Skills: turn your repeated workflows into a command

There's a kind of work we all repeat without noticing: the same report every Monday, the same way of summarizing a sales CSV, the same instructions we type to the assistant over and over. Every time we re-explain the same thing we lose a little time and, more importantly, we lose consistency. Minte Skills exist precisely to fix that.

A skill is a reusable capability that extends what Minte's chat and your agents can do. You define it once and then trigger it with a slash command —for example /sales-summary— from any conversation where it's available. Instead of rewriting the context each time, you wrap the whole workflow and invoke it like a custom command.

What a Skill wraps

In its simplest form, a skill is a set of instructions: the script the assistant should follow for a specific task. In fact, most skills are just that, and they work great. When a task needs precision, a skill can also include scripts —small Python or Bash programs that run deterministically in a sandboxed environment— ideal for exact calculations or transformations you don't want to leave to the model's interpretation. And if the workflow needs supporting data, it can carry resources like CSVs, images or PDFs that its scripts read at runtime.

The idea is simple: you define what to do once, then invoke it with /its-command whenever you need it.

Create a Skill by chatting

The best part is you don't need to know how to code to create one. The most natural way is to describe to the assistant what you want to automate: "create a skill that summarizes the week's sales from a CSV," and it prepares the proposal for you.

Before saving anything, Minte shows a confirmation card with the full detail of the skill —name, command, description, scope and, if applicable, the included scripts and resources—. You review the proposal and press Confirm to create it or Cancel to discard it. The assistant automatically picks the right approach: instructions only when guiding behavior is enough, a deterministic script when an exact calculation is needed, or a script with read-only search access when it has to look up information from the web, news or the knowledge base.

Likewise, you can ask it to modify an existing skill or to delete it, and you'll always see a confirmation card before the change is applied.

Share what works

A skill comes into its own when it stops being personal. You can share it within your organization with different visibility levels: private, just for you; for the managers across the whole company; or with specific groups and members. When someone shares a skill with you, it appears in your list as available and a single tap activates it so you can start using it with its command.

Turning a team's repeated workflows into shared skills is the best way to keep things consistent: everyone summarizes that report the same way, everyone follows the same process.

Install skills from outside

Beyond creating them, you can bring in existing skills from a public repository —the assistant searches the registry and installs the one you choose— or by uploading a ZIP file from the connectors section.

For security, installed skills are not activated automatically. A manager must review and approve them, including their scripts and tools, before they can be used. It's a deliberate barrier to avoid running unreviewed code inside your organization.

Skills inside your agents

Skills don't live only in the general chat: they're also part of an agent's capabilities. An agent exposes only the skills explicitly installed on it, so anyone chatting with it can invoke them via their slash command.

Installation happens from the agent's panel, in the Minte Library, under the Skills tab. There you see the ones already installed on that agent and a catalog of those available, with the option to preview them before adding. From that same place you install or remove skills, and the library summary tells you how many are active alongside the rest of your tools, MCP, integrations and automations.

Start simple

If you're going to try Skills, a couple of pointers. Use clear, memorable names for commands, like /monthly-report. Start with instructions: many tasks are solved with a good script and you only need code when exact calculations come into play. If you install third-party skills, read them before approving. And when a skill works well for you, share it: what's a shortcut for you can become a standard for your team.

Skills turn repeated knowledge into something you invoke with a slash. Define it once, and stop explaining it forever.

An open standard, not a Minte quirk

One thing worth highlighting: Minte Skills aren't a proprietary format — they follow the open Agent Skills standard. A skill is, at its core, a folder with a SKILL.md file that holds its name, description and instructions, and can bundle scripts and supporting resources. That same format is the one many of the largest AI platforms use today.

The standard originated at Anthropic and was released as an open specification, and since then it has been adopted by tools like Claude, GitHub Copilot, VS Code, Cursor, OpenAI Codex and Google Gemini, among many others. For you, that means the effort of defining a good workflow isn't locked into a single tool: what you learn and build with Minte Skills rests on a common language the whole ecosystem is converging on.