Skills

Skills are the building blocks of systemprompt.io. A skill is a reusable set of AI instructions that defines exactly how an agent should perform a specific task -- writing a blog post, analyzing a dataset, generating a customer email, or anything else you need done consistently.

You create a skill once, then use it over and over. Every time you or your team invokes a skill, the agent follows the same instructions and produces consistent, reliable results.

What makes up a skill

Every skill has three core components:

1. Instructions

The instructions are the heart of the skill. They tell the AI what to do, how to do it, and what rules to follow. Good instructions are specific, structured, and include examples of expected output.

For instance, a "Write Product Description" skill might include instructions about tone, length, required sections, and SEO guidelines. The more precise your instructions, the more consistent your results.

2. Parameters

Parameters are the input fields that make a skill flexible and reusable. Instead of hardcoding values into your instructions, you define parameters that get filled in at runtime.

A "Write Blog Post" skill might accept these parameters:

• topic -- The subject to write about
• audience -- Who the post is for
• word_count -- Target length
• tone -- Formal, casual, or technical

When someone runs the skill, they provide values for each parameter. The platform injects those values into the instructions before sending them to Claude.

3. Secrets

Secrets are secure credentials that a skill needs to access external services. API keys, authentication tokens, database connection strings -- anything sensitive goes here.

Secrets are stored encrypted and never exposed in plain text. When a skill runs, the platform injects the secret value at execution time without revealing it in logs or responses.

Learn more in Skill Secrets.

How skills work at runtime

When a skill executes, the platform follows this sequence:

1. Input -- You send a request, either through Claude (via the MCP server) or by triggering a skill directly from the dashboard
2. Assembly -- The platform takes the skill's instructions, substitutes the parameter values you provided, and injects any referenced secrets
3. Execution -- The assembled prompt is sent to Claude for processing
4. Result -- Claude's response is returned to you as the skill output

This happens in seconds. The skill abstraction means you do not need to write or manage prompts manually -- you define them once and the platform handles the rest.

Skill types

Skills fall into two categories based on visibility:

Personal skills are created by you and visible only to your account. Use personal skills for your own workflows, experiments, and tasks that are specific to your work. Every skill starts as a personal skill.

Published skills are shared with your team or the wider systemprompt.io community through the marketplace. Publishing a skill makes it discoverable and installable by others. You control whether a published skill is available to your team only or to all platform users.

Both types work identically at runtime. The only difference is who can discover and use them.

How agents use skills

Agents are the workers that execute skills. The relationship between agents and skills is many-to-many:

• A single agent can have multiple skills assigned to it
• A single skill can be assigned to multiple agents

When you interact with Claude through the MCP server, the agent automatically selects the appropriate skill based on your request. You can also invoke a specific skill by name if you want precise control.

The typical flow:

1. You describe what you need to Claude
2. The agent identifies the best matching skill from its assigned set
3. The agent prompts you for any required parameters (or fills them from context)
4. The skill executes and returns the result

Agents with well-organized skills become powerful assistants that handle complex workflows without manual prompt engineering.

Managing skills

You can manage skills through two interfaces, and they stay in sync automatically.

Dashboard -- The visual editor at systemprompt.io lets you create, edit, test, and publish skills from your browser. The dashboard gives you a form-based interface for defining instructions, parameters, and secrets. It is the best starting point for new users.

Claude via MCP -- The Skill Manager connector lets Claude create and manage skills directly through conversation. Ask Claude to "create a new skill for writing release notes" and it will walk you through defining the instructions, parameters, and configuration. This is faster for experienced users who prefer a conversational workflow.

Changes made in either interface are reflected in the other immediately. Edit a skill in the dashboard and it is available in Claude within seconds.

Choosing the right approach

If you want to...	Start here
Build your first skill	Creating Skills
Share a skill with your team or the community	Publishing Skills
Connect a skill to an external API	Skill Secrets
Browse skills others have built	Browse Plugins

Next steps

• Creating Skills -- Step-by-step guide to building your first skill from scratch
• Publishing Skills -- Share your skills with your team or the marketplace
• Skill Secrets -- Securely store and use API keys and credentials in your skills