2.5 Designing a Work: Assemble Skills Into One Business Chain
You installed a few Skills and they work well. But every time you start a task you still have to think it through: "which Skills does this one need?"
Once or twice, fine. Every day, it wears. And no two people on the team choose the same set — the same job runs two different ways.
Work exists for exactly this: a set of Skills packaged for a specific role or a complete piece of work.
What a Work Is: A Combination of Skills
One line to hold on to: a Skill defines one capability; a Work defines how those capabilities combine.
| Dimension | Skill | Work |
|---|---|---|
| Contains | How one class of task is done | A group of related Skills |
| Loading | Enters your capability set by default and activates on demand | Select it in a task and all its enabled Skills load together |
| Problem it solves | "How do I do this kind of thing?" | "Which capabilities does this whole job need?" |
| Best for | Everyday, scattered tasks | One complete business chain |
For scattered everyday work, a Skill is enough. But when a complete job is on the table — a weekly competitor report, or launching a product in three languages — you need a coordinated set. Picking Skills one at a time is far slower than loading a Work.
The rule of thumb
if the job runs three or more steps in sequence, use a Work; if it is a single small thing, use a Skill.
This lesson is about designing your own Work. For the basic concept, how it differs from a Skill, and how to use one inside a task, see 1.6 Work.
How to Use a Work
The three-step operation is in 1.6 Work. One point is worth repeating here: a Work does not call Skills for you — it only makes the capabilities available, and what actually happens is still decided by your instruction.
A recommended pattern: inside one workspace, create a separate conversation per task. Earlier conversations write data into Base; later conversations read from Base to assemble the summary.
That pattern deserves a sentence more: it turns "an analysis" into "a production line". Earlier conversations collect and clean; later ones aggregate and present — and they share the same workspace context.
For adding a Work, viewing its Skills, and selecting it in a task, see AOS Work: Work. For working with table data across tasks, start with the Base query and update exercise.
For continuing from a previous task’s file, see 2.1 Workspaces: report-to-checklist demo. Both conversations select the workspace named Product Team and the Weekly Report Work package. The second reads the first conversation’s report, Weekly Report.md, to prepare an owner checklist.
A Real Example: The Amazon Operations Work
Amazon operations is scattered but connected: monitor competitors, read reviews, optimise listings, watch ads, write reports. Each needs different capability, but together they form one operational chain.
Packaged as an "Amazon Operations Work", it holds six Skills: competitor monitoring, review analysis, listing optimisation, keyword research, ad performance analysis, and weekly reporting. Select the Work and all six load together.
Three Complete Action Chains
The three prompts below are usable as they stand. Note the shared structure: what to check, where to write the output, and where the data comes from.
Sub-scenario A: Daily Competitor and Review Check
1. Did any competitor change price overnight? What is my price gap?
2. Did any competitor change title or bullets? If so, list the changes.
3. How many new reviews did I and each competitor receive? What is the sentiment (positive/negative/neutral)?
4. Which issues dominate the negative reviews (sizing/fabric/logistics/other)?
Output: write into the "Daily Check" view in Base, one row per ASIN.
The agent calls the competitor-monitoring Skill, then the review-analysis Skill, and writes the results back to the table. Minutes later your Base holds an inspection table where the changes are visible at a glance.
Sub-scenario B: New Listing Launch
1. Title (≤200 characters, with core keywords)
2. Bullets (5, each ≤500 characters, features plus specs)
3. Product description (≤2000 characters)
4. Backend search terms (≤250 characters)
5. English → German → Italian versions
6. Constraint: keywords must come from real keyword-research Skill data
The agent calls the keyword-research Skill, the listing-optimisation Skill, and the multilingual copy Skill, then hands you a complete three-language listing set ready to paste into Seller Central.
Sub-scenario C: Friday Weekly Report
1. Sales overview: total revenue, order volume, return rate (compared with last week)
2. Review movement: new review count, average star rating, top three negative causes
3. Competitor changes: price adjustments, listing edits, new launches
4. Ad performance: total spend, ad spend, conversion rate, top five spending keywords
5. Next week's focus: three actions based on the above. Data sources: the "Daily Check" view, ad reports, and the inventory table in Base.
The agent reads your Base — data written by earlier conversations — and assembles a standard report using the operations-reporting Skill.
The precondition is that every earlier conversation wrote its data into Base. That is precisely why the recommended pattern above matters.
Designing Your Own Work
To package a Work of your own, walk these five steps:
| Step | The question to answer |
|---|---|
| 1 Define the job | What is the complete chain? Where does it start and where does it end? |
| 2 Break out the stages | Which steps are unavoidable along that chain? |
| 3 Match capability | What capability does each stage need, and does an existing Skill cover it? |
| 4 Find the gaps | Which stages have no Skill? Those you either write yourself or fill by hand |
| 5 Agree the handoffs | Where does data live between stages? A Base table? A Library folder? |
In step 5, name the handoff for each stage, such as a Base table and fields or a Library file path. Later steps use these locations as inputs.
Summary
A Skill is one capability; a Work is how they combine. Choose a Work and the whole set arrives at once.
What it really solves is not "fewer clicks" but keeping the way a job is done consistent across a team — everyone using the same capability set and running the same process.
That completes how capabilities are combined. The next lesson moves to the asset layer: team space, bulk work, and knowledge sources in Library.
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.
2.6 Library in Depth: Team Space, Bulk Work, and Knowledge Sources
Lesson 1.8.1 got your files in. This one covers what happens once you have thousands of files and a whole team.