> ## Documentation Index
> Fetch the complete documentation index at: https://klaw.sh/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Skills
> Composable capability bundles for agents
## What are Skills?
**Skills** are composable capability bundles that enhance agent functionality. They package together:
* A set of related tools
* System prompt additions
* Configuration parameters
* Optional external dependencies
Think of skills like npm packages or Kubernetes operators—reusable modules that extend the base system.
## Skill Architecture
```
┌─────────────────────────────────────────────────────┐
│ SKILL │
│ ┌───────────────────────────────────────────────┐ │
│ │ Metadata │ │
│ │ name, version, description, author │ │
│ └───────────────────────────────────────────────┘ │
│ ┌───────────────────────────────────────────────┐ │
│ │ Tools │ │
│ │ [tool1] [tool2] [tool3] │ │
│ └───────────────────────────────────────────────┘ │
│ ┌───────────────────────────────────────────────┐ │
│ │ System Prompt │ │
│ │ Instructions for using this skill │ │
│ └───────────────────────────────────────────────┘ │
│ ┌───────────────────────────────────────────────┐ │
│ │ Config │ │
│ │ Skill-specific parameters │ │
│ └───────────────────────────────────────────────┘ │
└─────────────────────────────────────────────────────┘
```
## Built-in Skills
klaw includes 10+ skills out of the box:
Search the internet and retrieve information.
**Tools:** `web_search`, `web_fetch`
Interactive web browsing and scraping.
**Tools:** `web_fetch`, `web_screenshot`
Execute code in sandboxed environments.
**Tools:** `code_run`, `code_eval`
Version control operations.
**Tools:** `git_status`, `git_commit`, `git_push`
Container management.
**Tools:** `docker_run`, `docker_build`, `docker_ps`
SQL database interactions.
**Tools:** `sql_query`, `sql_exec`
REST API interactions.
**Tools:** `http_get`, `http_post`, `http_request`
Slack messaging and management.
**Tools:** `slack_send`, `slack_read`
Email sending and reading.
**Tools:** `email_send`, `email_read`
Calendar management.
**Tools:** `calendar_create`, `calendar_list`
## Managing Skills
### List Available Skills
```bash theme={null}
klaw skill list
```
Output:
```
NAME VERSION STATUS DESCRIPTION
web-search 1.0.0 installed Search the internet
browser 1.0.0 available Web browsing capabilities
code-exec 1.0.0 installed Code execution sandbox
git 1.0.0 available Git version control
docker 1.0.0 available Docker container management
database 1.0.0 available SQL database interactions
api 1.0.0 available REST API client
slack 1.0.0 available Slack integration
email 1.0.0 available Email capabilities
calendar 1.0.0 available Calendar management
```
### Install a Skill
```bash theme={null}
klaw skill install browser
```
### Uninstall a Skill
```bash theme={null}
klaw skill uninstall browser
```
### View Skill Details
```bash theme={null}
klaw skill describe web-search
```
Output:
```yaml theme={null}
Name: web-search
Version: 1.0.0
Description: Search the internet and retrieve information
Author: each::labs
Status: installed
Tools:
- web_search: Search the web for information
- web_fetch: Fetch and process a URL
System Prompt:
You have access to web search capabilities. Use web_search
to find current information and web_fetch to read specific
pages. Always cite your sources.
Config:
max_results: 10
safe_search: true
```
## Using Skills with Agents
### At Creation Time
```bash theme={null}
klaw create agent researcher \
--model claude-sonnet-4-20250514 \
--skills web-search,browser,code-exec
```
### In Agent Definition
```toml theme={null}
# ~/.klaw/agents/researcher.toml
name = "researcher"
model = "claude-sonnet-4-20250514"
skills = ["web-search", "browser", "code-exec"]
[skill_config.web-search]
max_results = 20
[skill_config.database]
connection_string = "${DATABASE_URL}"
```
## Skill Composition
Skills can be combined to create powerful agents:
```
┌─────────────────────────────────────────────────────┐
│ FULL-STACK DEVELOPER AGENT │
│ │
│ Skills: code-exec + git + docker + database │
│ │
│ ┌─────────┐ ┌─────────┐ ┌─────────┐ ┌─────────┐ │
│ │code-exec│ │ git │ │ docker │ │database │ │
│ └────┬────┘ └────┬────┘ └────┬────┘ └────┬────┘ │
│ │ │ │ │ │
│ └───────────┴───────────┴───────────┘ │
│ │ │
│ ▼ │
│ Combined Tool Registry │
│ Merged System Prompts │
└─────────────────────────────────────────────────────┘
```
### Example: DevOps Agent
```bash theme={null}
klaw create agent devops \
--model gpt-4o \
--skills docker,git,api \
--task "Manage deployments and infrastructure"
```
### Example: Research Agent
```bash theme={null}
klaw create agent researcher \
--model claude-sonnet-4-20250514 \
--skills web-search,browser \
--task "Research topics and compile reports"
```
## Skill Definition Format
Skills are defined in TOML format:
```toml theme={null}
# skill.toml
name = "my-skill"
version = "1.0.0"
description = "My custom skill"
author = "Your Name"
# Tools this skill provides
tools = ["my_tool_1", "my_tool_2"]
# System prompt addition
system_prompt = """
You have access to my custom tools.
