On September 30, 2026, Google announced that it is rolling out "Skills," a feature for saving and reusing frequently used instructions, to regular Gemini chats and Google Workspace.
Gems, which let users create dedicated AI assistants, are scheduled to move to Skills in November for personal Google accounts.
With Skills, multiple sets of instructions can be combined within a single conversation. Skills also support the open SKILL.md format, which a growing number of other AI services are adopting.
But being able to carry instructions over in the same format is not the same as being able to reproduce your previous work exactly. How much of the tools and automation you used with Gems carries over will determine how practical Skills are after the move.
Combining Multiple "Work Procedures" in the Same Chat
You can invoke a Skill in Gemini's input box by typing "/" followed by its name.
Even if the user doesn't specify one, Gemini can also automatically apply a Skill it judges to be suited to the request.
The major difference from Gems is that you don't have to switch to a dedicated chat screen. You can add the instructions you need to a conversation already in progress.
For example, when writing, you can use a Skill that defines your personal writing style together with a Skill that defines your company's brand guidelines. This is a usage example Google itself has presented.
Compared with cramming style and brand standards into one long prompt, each can be edited independently, and the combination can be changed to suit the purpose.
The shift is away from switching between separate dedicated chatbots for each purpose and toward combining the instructions a single conversation needs.
You can create Skills without editing SKILL.md yourself.
You can ask Gemini in a chat to "turn these instructions into a Skill," save it, and then keep refining its contents conversationally.
You can also add reference materials such as text documents, PDFs and images. That means you can go beyond simply telling Gemini to "write like me" and give it concrete procedures along with actual writing samples or internal standards.
According to the official help page, using Skills in Gemini chats does not require a paid Google AI subscription.
For personal use, the current requirements are that you are 18 or older, signed in to a personal Google account, and have "Keep Activity" turned on. The rollout is being carried out in phases.
Separate requirements apply to features such as running tasks automatically or setting schedules in Gemini Spark. The requirements for using Skills themselves and those for task and schedule features need to be considered separately.
As of the announcement, Skill sharing, adding files from Google Drive and integration with Gemini Notebook were due to be added over the following few weeks.
At present, there is a difference between being able to upload local files directly and being able to reference materials stored in the cloud as they are.
What SKILL.md Standardizes Is the "Work Procedure"
In its announcement for Workspace, Google explains that Skills are based on the open SKILL.md standard.
Skills created on other platforms can be imported into the Gemini app and Workspace. In this respect, it is more than a simple renaming of Gems or a change to their interface.
Agent Skills is a mechanism that Anthropic announced on October 16, 2025, and released as an open standard on December 18 of the same year.
In its technical explanation, the company describes a design that bundles instructions, scripts, reference materials and more into a single folder to teach an AI how to do a job.
The central SKILL.md is a text file. Information such as the name and description goes at the top, followed by concrete procedures written in Markdown.
The standard specification recommends that the description field state not only what the Skill does but also when it should be used.
This serves as an explanation for humans and also as information the AI uses to decide when to apply the Skill.
Another feature is the idea of loading only the information that is needed, in stages.
Under the specification, the AI first reads the name and description of each Skill, then loads the detailed instructions in SKILL.md once it judges the Skill necessary. Reference materials and scripts are consulted only when the actual work requires them.
Instead of handing the AI a large set of internal manuals up front, the structure moves from an overview to the specific procedures and materials needed.
However, this is the mechanism presented in the Agent Skills specification. It has not been confirmed that Gemini loads information internally in exactly the same order or manner.
What Google has explicitly stated is support for the common file format, not a guarantee that each company's AI behaves the same way internally.
Still, with this format, writing style, review criteria, work procedures and the like can be saved as files rather than being locked inside one service's settings screen.
Related reference materials can be managed together, which makes the contents easier to review when moving to another AI service.
This can reduce the effort of rewriting long instructions from scratch, but using the same SKILL.md does not guarantee the same results in every AI. The final behavior also depends on the model used and the tools available.
The Same SKILL.md Doesn't Guarantee the Same Work Can Be Reproduced
In Gemini's Skills, you can also upload files that contain scripts such as .py and .sh files.
However, scripts that require an internet connection are not supported.
The upload requirements state explicitly that scripts included in a Skill cannot send requests to or interact with external websites.
In other words, simply putting a Python script that operates an external service into a Skill's folder does not let Gemini use that service.
The Agent Skills specification also provides fields for describing execution requirements, such as supported products, required software and network access.
Which scripting languages can be used also depends on how the AI service that runs the Skill implements it.
What is standardized is mainly the format for describing work procedures and passing them to the AI. The environment for carrying out those procedures is not automatically standardized.
Reference files have limits too.
A Skill uploaded to Gemini must be 100 MB or less in total across all files, and SKILL.md must be at the top level of the folder.
PDFs and images are supported, but binary formats such as .docx and .xlsx are not accepted for Skill uploads.
The 100 MB figure is only the upload size limit. It does not mean that all of the content is loaded into the AI's context at once.
