> ## Documentation Index
> Fetch the complete documentation index at: https://docs.zerotwo.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Create a Skill

> Build a custom skill with your own trigger, system prompt, tools, and model — invokable via @skill-name in any chat.

Custom skills let you package a repeatable task into a reusable `@trigger`. Write the instructions once, and invoke the skill with a single `@` in any future chat.

## Steps to Create a Skill

<Steps>
  <Step title="Open Skills settings">
    Go to **Settings → Skills** and click **Create New** or the **+** button.
  </Step>

  <Step title="Set the skill name (trigger)">
    Enter the skill's name — this becomes the `@trigger` used to invoke it in chat. Use lowercase and hyphens for multi-word names.

    Examples: `summarize`, `code-review`, `translate-to-spanish`, `extract-action-items`
  </Step>

  <Step title="Write a description">
    A short description helps you remember what the skill does and appears in the `@` picker tooltip.

    Example: "Summarizes text into 3–5 concise bullet points focused on key facts."
  </Step>

  <Step title="Write the system prompt">
    The system prompt is the core instruction set for the skill. It defines exactly how the skill processes your input. See examples below.
  </Step>

  <Step title="Enable tools (optional)">
    Toggle on any tools the skill needs: web search, canvas, deep research, file access, etc.
  </Step>

  <Step title="Select a base model">
    Choose the model that should power this skill. For format-sensitive tasks (summarization, translation), a reliable balanced model works well. For code tasks, a coding-optimized model may perform better.
  </Step>

  <Step title="Add example inputs/outputs (optional)">
    You can add one or two examples of ideal input/output pairs in the system prompt — this few-shot guidance significantly improves consistency.
  </Step>

  <Step title="Save">
    Click **Save**. The skill is immediately available via `@skill-name` in all your chats.
  </Step>
</Steps>

## System Prompt Examples

### @summarize

```
Summarize the provided text in 3–5 bullet points.
- Focus on key facts, main ideas, and important conclusions
- Start each bullet with an action verb or key concept
- Keep each bullet under 25 words
- Do not add information not present in the original text
```

### @explain-code

```
Explain the provided code clearly and concisely for a developer audience.

Structure your response as:
**Overview**: What the code does in 1–2 sentences
**How it works**: Step-by-step explanation of the logic
**Potential issues**: Any bugs, edge cases, or improvement opportunities

Be specific and technical. Do not oversimplify.
```

### @translate-to-spanish

```
Translate the provided text to Spanish.
- Maintain the original tone and formality level
- Preserve proper nouns, brand names, and technical terms unchanged (unless they have standard Spanish equivalents)
- Return only the translation — no explanations, notes, or the original text
```

### @extract-action-items

```
Extract all action items from the provided text (meeting notes, emails, messages, etc.).

Format as a numbered list where each item includes:
- The action to be taken (specific and actionable)
- The owner (if mentioned)
- The deadline or timeline (if mentioned)

If no owner or deadline is mentioned for an item, omit those fields.
Example:
1. Draft the Q2 proposal — Sarah — due Friday
2. Schedule kickoff meeting — unassigned
```

### @tone-formal

```
Rewrite the provided text in a formal, professional tone appropriate for business communication.
- Maintain all the original meaning and information
- Use complete sentences and avoid contractions
- Remove casual language, slang, or overly informal phrasing
- Return only the rewritten text — no commentary or explanation
```

## Tips for Effective Skill Prompts

**Focus on one task.** Skills work best when they do one thing well. Don't try to combine summarization + translation + sentiment analysis in a single skill — create separate skills for each.

**Specify the output format explicitly.** The more precisely you describe the output format in the system prompt, the more consistently the skill produces it. Use examples, numbering, and structure descriptions.

**Add "return only..." instructions.** Telling the skill to return only the result (no preamble, no "Sure! Here is...") produces cleaner output.

**Use few-shot examples for complex formats.** If your desired output has a specific structure, include an example directly in the system prompt to show the exact format.

**Keep prompts under 500 words.** Skill prompts don't need to be long. Concise, specific instructions work better than lengthy general ones.