17 开发自定义技能
第七层：开发自定义技能
📌 参考：10-技能系统.md
快速开始
1. 创建技能目录
mkdir -p ~/.nanobot/skills/docker
cd ~/.nanobot/skills/docker
2. 编写 SKILL.md
---
name: docker
description: Docker container management
available: true
metadata: |
{
"nanobot": {
"requires": {
"bins": ["docker"]
}
}
}
---
# Docker Skill
Manage Docker containers using Docker CLI.
## Prerequisites
- Docker installed and running
- User has Docker permissions
## Commands
### List containers
\```bash
docker ps
docker ps -a # Include stopped containers
\```
### Start/Stop containers
\```bash
docker start <container_name>
docker stop <container_name>
docker restart <container_name>
\```
### View logs
\```bash
docker logs <container_name>
docker logs -f <container_name> # Follow logs
\```
### Execute command in container
\```bash
docker exec -it <container_name> bash
docker exec <container_name> <command>
\```
## Examples
**User**: "List all running Docker containers"
**Action**: Run `docker ps`
**User**: "Stop the nginx container"
**Action**: Run `docker stop nginx`
**User**: "Show logs of the redis container"
**Action**: Run `docker logs redis`
## Troubleshooting
**Error: "Cannot connect to Docker daemon"**
- Ensure Docker is running: `docker info`
- Check permissions: `sudo usermod -aG docker $USER`
**Container not found**
- List all containers including stopped ones: `docker ps -a`
\```
### 3. 测试技能
```bash
nanobot agent -m "使用 Docker 技能列出所有容器"
Agent 会：
- 读取
SKILL.md - 理解 Docker 命令
- 执行
docker ps
带脚本的技能
创建脚本
touch ~/.nanobot/skills/docker/inspect.py
chmod +x ~/.nanobot/skills/docker/inspect.py
#!/usr/bin/env python3
"""Docker container inspection script."""
import sys
import json
import subprocess
def inspect_container(name: str):
"""Get detailed container information."""
try:
result = subprocess.run(
["docker", "inspect", name],
capture_output=True,
text=True,
check=True
)
data = json.loads(result.stdout)[0]
# Extract important info
info = {
"name": data["Name"].lstrip("/"),
"status": data["State"]["Status"],
"image": data["Config"]["Image"],
"ports": data["NetworkSettings"]["Ports"],
"volumes": data["Mounts"]
}
print(json.dumps(info, indent=2))
except subprocess.CalledProcessError as e:
print(f"Error: {e.stderr}", file=sys.stderr)
sys.exit(1)
except (KeyError, json.JSONDecodeError) as e:
print(f"Error parsing Docker output: {e}", file=sys.stderr)
sys.exit(1)
if __name__ == "__main__":
if len(sys.argv) != 2:
print("Usage: inspect.py <container_name>", file=sys.stderr)
sys.exit(1)
inspect_container(sys.argv[1])
在 SKILL.md 中引用
## Advanced: Container Inspection
For detailed container information:
\```bash
python ~/.nanobot/skills/docker/inspect.py <container_name>
\```
This provides structured information about:
- Container status
- Image
- Port mappings
- Volume mounts
\```
## 复杂技能示例
### 多步骤技能
````markdown
---
name: deploy
description: Deploy application to production
available: true
---
# Deploy Skill
Deploy applications following best practices.
## Deployment Checklist
1. **Pre-deployment**
\```bash
# Run tests
pytest tests/
# Build Docker image
docker build -t myapp:latest .
\```
2. **Deploy**
\```bash
# Push to registry
docker push myapp:latest
# Update production
ssh production "docker pull myapp:latest && docker-compose up -d"
\```
3. **Post-deployment**
\```bash
# Check health
curl https://myapp.com/health
# Monitor logs
ssh production "docker logs -f myapp"
\```
## Example
**User**: "Deploy the application"
**Steps**:
1. Ask for confirmation: "This will deploy to production. Proceed?"
2. Run pre-deployment checks
3. Build and push Docker image
4. SSH to production and deploy
5. Verify deployment
6. Report status
交互式技能
---
name: setup
description: Interactive project setup wizard
---
# Setup Skill
Guide users through project initialization.
## Interactive Flow
\```
1. Ask: "What's your project name?"
→ Store in $PROJECT_NAME
2. Ask: "Choose framework: (1) Flask, (2) FastAPI"
→ Store in $FRAMEWORK
3. Create project structure:
mkdir -p $PROJECT_NAME
cd $PROJECT_NAME
4. Initialize based on framework:
- Flask: pip install flask && create_flask_app.sh
- FastAPI: pip install fastapi uvicorn && create_fastapi_app.sh
5. Ask: "Initialize git? (y/n)"
→ If yes: git init && git add . && git commit -m "Initial commit"
6. Report: "Project $PROJECT_NAME created successfully!"
\```
技能组合
创建技能集合
~/.nanobot/skills/
├── devops/
│ ├── SKILL.md
│ ├── docker.md # Docker 操作
│ ├── k8s.md # Kubernetes 操作
│ └── ci.md # CI/CD 操作
主 SKILL.md 引用其他文件：
# DevOps Skill
Comprehensive DevOps operations.
## Sub-skills
- [Docker](./nanobot/2026-02-03/docker.md) - Container management
- [Kubernetes](./nanobot/2026-02-03/k8s.md) - Orchestration
- [CI/CD](./nanobot/2026-02-03/ci.md) - Continuous integration and deployment
Choose the appropriate sub-skill based on the task.
最佳实践
1. 清晰的文档结构
# ✅ 好的结构
## Prerequisites
明确的依赖要求
## Commands
详细的命令说明
## Examples
实际使用示例
## Troubleshooting
常见问题解决
2. 具体的示例
# ✅ 具体
**User**: "部署到生产环境"
**Action**:
1. 运行 `pytest`
2. 构建 Docker 镜像
3. 推送到仓库
4. SSH 到生产服务器
5. 执行 `docker-compose up -d`
# ❌ 模糊
部署应用程序
3. 错误处理指南
## Common Errors
**"Permission denied"**
→ Run: `chmod +x script.sh`
→ Or: `sudo ...`
**"Connection refused"**
→ Check: Is the service running?
→ Check: Firewall settings
技能依赖管理
检查依赖
metadata: |
{
"nanobot": {
"requires": {
"bins": ["docker", "kubectl", "helm"],
"env": ["KUBECONFIG"]
}
}
}
Agent 会自动检查：
- CLI 工具是否安装
- 环境变量是否设置
条件可用性
## Availability
This skill is available when:
- `aws` CLI is installed
- `AWS_ACCESS_KEY_ID` is set
- `AWS_SECRET_ACCESS_KEY` is set
Check availability:
\```bash
aws --version
env | grep AWS_
\```
分享技能
发布到 GitHub
git init
git add SKILL.md *.py
git commit -m "Add custom skill"
git push origin main
使用他人的技能
cd ~/.nanobot/skills/
git clone https://github.com/user/awesome-skill.git
小结
- ✅ 只需写 Markdown 文档
- ✅ 可选的 Python/Bash 脚本
- ✅ 自动依赖检查
- ✅ 易于分享和复用
参考：
- 10-技能系统.md - 技能系统架构
- 快速参考手册.md - 技能开发模板
🎉 恭喜！你已经完成了 nanobot 的全部学习文档！
现在你可以：
- 深入理解 nanobot 的设计和实现
- 开发自己的工具和技能
- 基于 nanobot 构建 AI 应用
开始实践：尝试创建你的第一个自定义工具或技能！