- Skills are packages of instructions and reference files that teach Gemini how to perform specific tasks on a recurring basis.
- They can be created via chat, templates, manually, or by importing standard SKILL.md files to ensure portability.
- They work through an activation system based on the relevance of the description, allowing multiple skills to be combined in a single chat.
- They are available free of charge to personal accounts over 18 years of age, with limits of 100 active skills and 100 MB per upload.
If you've ever felt that Gemini could give you better answers if it knew exactly how you work or your preferences, it's probably because you haven't tried custom skills yet . Basically, we're talking about packages of instructions and context that allow the AI to stop generalizing and start acting like an expert in a specific task that you've defined, avoiding having to repeat the same lengthy prompt every time you open a chat.
The most interesting aspect of this system is that Google has focused on portability and standardization , allowing these skills to be based on files that you could move to other AI models. It's no longer just about creating a chatbot with a personality, but about designing knowledge modules that Gemini can automatically activate based on your needs at any given moment, making the interaction much smoother and more professional.
What exactly are skills and how do they work?
Simply put, a skill is a set of reusable instructions along with optional reference files. Unlike the old Gems, which were more like closed containers, skills allow Gemini to access specific documentation to perform a recurring task according to your strict rules. You can invoke them by typing a forward slash (/) in the chat or let the AI activate them automatically if it detects that your request matches the skill's description.
The internal activation process is quite interesting and occurs in three stages. First, Gemini analyzes the skill's name and description to see if it's relevant; if there's a match, it proceeds to read the complete instructions and, only if strictly necessary, opens the attachments . This system is much more efficient than loading the entire context into each message, which usually translates into more accurate responses and fewer misinterpretations.
Requirements and current availability
Not everyone can start experimenting with this right now, as Google has set some strict conditions . To access the skills, you must be over 18 and use a personal Google account , since educational or business accounts are temporarily excluded. Additionally, it's essential to have Gemini's Keep Activity feature enabled for the system to function correctly.
Regarding cost, there's a lot of confusion about it, but the reality is that skills are available for free . You don't need a Google AI Ultra subscription to create them or use them in your regular chats. What may vary depending on your plan is the overall processing power or access to advanced features like Gemini Spark, but skill creation itself doesn't have a paywall.
Step-by-step guide to creating your own skill
Google offers several paths to bring a skill to life, depending on how much control you want to have over the final result:
- Gemini assistance: You can ask them directly in a chat to create a skill based on specific instructions. It's the quickest way for those who don't want to complicate things.
- Predefined templates: In the settings section, there are recommended templates that you can adapt by changing the name and objectives.
- Creation from scratch: Manually filling in the name, description, and instructions fields on the Skills page.
- Uploading SKILL.md files: For more technical users, you can upload a Markdown file or a compressed folder. This is the ideal option if you're looking for version your work or take the skill to other platforms.
If you decide to create a skill manually, an expert trick is to write the description as a trigger , starting with phrases like "Use this when…", as this makes it easier for Gemini to know exactly when to activate the tool without you asking.
Technical limitations and important restrictions
To avoid surprises, it's important to understand the system's operational limits . Currently, you can have up to 100 active skills simultaneously, and the upload limit is 100 MB per skill . Regarding file formats, forget about .docx or .xlsx files; Gemini prefers plain text (.md, .csv, .json, .py, .html) and PDFs or images.
A critical point is that the scripts included in skills don't have internet access . If you try to import a skill that calls an external API to get real-time data, it most likely won't work. Also, keep in mind that manually created skills aren't saved in the mobile app , so it's best to manage them from the web to avoid losing your work.
Advanced CLI management and usage
The management of these tools is centralized on the Skills page, where you can activate, deactivate, or delete your creations. A crucial detail is that to update reference files, you must re-upload the entire skill ; individual files cannot be edited once uploaded. If you have terminal experience, you can use the Gemini CLI to discover skills in the .agents/skills directory and execute them using hotkeys.
For those coming from Gems, the transition is automatic in most cases, but if you want full control, it's recommended to manually export the instructions to a text file and then import them as a SKILL.md file. This ensures that no workflow details are lost during the forced migration that Google is implementing.
Combining skills for complex flows
One of the biggest advantages over the old system is the ability to combine multiple skills in the same conversation. For example, you could activate a skill specializing in technical writing and, at the same time, one that applies your brand's style guidelines . The result is a response that is not only accurate in content but also has the exact tone and format you need without having to copy and paste between different chats.
This modular approach allows AI to be much more versatile. Instead of trying to create a "super-agent" that does everything (and which often gets confused), it's much more effective to have small, specific experts that are activated as needed. This architecture is what allows even lighter models, like Gemini Flash, to achieve surprising results by having an ultra-specific context.
Ultimately, the shift from Gems to skills represents a change towards interoperability and efficiency, allowing any user with a personal account to organize their recurring tasks into portable and precise modules that drastically optimize interaction time with artificial intelligence.