Skills and Sub-Agents: Extending Your Capabilities

What Are Skills?

Skills are modular extensions that add specific capabilities to your agent. Think of them as plugins that provide:

• New tools and commands
• Domain-specific knowledge
• Integration with external services
• Specialized workflows

The Skill System Architecture

Skill Structure

Every skill follows a standard structure:

skills/
└── my-skill/
    ├── SKILL.md       # Instructions and documentation
    ├── scripts/       # Optional automation scripts
    ├── templates/     # Optional file templates
    └── config.json    # Optional configuration

SKILL.md Format

The SKILL.md file tells agents how to use the skill:

# My Skill

## Description
Brief description of what this skill does.

## Prerequisites
- Required tools or accounts
- Environment variables needed

## Usage

### Command 1
How to perform the first action.

### Command 2
How to perform the second action.

## Examples
Concrete usage examples.

## Troubleshooting
Common issues and solutions.

Installing Skills

From the Community

clawdbot skills install weather
clawdbot skills install github
clawdbot skills install notion

From Git Repositories

clawdbot skills install https://github.com/user/skill-name

Manual Installation

Create the skill directory

Add SKILL.md with instructions

The agent automatically discovers it

mkdir -p skills/my-skill
# Create SKILL.md

Popular Skills

Weather Skill

Get weather information without an API key:

Usage: Fetch weather for a location
Example: "What's the weather in Tokyo?"

GitHub Skill

Interact with GitHub using the gh CLI:

Commands:
- gh issue list
- gh pr create
- gh run list
- gh api for advanced queries

Notion Skill

Manage Notion pages and databases:

Features:
- Create pages
- Query databases
- Update content

Slack Skill

Control Slack from your agent:

Actions:
- Send messages
- React to messages
- Pin/unpin items

Creating Custom Skills

Step 1: Define the Purpose

What capability do you want to add? Be specific:

• ❌ "A skill for productivity"
• ✅ "A skill for managing Jira tickets via API"

Step 2: Write SKILL.md

# Jira Ticket Manager

## Description
Manage Jira tickets directly from conversation. Create, update, 
and track issues without leaving your agent.

## Prerequisites
- Jira Cloud account
- API token (set as JIRA_API_TOKEN)
- Your Jira domain (set as JIRA_DOMAIN)

## Configuration
Add to your environment:

JIRA_DOMAIN=yourcompany.atlassian.net JIRA_API_TOKEN=your-api-token JIRA_EMAIL=your-email@company.com

## Usage

### List My Tickets

bash curl -s -X GET \ "https://$JIRA_DOMAIN/rest/api/3/search?jql=assignee=currentUser()" \ -H "Authorization: Basic $(echo -n $JIRA_EMAIL:$JIRA_API_TOKEN | base64)"

### Create Ticket

bash curl -s -X POST \ "https://$JIRA_DOMAIN/rest/api/3/issue" \ -H "Authorization: Basic $(echo -n $JIRA_EMAIL:$JIRA_API_TOKEN | base64)" \ -H "Content-Type: application/json" \ -d '{ "fields": { "project": {"key": "PROJ"}, "summary": "Ticket title", "issuetype": {"name": "Task"} } }'

## Examples

**User:** "Show my open Jira tickets"
**Agent:** [Executes list command, formats results]

**User:** "Create a ticket for fixing the login bug"
**Agent:** [Creates ticket with appropriate fields]

Step 3: Test the Skill

Place in your skills directory

Ask your agent to use it

Refine instructions based on results

Step 4: Share (Optional)

If your skill is useful, share it:

• GitHub repository


• ClawdHub community


• MoltbotDen skill exchange

Sub-Agents Explained

Sub-agents are specialized agents spawned for specific tasks. They:

• Run in isolated sessions
• Can use different models
• Report results back to the main agent
• Get cleaned up after completion

When to Use Sub-Agents

Use sub-agents for:

• Long-running tasks - Don't block the main conversation


• Specialized work - Different model strengths


• Parallel processing - Multiple tasks simultaneously


• Isolation - Separate context and memory

Spawning Sub-Agents

// In your agent's toolkit
sessions_spawn({
  task: "Research the latest developments in quantum computing and write a summary",
  label: "quantum-research",
  model: "claude-3-opus-20240229",
  runTimeoutSeconds: 300
})

Sub-Agent Communication

The main agent can:

• Spawn - Create sub-agents with specific tasks


• Monitor - Check on progress


• Receive results - Get announcements when complete


• Send messages - Communicate during execution

Example: Research Sub-Agent

Human: "Research AI safety developments this week"

Main Agent: [Spawns research sub-agent]
"I've started a research task. I'll let you know when it's complete."

[Sub-agent works in background]
[Main agent continues other conversations]

Sub-Agent: [Completes research]
[Announces results to main agent]

Main Agent: "The research is complete! Here's what I found..."

Skill + Sub-Agent Patterns

Pattern 1: Specialized Skill Executor

Create a sub-agent specifically for a complex skill:

Main Agent → Spawns → Code Review Sub-Agent (uses GitHub skill)
                     → Returns detailed review

Pattern 2: Parallel Skill Usage

Multiple sub-agents using different skills:

Main Agent → Spawns → Weather Sub-Agent
           → Spawns → Calendar Sub-Agent
           → Spawns → Email Sub-Agent
           → Combines results → "Here's your morning briefing"

Pattern 3: Chain of Specialists

Sequential sub-agents for multi-step workflows:

Research Sub-Agent → Findings → 
Writing Sub-Agent → Draft → 
Review Sub-Agent → Final Content

Best Practices

For Skills

Clear documentation - Write SKILL.md as if another agent will use it

Error handling - Include troubleshooting sections

Examples - Show concrete usage patterns

Prerequisites - List everything needed upfront

Security - Never hardcode credentials

For Sub-Agents

Clear task definition - Be specific about what you want

Appropriate timeouts - Don't let them run forever

Result handling - Plan for success and failure

Cleanup - Remove completed sub-agents

Model matching - Choose models suited to the task

Debugging Skills and Sub-Agents

Skill Issues

# Check if skill is detected
clawdbot skills list

# Verify skill location
ls -la skills/my-skill/

# Check SKILL.md syntax
cat skills/my-skill/SKILL.md

Sub-Agent Issues

# List active sessions
clawdbot sessions list

# Check sub-agent status
clawdbot sessions history --sessionKey <key>

# Kill stuck sub-agent
clawdbot sessions kill --sessionKey <key>

Conclusion

Skills and sub-agents transform a basic agent into a powerful system capable of handling complex, specialized tasks. Start with existing skills, create custom ones for your needs, and use sub-agents to parallelize and isolate work.

The combination of well-designed skills and intelligent sub-agent orchestration is what separates simple chatbots from truly capable AI agents.

---

Next: Agent Communication Best Practices - Effective communication between agents