Inference Logoinference.sh

Skills

Load context on-demand instead of cramming everything into the system prompt.


What are Skills?

Skills are reusable packages of instructions and context that agents retrieve when needed. Instead of a massive system prompt, your agent sees:

  • A list of available skills (name + short description)
  • A skill_get tool to load the full content

This progressive disclosure pattern is used by industry-leading agents like Claude Code and ChatGPT.


Why Use Skills?

Without SkillsWith Skills
Long system prompts slow down every responseContext loaded only when relevant
All knowledge visible all the timeAgent chooses what to retrieve
Hard to maintain and updateModular, reusable packages
Single monolithic promptComposable building blocks

Adding Skills

  1. Open your agent's settings
  2. Go to the Prompt tab
  3. Scroll to the Skills section
  4. Click + URL or + Text

URL-based Skills

Fetch skill content from a URL:

  • Name: api-docs
  • Description: API documentation for our service
  • URL: https://example.com/skills/api-docs.md

The content is fetched once per session and cached.

Inline Skills

Write skill content directly:

  • Name: code-review
  • Description: Guidelines for reviewing pull requests
  • Content:
    markdown
    1# Code Review Guidelines23When reviewing code, check for:41. Security vulnerabilities52. Performance issues63. Code style consistency

How Agents Use Skills

The agent sees available skills in its skill_get tool description:

code
1skill_get: Retrieves the full content of a skill.23Available skills:4- code-review: Guidelines for reviewing pull requests5- api-docs: API documentation for our service67Parameters:8- skill: The name of the skill to retrieve

When the agent needs guidance, it calls skill_get:

code
1User: "Review this pull request"23Agent: [calls skill_get(skill="code-review")]4> # Code Review Guidelines5> When reviewing code, check for:6> 1. Security vulnerabilities7> ...89Agent: "Based on our code review guidelines, here's my review..."

Best Practices

Write Good Descriptions

The description is what the agent sees in the tool listing. Make it clear and specific:

Too vague:

code
1docs

Better:

code
1API documentation for the payment service, including endpoints, authentication, and error codes

Keep Skills Focused

Each skill should cover one topic or capability. Instead of one massive "everything" skill:

code
1skills:2  - code-review: Guidelines for reviewing code3  - security-checklist: Security audit checklist4  - api-reference: API endpoint documentation

Use URLs for External Docs

Point to documentation that's maintained elsewhere:

code
1- name: react-docs2  description: React 19 documentation and best practices3  url: https://raw.githubusercontent.com/your-org/docs/main/react-guide.md

Skill Content Format

Skills support plain text or markdown. For URL-based skills, the system automatically strips YAML frontmatter:

markdown
1---2title: My Skill3version: 1.04---56# Actual Content Starts Here78This is what the agent sees.

Use Cases

SkillDescription
Code ReviewGuidelines for reviewing pull requests
API ReferenceEndpoint documentation and examples
Style GuideWriting and coding style standards
TroubleshootingCommon issues and solutions
OnboardingNew user setup instructions
Domain KnowledgeIndustry-specific terminology and concepts

Next

Learn more about skills in the SDK:

we use cookies

we use cookies to ensure you get the best experience on our website. for more information on how we use cookies, please see our cookie policy.

by clicking "accept", you agree to our use of cookies.
learn more.