AI Skills Compared: Pros and Cons by Service
This page compares the skill mechanisms of each AI service. Major AI services each have a way to save frequently used instructions and procedures for reuse. In the second half of 2026, many of those mechanisms started converging on SKILL.md "skills."
- ChatGPT plans to retire custom GPTs on December 11, 2026. Their instructions move into skills inside plugins (Source 1).
- Gemini is moving Gems to skills. For personal accounts, Gems are removed on December 17, 2026 (Source 2).
- Copilot Studio accepts SKILL.md skills as a preview (Source 3).
For the steps to install and invoke named skills, see the Skills Guide.
Skills in chat AI
In AI you use through a browser or app chat, you create a skill as a set of instructions and materials. You can build one without writing code. How you share it, and where you can use it, differs by service.
ChatGPT is moving from custom GPTs to plugins. A plugin can hold skills, service connections, and reference files. You install plugins from the directory that ChatGPT shares with Codex.
Skills in plugins
You write instructions and procedures as skills and use them inside a plugin. In chat, type @ to pick a skill.
ProsThe same plugin works in both ChatGPT and Codex
ConsWhat you can use depends on your account or workspace settings
Best forHanding off a repeated workflow in a fixed form
Get itInstall from the Plugins tab
Custom GPTs (retiring)
Dedicated GPTs with instructions and knowledge files. They are scheduled to retire on December 11, 2026, and work as before until then.
ProsCreators get a migration flow to plugins
ConsCustom actions and sharing settings don't carry over
Best forMigrating to a plugin and testing it before retirement
Get it"Migrate to plugin" in My GPTs
Claude has two mechanisms with different roles. Projects fix the context by bundling materials and instructions, while Skills reuse procedures as SKILL.md. Skills you enable on claude.ai are also available in Claude Code when you sign in with the same account.
Skills
Official skills for PDF, Word, PowerPoint, and Excel, plus skills you create yourself. They start automatically when you describe the task.
ProsThe same SKILL.md can be shared with Claude Code
ConsThe official example skills require a paid plan
Best forCreating and extracting routine documents such as invoices and reports
Get itEnable them in claude.ai settings
Projects
Bundles related materials and instructions in one place so you can reuse a long context. Unlike skills, it fixes the context itself.
ProsKeeps assumptions and materials without re-pasting
ConsWhen to use it versus Skills is not obvious at first
Best forLong projects that keep referring to the same materials
Get itCreate from the claude.ai sidebar
Gemini is moving from Gems to skills. You invoke a skill by typing / and its name in chat. Google says this will soon change to @.
Skills
Saved instructions you reuse. Gemini applies a skill automatically when it is relevant, and you can combine several skills in one chat.
ProsUsed automatically when relevant, without invoking
ConsFor now, only personal accounts of users 18 and over
Best forTasks where you keep writing the same context
Get itCreate a skill in the Gemini app
Gems (retiring)
Custom assistants with saved instructions. For personal accounts, they become read-only on November 17, 2026 and are removed on December 17. Work and school accounts follow in 2027.
ProsGoogle recreates your Gems as skills automatically
ConsYou have to download knowledge files and add them again yourself
Best forSaving knowledge files before the deadline
Get itOpen from the Gem list in Gemini
In Copilot Studio you can build agents with their own instructions, knowledge, and connections to business systems. Agents that run on the GitHub Copilot harness can use SKILL.md skills as a preview.
Copilot Studio skills (preview)
Upload a SKILL.md, or a ZIP that contains one, to an agent. Skills are handled separately from tools that connect to external services.
ProsYou can bring SKILL.md files written for other tools
ConsIt is a preview, so details may change
Best forTeaching business agents your internal procedures
Get it"Upload a skill" in the agent's Build tab
Microsoft 365 agents
Business agents that answer from internal data such as SharePoint and Outlook.
ProsDeep integration with Office apps
ConsRequires a Microsoft 365 license
Best forAnswering questions grounded in internal documents
Get itBuild in Copilot Studio
Coding-agent approaches
In developer AI, skills have converged on a shared format, SKILL.md.
What comes built in, and where you add skills from, still differs by tool.
Here we compare each tool's skill mechanism, not individual skills.
How you invoke a skill also depends on where you use it.
In Codex, you call skills with /skills or $skill-name in the CLI, and you can open them from Skills in the sidebar of the ChatGPT desktop app (Source 4).
