Skills and Tools

Skills and tools are what transform agents from conversational interfaces into productive teammates. While agents can chat and reason, skills give them expertise and tools give them capabilities to take action.

Think of skills as what an agent knows and tools as what an agent can do.

Skills

What Are Skills?

Skills are structured knowledge packages that enhance an agent's capabilities. They typically include:

Expert prompts and instructions
Domain-specific knowledge
Best practices and methodologies
Templates and frameworks
Reference materials and examples

Types of Skills

<AccordionGroup> <Accordion title="Built-in Skills"> Pre-installed skills that ship with LobeHub:

- **Artifacts Skill**: Enables creation of interactive content, visualizations, and documents
- **Core reasoning skills**: Enhance problem-solving and analysis

Built-in skills are maintained by the LobeHub team and automatically updated.

</Accordion> <Accordion title="Market Skills"> Community-created skills available from the LobeHub marketplace:

- Code review methodologies
- Writing style guides
- Industry-specific frameworks
- Language and translation guides

Browse, install, and share skills with the community.

</Accordion> <Accordion title="User Skills"> Custom skills you create for your specific needs:

- Company-specific knowledge
- Personal methodologies
- Team guidelines
- Project templates

User skills give you complete control over agent expertise.

</Accordion> </AccordionGroup>

Using Skills

Attach skills to agents to enhance their expertise:

Open Agent Settings
Go to the Skills tab
Browse and select skills to attach
Skills are automatically injected into agent context

Skills are injected at runtime, which means agents always use the latest version of the skill.

Tools

What Are Tools?

Tools are functions that agents can call to perform actions. When you ask an agent to do something beyond conversation, it uses tools:

"Search for the latest news on AI" → Web Search tool
"Create an image of a sunset" → Image Generation tool
"Save this to my notes" → File Management tool
"Remember that I prefer dark mode" → Memory Management tool

How Tools Work

<Steps> ### Agent Receives Request

You ask the agent to perform an action.

Tool Selection

The agent determines which tool(s) to use based on your request. This is automatic — you don't need to tell the agent which tool to use.

Function Calling

The agent calls the tool with appropriate parameters.

Tool Execution

The tool performs the action and returns results.

Response Generation

The agent incorporates tool results into its response. </Steps>

Built-in Tools

LobeHub includes several powerful built-in tools:

Web Search — Search the internet for current information, news, and data
Image Generation — Create images using DALL-E, Midjourney, or other providers
Memory Management — Store and retrieve information from personal memory
Page Agent Tools — Initialize and edit documents in the page editor
Agent Builder Tools — Create and configure new agents
Knowledge Base — Search and read from uploaded documents

Plugin System

Extend agent capabilities with plugins from the LobeHub marketplace:

Shopping tools: Product search and comparison
Video tools: YouTube transcript extraction
Weather: Current weather information
SEO Assistant: Keyword research and optimization
And many more community-contributed plugins

MCP (Model Context Protocol)

MCP is LobeHub's system for connecting agents to external tools and services — databases, APIs, file systems, development tools, and business applications.

See MCP Marketplace for details on installing and using MCP plugins.

Tool Configuration

Configure which tools are available to each agent:

<AccordionGroup> <Accordion title="Enable/Disable Tools"> In agent settings, toggle tools on or off:

- Enable only tools the agent needs
- Too many tools can slow responses
- Some tools have API costs

</Accordion> <Accordion title="Tool Parameters"> Configure tool behavior:

- **Web Search**: Number of results, search region
- **Image Generation**: Model, size, quality
- **Memory**: Retention period, categories

</Accordion> <Accordion title="Tool Permissions"> Some tools require permissions:

- File access
- API keys
- External service credentials

Review and approve before enabling.

</Accordion> </AccordionGroup>

Tool Chaining

Agents can use multiple tools in sequence to handle complex requests:

<Steps> ### User Request

"Find recent research papers on quantum computing and summarize the key findings"

Web Search Tool

Searches for research papers

Knowledge Base Tool

Retrieves relevant documents from uploaded files

Response

Agent synthesizes a summary from all sources </Steps>

Enable Chain of Thought to see which tools the agent selects, what parameters it uses, and the results at each step.

Best Practices

For Skills

Keep skills focused — Each skill should address one area of expertise
Use clear, structured content — Headings, lists, examples, and step-by-step instructions
Update regularly — Review skills periodically and remove outdated information
Share valuable skills — Publish to the marketplace to help the community

For Tools

Enable only necessary tools — More tools = slower responses and potential confusion
Configure tool parameters — Set sensible defaults for search limits, quality settings, etc.
Test tools before production — Verify tools work as expected with correct permissions

Troubleshooting

<AccordionGroup> <Accordion title="Skill Not Appearing in Agent"> - Verify skill is attached in agent settings - Check skill content is not empty - Restart the agent session </Accordion> <Accordion title="Tool Not Being Used"> - Confirm tool is enabled in agent settings - Make request more explicit: "Use web search to find..." - Check tool permissions are granted - Verify API keys are configured </Accordion> <Accordion title="Tool Errors"> - Review error message in Chain of Thought - Check API quotas and rate limits - Verify authentication credentials - Test tool configuration </Accordion> </AccordionGroup> <Cards> <Card href={'/docs/usage/community/mcp-market'} title={'MCP Marketplace'} />