What counts as an AI skill for HR work: the four kinds
An AI skill for HR can be a SKILL.md file, a custom GPT, a Copilot agent, or a plugin or connector. Here's what each one is, where it runs, and what it reaches.
AI HR Skills Editorial TeamAn AI skill for HR work is any AI-powered building block that helps with a people task. In practice it’s one of four things: a SKILL.md file, a custom GPT, a Copilot agent, or a plugin or connector. They differ in what they’re made of, where they run, and what data they can reach.
The names overlap, and they keep changing. To some people, “skill” means one file format; to others, it means anything AI-flavored. OpenAI plans to retire custom GPTs and help creators move them into plugins. A Microsoft 365 Copilot plugin can bundle an agent with skills and connectors.
Here’s each kind in plain words, so you know what you’re looking at before you use one.
Key takeaways
- A SKILL.md file is a folder of instructions an AI agent loads when a task needs it, in an open format from Anthropic.
- OpenAI plans to retire custom GPTs on December 11, 2026, and help creators move them into plugins.
- In Microsoft 365, the usual form is an agent, built as a declarative agent or in Copilot Studio, Microsoft’s low-code tool.
- Plugins bundle these pieces, and connectors reach outside data, so connectors deserve the closest look.
A SKILL.md file: open agent skills and Claude skills
A skill in the narrow sense is a folder with a file called SKILL.md inside. According to the Agent Skills specification site, that file holds a name, a description, and instructions that tell an agent how to do a task. The folder can also include scripts, reference documents, and templates.
Anthropic developed the format and released it as an open standard. That’s why you’ll see open agent skills and Claude skills used for the same kind of thing. Anthropic’s documentation says Claude reads only each skill’s name and description until a request matches, then loads the full instructions.
For HR, a skill might be a written method: how to draft a structured interview guide, say, or how to summarize survey comments without quoting anyone. Because it’s a set of files, you can read all of it before you use it.
A custom GPT
A custom GPT is a version of ChatGPT set up for a specific purpose. OpenAI’s help center lists what it can include: instructions, conversation starters, uploaded knowledge files, and tools such as web search. It can also use apps or actions that connect to outside services.
That format is ending. OpenAI’s retirement FAQ says custom GPTs are scheduled to retire on December 11, 2026, or on February 11, 2027 for some workspaces. When a creator migrates a GPT, its instructions become a skill inside a plugin, and its knowledge files are copied over as reference files. Custom actions don’t transfer, so they have to be rebuilt.
If your HR team relies on someone else’s GPT, check the notice on it and ask its creator what replaces it.
A Copilot agent: declarative agents and Copilot Studio
In Microsoft 365, the common form is an agent. You’ll mostly hear about two kinds: declarative agents and agents built in Copilot Studio.
A declarative agent is Microsoft 365 Copilot configured for one job. According to Microsoft’s documentation, you define its purpose, instructions, knowledge sources, and actions. It uses Copilot’s own model and orchestrator, and people use it in Microsoft 365 Copilot and supported Microsoft 365 apps. Microsoft’s own example is an employee-support agent that answers common questions from approved company knowledge.
Copilot Studio is Microsoft’s low-code studio for building agents and workflows and publishing them to places like Teams, Microsoft Copilot, and websites. If someone in your organization built an agent without writing code, it was likely made here.
A third route, the custom engine agent, brings its own model and orchestration instead of Copilot’s.
A plugin or a connector
A plugin is a package. In ChatGPT, OpenAI says a plugin can include skills, connected apps, and other extensions. In Microsoft 365 Copilot, Microsoft says a plugin can include an agent along with skills, Copilot connectors, or MCP-based tools. One of Microsoft’s preview developer tools builds those skills on SKILL.md files.
A connector is what reaches your data. Microsoft’s Copilot connectors come in two kinds. Synced connectors copy outside data into Microsoft 365 so Copilot can search it.
Federated connectors fetch data live from the source system through the Model Context Protocol (MCP), a protocol that lets AI tools connect to outside services. Either way, Microsoft says people only see content they already have permission to access.
Kind of skill
SKILL.md file
- Made of
- A folder of instructions, sometimes with scripts and reference files
- Runs in
- Claude and other agent products that support the open format
- Check first
- Every file, especially scripts and web addresses
Kind of skill
Custom GPT
- Made of
- Instructions, knowledge files, tools, and apps or actions
- Runs in
- ChatGPT, until its retirement date
- Check first
- Which outside services it calls, and what replaces it
Kind of skill
Copilot agent
- Made of
- Instructions, knowledge sources, and actions, built as a declarative agent or in Copilot Studio
- Runs in
- Microsoft 365 Copilot, Teams, and other Microsoft channels
- Check first
- Its knowledge sources, its actions, and who in IT approved it
Kind of skill
Plugin or connector
- Made of
- A package of skills, apps, or agents, or a link to an outside data source
- Runs in
- ChatGPT or Microsoft 365 Copilot, depending on who built it
- Check first
- What data it reaches and whether it copies that data or fetches it live
A hypothetical example
Imagine a hypothetical HR coordinator who asks IT whether the team can “install a skill” for writing onboarding checklists. IT’s first question is which kind.
If it’s a SKILL.md file, IT can read every file in it. If it’s a custom GPT, it’s scheduled to retire, so IT will ask what replaces it. If it’s a Copilot agent or plugin, IT will ask who owns it and which connectors it uses. A connector might reach the HR system.
Same word, four different conversations.
Why the label matters less than what it touches
We use “skill” as the umbrella term in the AI HR Skills library because the formats keep shifting. A file format that’s new this year may be packaged differently next year. What stays useful is knowing what a tool is made of, where it runs, and what it can reach.
Each listing in the library shows what kind of tool it is, along with a named author and a working source. That’s part of how listings are reviewed. A listing still isn’t a guarantee of fit, so check what each tool touches before your team relies on it.
If you’ve built any of these four kinds of tool for HR work, submit your own skill.
FAQ
Is a SKILL.md file only for Claude?
No. Anthropic developed the format and released it as an open standard that other agent products have adopted. One of Microsoft’s preview developer tools for Microsoft 365 Copilot plugins also builds skills on SKILL.md files.
Are custom GPTs going away?
Yes. OpenAI plans to retire them on December 11, 2026, with a later date for some workspaces. A migrated GPT becomes a plugin, and its custom actions have to be rebuilt.
What’s the difference between a declarative agent and Copilot Studio?
A declarative agent is a type of agent that runs on Microsoft 365 Copilot’s own model with your instructions, knowledge, and actions. Copilot Studio is a low-code tool for building agents and workflows.
What is a connector?
A Copilot connector brings data from outside Microsoft 365 into Copilot. It either copies the data in or fetches it live through MCP.
What to do next
When someone offers your team an AI skill, ask which of the four kinds it is. Each kind needs a different check. Anything with a connector reaches data outside the chat.
To see real examples, browse the full library of HR skills and plugins, or see other guides in the resources section.
Sources
- Source: Agent Skills overview, agentskills.io
- Source: Agent Skills overview, Claude Docs
- Source: GPTs in ChatGPT, OpenAI Help Center
- Source: Custom GPT retirement and migration FAQ, OpenAI Help Center
- Source: Plugins in ChatGPT, OpenAI Help Center
- Source: Declarative agents for Microsoft 365 Copilot, Microsoft Learn
- Source: Compare declarative and custom engine agents, Microsoft Learn
- Source: Overview, Microsoft Copilot Studio, Microsoft Learn
- Source: Build or reuse skills for a Microsoft 365 Copilot plugin, Microsoft Learn
- Source: Copilot connectors overview, Microsoft Learn