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

# Skills

> Create custom skills to give AI agents your organization's domain-specific knowledge — coding standards, review checklists, and operational procedures.

## What You'll Set Up

By the end of this tutorial, you'll have a custom skill that gives AI agents your organization's specific knowledge — applied automatically whenever they run tasks for the features you choose.

<Steps>
  <Step title="Navigate to Customize">
    Go to **Customize** from the sidebar menu. You'll see the **Custom Skills** section where all workspace skills are managed.
  </Step>

  <Step title="Create a Skill">
    Click **New Skill** and choose a creation method:

    * **Create with AI**: Describe what you need and let AI generate the skill
    * **Write skill instructions**: Write the skill manually using a form

    For your first skill, try **Create with AI** — describe your team's coding standards or review rules in plain language, and the agent will generate a structured SKILL.md for you.
  </Step>

  <Step title="Review the Generated Skill">
    After creation, your skill appears as a card on the Custom Skills page. Click the card to open the detail view and review the instructions.

    The detail view shows a split-pane layout with the file tree on the left and rendered markdown on the right. Make sure the instructions match your expectations.
  </Step>

  <Step title="Enable the Skill">
    Toggle the skill **on** using the switch on its card. Disabled skills are saved but not applied by agents.
  </Step>

  <Step title="Assign to Features">
    Use the feature checkboxes on the skill card to control where it applies:

    * **Code Review** — applied during PR analysis
    * **Incident** — applied during incident investigation
    * **Assessment** — applied during infrastructure assessments

    Select the features where your skill is relevant. A skill can be assigned to multiple features.
  </Step>

  <Step title="Verify It Works">
    Trigger a task for one of the assigned features — for example, open a PR if you assigned the skill to Code Review. Check the agent's output to confirm it follows your skill instructions.
  </Step>
</Steps>

***

## How It Works

```
Create skill → Enable → Assign to features → Agent loads instructions automatically
```

When an agent runs a task for an assigned feature, it loads the skill instructions alongside its default behavior. Your domain knowledge is applied consistently without repeating context in every conversation.

***

## Writing Effective Skills

A skill is a Markdown file with YAML frontmatter:

```markdown theme={null}
---
name: api-review-standards
description: Enforce API design standards during code review
---

## Naming

- Use plural nouns for collection endpoints (`/users`, not `/user`)
- Use kebab-case for multi-word paths (`/user-settings`)

## Error Handling

- Return consistent error response shape with `code`, `message`, `details`
- Use 4xx for client errors, 5xx for server errors only
```

Tips for writing skills that agents follow reliably:

* **Be specific** — concrete rules and examples work better than vague guidelines
* **Keep it focused** — one skill per domain (separate security from performance)
* **Show examples** — agents follow examples more reliably than abstract explanations
* **Every token counts** — write only what the agent doesn't already know

<Card title="Skill Format Reference" icon="file-code" href="/guide/skills/skill-format">
  Full SKILL.md structure, frontmatter reference, and best practices
</Card>

***

## Tips

* **Start with one skill**: Create a focused skill for your most common review concern and see how the agent applies it before creating more
* **Use AI to iterate**: Click **Edit with AI** on any skill to refine it through conversation
* **Test on a single feature first**: Enable a new skill on one feature and review the output before assigning it broadly
* **Check the detail view**: Always review the rendered instructions to make sure they read clearly

***

## Next Step

<Card title="Deep Response Engine Setup" icon="triangle-exclamation" href="/guide/tutorial/incident-response">
  Set up Pulse signal sources and AI-powered incident investigation
</Card>