Authoring and publishing skills

A workspace with no skills in it is a chat window. The skills program is what turns an installation into something the client keeps using after the engagement ends, and it is the part most implementations under-resource.

Background on the feature itself is in skills. This page covers running it as a program.

Where skills come from

  • From observation

    The Copilot watches how the team works and proposes a skill when it sees a pattern repeat. Nobody has to spot the opportunity first.

  • From instruction

    Describe the automation in plain language and the workspace builds it. No code, no scripting, no configuration panels.

  • From a template

    Pre-built starting points for common workflows, adapted to the client tools and data rather than used as-is.

For a client implementation, the third route is the fastest start and the worst finish. Templates are starting points, not finished skills: they have to be adapted to the client actual tools, data sources, and conventions before they are worth publishing.

The lifecycle

step 01

Author against a real workflow

Build the first skills around work the client already does badly or slowly, not around what demonstrates well. A skill that saves a named person a real hour each week survives handover. A polished demo does not.

step 02

Review before approving

Approval is a security decision, not a formality. A skill can only reach the tools and data the approving user already has permission to use, so approving as an Admin can give a skill materially broader reach than the person who will run it. Approve as the narrowest role that still works.

step 03

Publish to the right scope

On Teams and Enterprise, a published skill becomes available across the workspace through unified memory. Decide deliberately whether a skill belongs to one project, one department, or the whole organization, using the structure from organization structure.

step 04

Hand over ownership

Name a client-side owner for every published skill before the engagement ends. Skills the client cannot maintain are skills that quietly stop being used the first time a connected tool changes.

Governance

Skills operate inside the workspace governance framework and inherit its access controls. Permissions flow through automatically, so an automation can never become a route around the permission model. That property is worth showing a client security team directly, because it is usually assumed to be untrue of low-code automation.

Why this compounds

Skills accumulate. The library grows into a record of how the client actually operates, and when someone leaves, the skills they built stay behind. That is the argument for treating the skills program as a deliverable with an owner and a budget rather than as configuration work done in the last week.

Back to implementing Skynet.

The industry’s most generous  AI app

We offer 10x more value than any mainstream AI provider

  1. 1. Register

    No Upfront payment required

  2. 2. Create new Chat

    Start chatting with our models

  3. 3. Get more things done

    Get things done with mini apps and agents