In Claude Code, you call skills with a slash and the name, and in the desktop app and on the web they also start when you ask in plain language.
For the list of skills you can use and how to install them, see the Skills Guide.
Claude Code ships with bundled skills for review, verification, and more. Official skills such as the document skills, and skills made by others, are added through plugins (Source 5).
Bundled skills
/code-review, /verify, /loop, /debug, and others work without installing anything.
ProsReview and verification work with no setup
ConsMatching your own procedures takes custom skills
Best forGetting started with the built-in features
Get itSee the Skills Guide for the list and usage
Plugins
Install skills, subagents, hooks, and MCP servers together as one package. The official marketplace is registered by default.
ProsA team can share the same toolset
ConsRead community plugins before installing them
Best forAdding specialized workflows that aren't built in
Get itPick one in /plugin
Subagents
Split work across role-specific helpers and run research and implementation in parallel.
ProsParallelizes work that spans many files
ConsMore token use and more design effort
Best forSplitting large research or refactoring jobs
Get itBuilt in, no extra install
Codex now packages skills for sharing as plugins and installs them from the directory it shares with ChatGPT. Skills for your own use are placed as folders (Source 4, Source 6).
Plugins
Distribute skills, MCP servers, and hooks together. ChatGPT and Codex use the same directory.
ProsThe same skills work in chat and in coding
ConsThe IDE extension doesn't support plugins
Best forSharing skills with a team
Get it/plugins in the CLI
Local skill folders
Place folders in .agents/skills/ to load them. Codex reads from repository, user, and admin locations.
ProsWorks as soon as you place it, and changes load automatically
ConsNot suited to sharing with others
Best forRepository-specific procedures
Get itPlace it in ~/.agents/skills/ or similar
System skills
Comes with skill-creator for creating skills and skill-installer for installing them locally.
ProsCreate and try skills without leaving Codex
ConsYou need to learn the $-prefixed calls
Best forQuickly creating your own skills
Get it$skill-creator
Cursor separates rules it reads every time from skills it picks for the task. It loads skills from .cursor/skills/ and .agents/skills/, and also reads the Claude Code and Codex skill locations (Source 7).
Project rules (.cursor/rules)
Files in .cursor/rules/ that apply naming conventions and coding policy from the start of every conversation.
ProsApply consistently to every action in the editor
ConsMore rules mean more to read every time
Best forAligning coding standards across a team
Get itPlace files in .cursor/rules/
Agent Skills
Place SKILL.md folders and the agent loads them to match the task. Unlike rules, they are read only when needed.
ProsSkills written for other tools work as is
ConsYou need to decide what goes in rules and what goes in skills
Best forLong procedures you use now and then
Get itPlace the folder and call it by typing / and the name in chat
AGENTS.md
An instruction file describing the repository's context, usable across tools.
ProsCan be shared with other AI tools
ConsIts role easily blurs with editor-specific settings
Best forSharing context across several tools
Get itPlace AGENTS.md at the repository root
Which to choose
The deciding factor is how far you want to carry the skills you build. As of October 2026, ChatGPT, Gemini, Copilot Studio, Claude, and the coding tools all accept the SKILL.md format. So if you write your procedures as SKILL.md, switching services takes little rewriting. Anything you built as a custom GPT or a Gem is safer moved to skills before its retirement date.
Once you've chosen an approach, check how to get and invoke named skills in the Skills Guide.
Sources
- OpenAI, "Custom GPT retirement and migration FAQ" (checked October 26, 2026)
https://help.openai.com/en/articles/20001519-custom-gpt-retirement-and-migration-faq - Google, "About the transition from Gems to skills" (checked October 26, 2026)
https://support.google.com/gemini/answer/18560919?hl=en - Microsoft, "Skills overview for agents" (checked October 26, 2026)
https://learn.microsoft.com/en-us/microsoft-copilot-studio/agents-experience/skills-overview - OpenAI, "Build skills" (checked October 26, 2026)
https://developers.openai.com/codex/skills - Anthropic, "Extend Claude with skills" (checked October 26, 2026)
https://code.claude.com/docs/en/skills - OpenAI, "Plugins" (checked October 26, 2026)
https://developers.openai.com/codex/plugins - Cursor, "Agent Skills" (checked October 26, 2026)
https://cursor.com/docs/skills