Use my_tool_1 for X and my_tool_2 for Y.
Always follow best practices when using these tools.
"""
# Configuration schema
[config]
api_key = { type = "string", required = true, env = "MY_API_KEY" }
timeout = { type = "integer", default = 30 }
```
## Skill Registry
Skills are stored and loaded from:
```
~/.klaw/
├── skills/
│ ├── installed.json # Installed skills registry
│ ├── web-search/ # Skill files
│ ├── browser/
│ └── custom-skill/
```
### installed.json
```json theme={null}
{
"skills": {
"web-search": {
"version": "1.0.0",
"installed_at": "2024-12-14T10:00:00Z",
"source": "builtin"
},
"custom-skill": {
"version": "1.0.0",
"installed_at": "2024-12-14T11:00:00Z",
"source": "local"
}
}
}
```
## Creating Custom Skills
### Step 1: Create Skill Directory
```bash theme={null}
mkdir -p ~/.klaw/skills/my-skill
```
### Step 2: Define the Skill
```toml theme={null}
# ~/.klaw/skills/my-skill/skill.toml
name = "my-skill"
version = "1.0.0"
description = "My custom skill for doing X"
tools = ["my_tool"]
system_prompt = """
You have access to my_tool which does X.
Use it when the user asks about Y.
"""
```
### Step 3: Implement Tools (Optional)
For custom tools, create a Go file:
```go theme={null}
// ~/.klaw/skills/my-skill/tools.go
package myskill
type MyTool struct{}
func (t *MyTool) Name() string { return "my_tool" }
// ... implement Tool interface
```
### Step 4: Install
```bash theme={null}
klaw skill install ./my-skill
```
## Remote Skills
Install skills from the marketplace:
```bash theme={null}
# From skills.sh marketplace
klaw skill install skills.sh/eachlabs/notion
# From GitHub
klaw skill install github.com/user/repo
# From URL
klaw skill install https://example.com/skill.tar.gz
```
## Skill Configuration
Configure skill behavior per-agent:
```toml theme={null}
# ~/.klaw/agents/my-agent.toml
name = "my-agent"
skills = ["database", "api"]
[skill_config.database]
connection_string = "${DATABASE_URL}"
max_connections = 10
[skill_config.api]
base_url = "https://api.example.com"
auth_header = "Bearer ${API_TOKEN}"
```
## System Prompt Composition
When multiple skills are used, their system prompts are merged:
```
┌─────────────────────────────────────────────────────┐
│ FINAL SYSTEM PROMPT │
├─────────────────────────────────────────────────────┤
│ Base Agent Prompt │
│ ───────────────────────────────────────────────────│
│ You are a helpful assistant... │
├─────────────────────────────────────────────────────┤
│ Skill: web-search │
│ ───────────────────────────────────────────────────│
│ You can search the web using web_search... │
├─────────────────────────────────────────────────────┤
│ Skill: code-exec │
│ ───────────────────────────────────────────────────│
│ You can execute code using code_run... │
├─────────────────────────────────────────────────────┤
│ Workspace Context │
│ ───────────────────────────────────────────────────│
│ [SOUL.md content] │
└─────────────────────────────────────────────────────┘
```
## Best Practices
Each skill should do one thing well. Don't create monolithic skills.
Use environment variables for API keys and secrets, never hardcode them.
Write clear system prompts that explain when and how to use each tool.
Test skills locally before installing on production agents.
## Next Steps
Learn about persistent agent context
Build your own tools and skills