According to the migration help page, Skills currently do not support Canvas or Deep Research.
Among the default tools available in Gems, video creation, music creation, Canvas, Deep Research and guided learning are also not currently available from Skills.
On the other hand, Skills can be used together with Connected Apps.
It would therefore be a mistake to read the restriction that scripts inside a Skill cannot access the external internet as meaning that Gemini cannot use any external apps.
For example, even if you can save the procedure "research a company and write a report in a set format" as a Skill, that Skill may not be able to run Deep Research.
If you use existing Gems for work, you need to check separately whether the instructions migrated correctly and whether the tools that work requires are still available afterward.
What Migrates Automatically From Gems, and What Doesn't
According to Google's guidance as of October 1, 2026, regular Gems will be automatically migrated to Skills.
On the other hand, workflows created in Opal from Google Labs will not be migrated automatically, and Skills created in the Gemini app are not currently synced with Skills on the Workspace side.
Organizing Google's Gemini migration help, Workspace announcement, admin transition guide and Opal FAQ gives the following picture.
| Item | What carries over | Conditions for continued use |
|---|---|---|
| Regular Gems on personal accounts | Gems and their reference files are automatically migrated to Skills | Some default tools are not supported in Skills. GitHub files are not currently supported in Skills |
| Gems created in the Gemini app on Workspace accounts | Name, description, instructions, supported knowledge files, etc. are migrated to Skills | After migration they become disabled drafts that the creator must review and enable. Skills for use in Workspace apps must be created separately |
| Opal / Gems by Google Labs | Saved workflow files remain in Google Drive | Not automatically migrated to Skills; running them in Opal ends on November 17, 2026 |
For regular Gems, saved instructions and supported files are migrated automatically, so there's no need to rebuild everything from scratch.
On Workspace accounts, however, migrated Skills are created as drafts that must be reviewed and enabled before use.
Because Skills in the Gemini app and in Workspace apps are currently not synced, you need to prepare them separately if you want to use the same work procedure in both.
Opal is a different case.
The saved workflow files themselves remain in the "Opal" folder in Google Drive, but they are not automatically converted to Skills.
The Opal service and "Gems made by Labs" will end on November 17, 2026, after which saved workflows can no longer be run in Opal.
In other words, having your data saved is not the same as being able to keep running it as is.
The schedules also differ between individuals and organizations.
Gems for personal use are scheduled to move to Skills in November 2026.
In Workspace, Gems will move to the Gemini app's settings screen on November 17, but can still be created, edited and used at that point.
Gems are scheduled to end for Business and Enterprise no earlier than March 1, 2027, and for Education no earlier than June 1, so they will not all end at once in November for every account.
Skills for Workspace begin rolling out on October 5 for rapid-release organizations and October 19 for scheduled-release organizations.
The rollout to Workspace users in the Gemini app begins on October 13 and is scheduled to be completed by mid-November.
Because Skills created in the Gemini app and Skills on the Workspace side don't sync, teams that use the same work standards in both need to check that the contents of the Skills they created in each place match.
If you automate with Workspace Studio, further action is needed.
The "Ask a Gem" step can no longer be added to new flows, and "Ask a Gem" steps in existing flows are scheduled to stop working when Gems end.
Waiting for Gems to be migrated automatically to Skills will not automatically replace existing business workflows.
OpenAI Is Also Moving From Custom GPTs to a "Procedures Plus Tools" Setup
OpenAI has also laid out a plan to migrate custom GPTs to Plugins.
According to its official FAQ, the standard retirement date for custom GPTs is December 11, 2026, and February 11, 2027 for some Enterprise workspaces that have been approved for an extension.
During migration, the instructions set in a GPT become a Skill within the new Plugin, and files registered in Knowledge are copied as reference materials.
Custom actions that operate external services, on the other hand, are not migrated automatically.
You need to prepare an app or another connection method that provides the required functionality, and also confirm that users after the migration have permission to access them.
The migration mechanisms and offerings differ between Google and OpenAI.
Still, they share one point: moving instructions to the AI and moving the connections to the tools and external services those instructions depend on are separate tasks.
The customization features of Google and OpenAI are moving closer to the idea Anthropic presented with Agent Skills: giving a general-purpose AI agent the work procedures it needs, when it needs them.
Rather than only building a dedicated AI bot for each purpose, approaches that combine the necessary procedures within the same conversation are spreading.
As this change advances, it should become easier for companies and individuals to separate the business knowledge they've accumulated from any particular bot and reuse it across multiple AI services.
However, the test of whether a migration has succeeded is not whether a Skill with the same name was created.
You need to check in practice whether the right Skill is selected for your usual requests, whether it correctly references the necessary materials, and whether the required tools and external services are available.
For a writing Skill, give the same request before and after migration and compare the output. For a Skill that includes automation, test it including connections to external services and permissions.
Only when both the instructions and the execution environment are in place can the work procedures you've built up continue to be used in the new AI environment.