skills/
├── google-calendar/
│   ├── SKILL.md          # Documentation for Claude
│   ├── list-events.js    # Tool implementation
│   ├── create-event.js   # Tool implementation
│   └── package.json      # Dependencies
├── spotify/
│   ├── SKILL.md
│   ├── play.sh
│   └── current-track.sh
└── custom-api/
    ├── SKILL.md
    └── query.py

ls ~/openclaw-workspace/skills/

cd ~/openclaw-workspace/skills/
git clone https://github.com/openclaw-community/skill-weather weather

mkdir -p ~/openclaw-workspace/skills/my-api

mkdir -p ~/openclaw-workspace/skills/my-database
cd ~/openclaw-workspace/skills/my-database

#!/usr/bin/env python3
import sys
import json
import psycopg2
import os

conn = psycopg2.connect(os.environ["DATABASE_URL"])
query = sys.argv[1]

try:
    cursor = conn.cursor()
    cursor.execute(query)
    results = cursor.fetchall()
    columns = [desc[0] for desc in cursor.description]
    rows = [dict(zip(columns, row)) for row in results]
    print(json.dumps(rows, indent=2, default=str))
except Exception as e:
    print(json.dumps({"error": str(e)}))
finally:
    conn.close()

chmod +x query.py
pip install psycopg2-binary

# my-database Skill

## Overview
Query the production PostgreSQL database using read-only SQL.

## Tools

### query.py
Execute a SQL SELECT query against the database.

**Usage:**
```bash
python ~/openclaw-workspace/skills/my-database/query.py "SELECT * FROM users LIMIT 10"

That's it. Your database skill is ready. Next session, Claude will read this SKILL.md and know exactly how to query your database by calling `query.py`.

## Step 4: Advanced Skill Features

![CLI commands showing advanced skill configuration and tool registration](/images/terminal-ai-commands.jpg)
*Advanced skills include state management, authentication handling, and multi-tool workflows*

More sophisticated skills include:

**State management** — Skills can write state files to maintain context between tool calls. A shopping skill might track a cart in `skills/shopping/state/cart.json`.

**Authentication** — Skills that require OAuth (Google, Notion, Spotify) include auth setup scripts. The SKILL.md documents the one-time auth flow.

**Multi-tool skills** — Complex integrations expose multiple tools. A GitHub skill might have separate scripts for `create-issue.js`, `list-prs.js`, `merge-pr.js` — each documented separately in SKILL.md.

**Streaming output** — For tools that produce continuous output (logs, real-time data), skills can stream results to a file that Claude monitors.

For managing tool behavior across sessions, see the [OpenClaw CLI commands reference](/openclaw-cli-commands-reference) and [OpenClaw heartbeat system guide](/openclaw-heartbeat-system-guide).

## Real-World Skill Examples

Skills solve problems that built-in tools can't. Here are patterns from the community:

- **Home automation skill** — Wraps Home Assistant's REST API to control lights, thermostat, locks from Claude
- **Analytics skill** — Queries Google Analytics or Posthog and formats results as readable reports
- **Customer database skill** — Lets Claude look up customer records, order history, support tickets
- **CI/CD skill** — Triggers GitHub Actions workflows, reads build logs, reports deployment status
- **Trading skill** — Queries portfolio data, calculates P&L, checks market prices

The pattern is always the same: a script that calls an API or system, a SKILL.md that explains how Claude should call that script.

## Frequently Asked Questions

![FAQ section for OpenClaw skills system](/images/ai-assistant-faq.jpg)
*Common questions about creating and managing OpenClaw skills*

**What languages can skills be written in?**
Any language that produces executable output — Python, JavaScript, Bash, Go, Ruby. Whatever runs on your system.

**How does Claude know which skills are available?**
OpenClaw loads all `SKILL.md` files from the skills directory into the system context at session start. Claude reads them and learns the available tools.

**Can skills call other skills?**
Yes. A skill script can call other skill scripts directly. Be careful about circular dependencies and excessive nesting.

**How do I handle API keys securely?**
Store API keys in environment variables (e.g., in `~/.bashrc` or a `.env` file loaded at startup). Never hardcode credentials in skill scripts or SKILL.md files.

**Can I share skills with other OpenClaw users?**
Yes. The OpenClaw community shares skills as GitHub repositories. Follow the standard structure (directory + SKILL.md + executable scripts) and others can install it with a git clone.

## Conclusion

![OpenClaw conclusion showing extended AI assistant with custom skills deployed](/images/cli-conclusion.jpg)
*The skills system transforms OpenClaw from a general-purpose AI into a personalized assistant tuned to your exact environment*

The skills system is what turns OpenClaw from a general-purpose AI assistant into a personalized one. By teaching Claude the tools specific to your environment — your databases, your APIs, your automation systems — you create an assistant that can actually act in your world, not just talk about it.

Start with a simple skill: wrap one API endpoint, write a minimal SKILL.md, and test it. Once you see how quickly it integrates, you'll find yourself building skills for everything.

For further reading, see the [OpenClaw file tools tutorial](/openclaw-file-tools-tutorial) for understanding built-in tool capabilities, and the [multi-agent AI systems guide](/multi-agent-ai-systems-guide) for orchestrating complex workflows across skills. The [MCP protocol tutorial](/mcp-model-context-protocol-tutorial) also covers a standardized approach to tool integration that complements the skills system.