Part 2 of 4 · The repeatable work
A Skill explains one focused task. A Workflow connects Skills into a complete journey. Together, they are the special sauce in your Desktop Workspace: plain-language Agent Instructions saved in the Local Folder that tell the AI Agent how to use the Workspace Information, update the Local Site, and behave consistently.
The Workspace Wizard asks about the work you repeat and the results you want, then creates a generated package for your AI Agent. Extract and open the package folder, type “Build It,” and your AI Agent will create the first Skills and Workflows for you to improve over time.
Why they matter
Skills and Workflows define how your information should be used and how the AI Agent should work with your workspace. They provide guardrails, approval stops, checks, and a consistent process the AI Agent can follow every time.
Instead of explaining the same steps in every prompt, you can simply ask for the result you want. The relevant Skill or Workflow automatically takes over, guides the AI Agent through the saved process, and keeps the workspace working consistently.
A Skill handles one specific task. A Workflow connects several Skills into a complete process. Both use plain language, so you can open, read, and improve them whenever you want.
Bookkeeper Skills include record a received document, prepare a follow-up list, and update a client contact.
Skill: Record a received document
Bookkeeper Workflows include onboard a client, run month-end follow-up, and close a filing month.
Workflow: Month-end document follow-up
It calls:
It decides what comes next and stops before sending.
One Workflow, several Skills
Learn in more detail
The diagram above shows the relationship. Use these details when you are ready to write, review, or improve a Skill or Workflow.
The request that starts it.
The information and approval it needs.
The steps in order.
Actions requiring approval.
How it confirms the result.
What it explains afterward.
I say a client sent a document.
Client, month, document, and received date. Ask for anything missing.
Find the request, mark it received, add the date, and update the client page.
Changing another client or removing a record.
The saved record and Local Site must match.
Tell me what changed and what remains missing.
Approval stops
Require current approval before sending messages, publishing, deleting records, spending money, or contacting customers. The AI Agent stops and explains the proposed action. You choose where every stop belongs.
What to ask the AI Agent to create
Show the task, information, result, and decisions.
Make the job a Skill. Let a Workflow connect Skills.
Save named .md files directly inside skills or workflows.
Use a different client and month. Fix any unclear Skill or Workflow step.
How to manage them