Mpromp2 — the AI Prompt GeneratorSkills Library / Microsoft Copilot
Microsoft Copilot Agents & Instructions
Inside Microsoft 365, Copilot's skills layer consists of declarative agents: named assistants defined by an instruction set, grounded in the files, sites, and data you point them at. This page explains the setup and gives you an instruction set shaped for work documents.
How it works
A Copilot agent is built in Copilot Studio (or the agent builder in Microsoft 365 Copilot): you describe the agent, write its instructions, and connect knowledge sources such as SharePoint sites, files, and Dataverse. The agent then answers and acts within that grounding. The instruction craft is identical in spirit to SKILL.md: define the job, the sources it may use, the format of its answers, and the boundaries - what it must not answer from its own knowledge when the sources are silent.
Setup, step by step
- Define the agent's job in one sentence a manager could approve
- Write instructions: role, sources it may use, answer format with citations to the source, and the fallback line when sources are silent
- Connect only the knowledge sources the job needs - breadth is a liability here
- Test against questions whose answers exist in the sources and questions whose answers do not
- Publish to the team scope the job belongs to, not the whole tenant by default
Copy-ready starter file
Field notes
Grounding discipline is the whole value of a Copilot agent: an agent that improvises from general knowledge is just a chatbot with a logo. The silent-source rule in the template is the line that makes it trustworthy.
Keep going
Mpromp2 — the AI Prompt Generator. Every guide and skill file here is free to use, including commercially.