Claude Code skills
Claude Code skills and Microsoft Copilot agents solve the same problem: you have a job you want done the same way every time, and you are tired of retyping the prompt. Here is how to build and manage both, based on Anthropic's and Microsoft's own documentation.
Build a Claude Code skill
Create a SKILL.md
A skill is a folder containing a SKILL.md file. The file starts with a short frontmatter block (a name and a one-line description of when to use it) followed by the instructions themselves. Claude reads the description to decide when the skill is relevant, so write it as a trigger, not a title.
Put it where Claude can find it
Personal skills live in your user-level skills folder and follow you across projects. Project skills live in the repository's skills folder, so everyone working on that codebase gets the same guidance. Claude Code ships with bundled skills as well.
Bundle supporting files
A skill can be more than one file: scripts to run, templates to fill, reference documents to consult. Keep SKILL.md as the short entry point and move long material into sibling files the skill links to.
Invoke it directly or let it trigger
Once installed, a skill is used automatically when the task matches its description, or you can call it by name like a slash command. Test it on a real task and tighten the description if it triggers too often — or not at all.
Share and version it
Because a skill is just files, it versions like code: commit it with the project, or publish the folder for others to drop into their own setup. Agent Skills were published as an open standard in December 2025, so a well-formed skill travels between tools.
Do the same in Microsoft Copilot
Microsoft's equivalent of a skill is an agent: a set of instructions plus the knowledge and actions it may use. You author it in Copilot Studio, or package it as a declarative agent that runs inside Microsoft 365 Copilot.
Start with instructions
In Copilot Studio, instructions are the central directions an agent follows: which tools, knowledge sources, and topics to call, and how to shape the response. Microsoft's guidance: assign a role, specify format and style, and state clearly what the agent should not do.
Build a declarative agent for Microsoft 365 Copilot
A declarative agent is a customized version of Microsoft 365 Copilot defined by a manifest: a machine-readable document declaring the agent's instructions, knowledge, and actions. You build it with the Microsoft 365 Agents Toolkit and it runs inside the Copilot your organization already uses.
Answer three questions before writing
Microsoft's instruction-writing guidance comes down to: what goal must the agent accomplish, what workflows will users go through, and what business logic belongs in the loop. Answer those first and the instructions almost write themselves.
Scope the knowledge
Declarative agents ground their answers in the knowledge you attach — SharePoint content, files, or other sources — so keep that set tight and current. A focused agent with a few good sources outperforms a broad one.
Test like a user, not an author
Run the workflows you listed in step three end to end. Where the agent picks the wrong tool or format, fix the instructions rather than adding exceptions — short, clear directions beat a wall of rules.
Claude Code skill vs. Copilot agent
- Format: a Claude skill is a folder of files led by SKILL.md; a Copilot agent is declared in a manifest plus instructions authored in Copilot Studio.
- Where it runs: skills run wherever Claude Code runs; Copilot agents run inside Microsoft 365 Copilot and Copilot Studio.
- Sharing: skills move as plain files and follow an open standard; Copilot agents deploy through your Microsoft 365 tenant.
- Same rule for both: one repeatable job, a clear trigger description, and short instructions win.