2.3 Creating a Skill: From One Success to a Reusable Method
Adding and using Skills was covered in 1.5. This section focuses on creating your own.
Using Someone Else's Skill Is the Starting Point
Say you have a sales table and want the agent to analyse trends and anomalies. Choose and add a suitable data-analysis Skill in 【Explore】, select it in the conversation, then enter:
AOS Work executes using that Skill's built-in methodology — for example: "profile the data → compute key metric distributions → flag anomalies with cause hypotheses → output a structured analysis report".
You get a structured report: a data overview, a list of anomalous SKUs, trend charts, and recommended actions. You never wrote a single analysis step — the steps came with the Skill. You supplied the data and reviewed the output.
That is what using someone else's Skill gives you: a low starting point and a fast payoff.
From Using to Writing: Where the Real Value Is
But someone else's Skill solves general problems. The way your company specifically works — field definitions, approval habits, brand voice, industry vocabulary — no off-the-shelf Skill will write that for you.
The real value comes from writing your own.
Ask the agent to generate a SKILL.md draft. This lesson uses WorkSkill Maker as an example creation tool: if it is available in your 【Explore】, add it and select it in the conversation. Otherwise, ask the agent directly for a draft using the structure below.
For adding and selecting a resource, see AOS Work: Skill. After generating the file, open SKILL.md using Resource Preview.
Five Steps to Your First Skill
Step 1: Open a Conversation
Click 【New Chat】 and request a SKILL.md draft. If using WorkSkill Maker, first select the resource you have added:
Ask the agent to list any missing requirements before generating the file.
Step 2: Describe It in Plain Language
Give the agent the following information: what problem does this Skill solve? What is the input? What is the output? Which rules must always hold?
You answer the way you would talk to a colleague. It formats everything into a SKILL.md.
Step 3: Review the Generated SKILL.md
Once it is generated, check these parts one by one:
| What to check | What to look for |
|---|---|
| name | Short, clear, and indicative of purpose |
| description | Purpose and trigger conditions — state suitable tasks and include expressions users would use |
| Steps | Is the execution flow complete? Did it leave out your experience? |
| Constraints and prohibitions | Did it capture the pitfalls — for example "do not modify the original file" or "confirm before executing"? |
| Output format | Are the structure and naming of the deliverable explicit? |
| Reference material | Are the templates and glossaries it needs attached? |
Review the steps, scope, output format, and references, and fill in the method for this type of task.
Step 4: Understand the Activation Mechanism
The description field states the Skill's purpose and suitable tasks. If you are creating a product-image renaming Skill, the description should read:
Activate when the user mentions image renaming, file naming conventions, or product image renaming.
This helps the agent judge whether the Skill suits "rename these product images". For testing, explicitly select the Skill in the conversation and check the result.
If the Skill is not used as expected, check that it is added, enabled, and selected in the current conversation, then check the description.
Step 5: Test and Iterate
Test with a set of practice images and check the names. Revise SKILL.md when needed. Check whether the current resource supports editing; if it does not, keep the revised file and ask its maintainer to update the resource before testing again.
For the iteration process, see 2.4.
Exercise: Draft and Compare Two Rule Versions
Create two drafts of a product-image naming Skill in the demo workspace named Product Team:
- Specify filenames built from category, product code, color code, and date (
category_SKU_color_date), and generateSKILL-v1.md. - Use the product fields for
front.jpgto list its original and proposed filenames. - Change the color-code prefix from
DtoC, retain the original filename column in the output, and save the revised rules asSKILL-v2.md. - Apply both sets of rules to the same product information and save their filename comparison as
Comparison.md.
A Good Description Versus a Bad One
For the same Skill, how you write the description decides whether it is usable:
| Approach | Description | Result |
|---|---|---|
| Too narrow | "Rename files" | Does not describe the product-image scenario or naming convention |
| Too broad | "Handle image-related tasks" | Does not distinguish renaming from other image tasks |
| Right | "Bulk-rename product images following the category_SKU_color_date convention. Activate when the user mentions image renaming, file naming conventions, or product image renaming." | The scenario is specific and the user's likely keywords are listed |
A practical formula for the description: what it does + which convention it follows + how you will refer to it.
A First Skill Usually Needs Iteration
A first version rarely lands in one pass. In real use you will usually find:
- Some steps get misread — they need rewording
- Some experience never made it in — it skips a step
- Constraints are missing — "do not modify the original file", "confirm with me before executing"
- The description is too narrow and misses scenarios, or too broad and fires when it should not
All of that is normal. The value of a Skill is continuous improvement. You can give feedback directly in the conversation:
1. keep the original file name as a note when renaming
2. use the C prefix for colour codes, not D.
Or more directly:
Ask the agent for a revised SKILL.md, review the changes, and update the resource you actually use. Test again with the same practice images to confirm that the new rules take effect.
Summary
Use someone else's Skill for general problems; write your own for your company's problems.
Five steps get you the first version. The value lies in every iteration after that.
The Skill thread has one link left: add and use in 1.5, look inside in 2.2, build your own in 2.3 — the next lesson covers tuning it when the standard version is not enough.
2.2 Understanding a Skill: Structure, Progressive Disclosure, and Reuse Value
Lesson 1.5 covered what a Skill is and how to install one. This lesson looks inside — how it is organised and why.
2.4 Tuning Skills: When Standard Isn't Enough
Before you tune a Skill, look at its current files first — see "How to Look Inside a Skill" in 2.2.