style.md
```
And if it needs to check for AI flavor, it executes:
```
scripts/check-ai-phrases.py
```

Standard specifications suggest that all Skills load only dozens to hundreds of tokens of metadata on startup, load full instructions after activation, and load other resources and scripts only on demand. OpenAI Codex also uses a similar mechanism, exposing only the name, description, and path to the model first, and reading the full content only when the model decides to use it.

Therefore, Skills solve a long-standing issue in Agent engineering: How to provide the Agent with extensive knowledge and workflows without stuffing everything permanently into the Prompt.

Key Differences Between Skills and Prompts

This is the core question.

Dimension	Prompt	Skill
Core Role	Describe the current task	Define how a class of tasks is performed
Lifecycle	Typically for a single request	Long-term reuse
Trigger Method	Active user input	Agent auto-selection or explicit user call
Carrier	Mainly text	Directory folder structure
Contents	Instructions, context, examples	Instructions, scripts, references, templates, assets
Context Consumption	Directly loaded into context	Loaded progressively on demand
Reuse	Often copied and pasted	Natively reusable
Sharing	Prompt text block	Complete capability package
Execution	Instructions only	Can invoke attached scripts and tools
Model Parameters	Unchanged	Unchanged

The simplest explanation:

Prompt = Task
Skill = Process / Method

Benefits and Applications of Skills

First, supplementing procedural knowledge. Large language models know a lot of general knowledge, but they do not know how your company specifically operates. For example, it knows SQL, but it doesn't know:

Which table has the canonical user_id
The subscriptions table is append-only
Refund queries must exclude a specific status
What the corresponding Grafana dashboard ID is

This type of knowledge is perfect for Skills. Anthropic uses hundreds of internal Skills, categorized into nine scenarios: API and library usage, product validation, data analysis, business process automation, code scaffolding, code review, CI/CD, runbooks, and infrastructure maintenance.

Second, fixing complex workflows. For instance, conducting an industry research task. A general prompt might be:

Detailed research on WorkBuddy

The model would have to think every time:

Where to find materials
What to search first
How to validate
Who to compare with
What structure to output

A Skill can lock down the workflow:

1. Official website
2. Official account and release events
3. Product documentation
4. Actual product tests
5. Competitor comparisons
6. Core insights extraction
7. Fact checking

This capability is called an Encoded Preference Skill. The model can perform each individual step, but the Skill structures them according to how your team works. Another category is a Capability Uplift Skill, which supplements abilities the model inherently struggles with (e.g. complex PDF and PPT handling).

Third, reducing repetitive Prompts. When working with AI, you often repeat rules:

Do not write in an AI tone
Mix long and short sentences
Do not over-use bullet points
Provide your own analysis
Be conservative with technical claims
Do not make up examples

These rules naturally fit into a writing-style Skill. In the future, your prompt only needs to be:

Write a WorkBuddy article

Writing habits, material standards, prohibited phrases, and article flow are all provided by the Skill.

Fourth, turning personal and organizational experiences into assets. General prompts are easily scattered across:

Chat history logs
Docs and Wikis
Notion pages
Individual brains

Since a Skill is structured in files, it supports:

Git version control
Version rollbacks
Team sharing
A/B testing
Automated evaluations
Continuous updates

This is critical for scaling workflows.

Locating Skills on SkillHub Marketplace

Open "Experts, Skills, Connectors" on the left sidebar to search the marketplace or describe your requirements. You can also find suitable Skills on the SkillHub platform. Besides installing from the recommended list, you can import custom skills you downloaded. If you find a skill package zip file online, click "Upload Skill" and load the zip file.

Summoning and Using a Skill in Tasks

For example, after generating an article, if you need to remove the AI tone, install the "De-AI Editor" Skill. During usage, type / in the chat to summon it. Reference the Skill and provide the article. WorkBuddy loads the Skill, executes the rules (such as removing "not... but rather" patterns), and outputs the polished article.

Disabling and Uninstalling Skills

Go to installed skills under "All Skills", click toggle to disable, or click "..." to edit or delete the Skill.

References

Part of this series

WorkBuddy In Action

View all guides in this series →