Creating Skills

This guide walks you through creating a skill from scratch using the systemprompt.io dashboard. By the end, you will have a working skill that you can test, refine, and assign to an agent.

Prerequisites

• A systemprompt.io account — register here
• The MCP server connected (see Quick Start)

Step 1: Open the skill editor

1. Log in to your systemprompt.io dashboard
2. Navigate to Skills in the sidebar
3. Click Create Skill

The skill editor opens with a blank form ready for you to fill in.

Step 2: Set the basics

Give your skill a name and description. These are visible in the dashboard and marketplace, so make them clear and specific.

• Name — A short, descriptive title. Example: "Write Blog Post"
• Description — One or two sentences explaining what the skill does. Example: "Writes a structured blog post on a given topic, tailored to a specific audience and word count."

Step 3: Write the instructions

The instructions field is where you define exactly what the AI should do. Write in plain language. Be specific about the output format, tone, and any rules the AI should follow.

Here is an example for a "Write Blog Post" skill:

Write a blog post about the provided topic.

Rules:
- Target the specified audience
- Match the requested word count (within 10%)
- Use a conversational, professional tone
- Include an introduction, 3-5 subheadings with body paragraphs, and a conclusion
- Do not use jargon unless the audience is technical
- Format the output in Markdown

Structure:
1. Start with a compelling hook
2. Break the topic into logical sections
3. End with a clear call to action or summary

Good instructions share a few traits: they state the goal, define constraints, specify the output format, and set the tone. The more specific you are, the more consistent the results.

Step 4: Add parameters

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

Click Add Parameter and configure each one:

Field	Description
Name	The parameter identifier (e.g., topic)
Description	What the parameter expects (e.g., "The subject of the blog post")
Type	The data type — usually string, but can be number, boolean, or enum
Required	Whether the parameter must be provided every time the skill runs

For the "Write Blog Post" example, you would add three parameters:

1. topic (string, required) — "The subject of the blog post"
2. audience (string, required) — "The target reader, e.g. 'small business owners' or 'software engineers'"
3. word_count (number, optional) — "Target word count. Defaults to 800 if not specified."

Parameters are automatically injected into the prompt at runtime. Reference them in your instructions by name and the platform handles the substitution.

Step 5: Test your skill

Before assigning the skill to an agent, test it directly.

1. Click Test in the skill editor
2. Fill in the parameter values (e.g., topic: "Remote Work Productivity", audience: "Managers", word_count: 1000)
3. Click Run
4. Review the output

If the result is not what you expected, go back and adjust the instructions. Common fixes:

• Output too long or short — Add an explicit word count instruction
• Wrong tone — Add tone guidance (e.g., "Write in a casual, friendly voice")
• Missing structure — Define the exact sections you want
• Too generic — Add more specific constraints or examples

Step 6: Iterate and refine

Good skills rarely come out perfect on the first try. Run several tests with different parameter values and refine your instructions based on the results.

Tips for iteration:

• Test edge cases (very short topics, unusual audiences)
• Add negative instructions for things the AI should avoid ("Do not include disclaimers")
• If a section is consistently weak, add more detail to that part of the instructions
• Keep instructions under 500 words — shorter instructions tend to produce more focused results

Step 7: Assign to an agent

Once you are satisfied with the skill:

1. Go to Agents in the sidebar
2. Select the agent you want to use (or create a new one)
3. Click Add Skill and select your new skill
4. Save the agent

The skill is now live. When you interact with that agent through Claude, it can use the skill to handle matching requests.

Example: complete "Write Blog Post" skill

Here is what the finished skill looks like:

• Name: Write Blog Post
• Description: Writes a structured blog post on a given topic, tailored to a specific audience and word count.
• Instructions: (as shown in Step 3 above)
• Parameters:
  • topic (string, required)
  • audience (string, required)
  • word_count (number, optional, default: 800)

Creating skills with Claude

You can also create skills entirely through conversation. With the MCP server connected, ask Claude:

Create a new skill called "Write Blog Post" that writes structured blog posts.
It should accept a topic, target audience, and optional word count.

Claude will create the skill, set up the parameters, and save it to your account. You can then refine it in the dashboard or through further conversation.

Troubleshooting

• Skill not appearing in agent — Make sure you assigned the skill to the agent and saved the changes
• Parameters not being used — Verify the parameter names in your instructions match the parameter names you defined
• Inconsistent output — Add more constraints to your instructions; vague prompts produce varied results
• Test button not working — Check that you are logged in and your session has not expired; refresh the page and try again