new-agent

X-McKay 的头像作者 X-McKay

从模板创建新的 AI 代理。用于向集群添加新代理、启动新的监控或自动化项目。

📖 中文导读:从模板创建新的 AI 代理。用于向集群添加新代理、启动新的监控或自动化项目。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: new-agent description: Create a new AI agent from template using kubani-dev CLI. Use when adding a new agent to the cluster, starting a new monitoring or automation project.

Create New AI Agent

Scaffold a new AI agent using kubani-dev CLI or manual setup.

# Create agent with default template
kubani-dev new my-agent

# Create with federated template (includes Sentinel, Healer, Explorer)
kubani-dev new my-agent --template federated

# Create minimal agent
kubani-dev new my-agent --template minimal

Arguments

  • agent-name: Name for the new agent (lowercase, hyphenated)
  • --template: Template type (default, federated, minimal)

Templates

Default Template

Standard agent with:

  • Temporal worker setup
  • Basic activities and workflows
  • Test structure
  • GitOps manifests

Federated Template

Advanced agent with:

  • Federated agents (Sentinel, Healer, Explorer)
  • Triage Graph workflow
  • A2A communication
  • Continuous learning integration

Minimal Template

Lightweight agent with:

  • Basic worker
  • Single activity
  • Minimal dependencies

Manual Creation

If you need to create an agent manually:

1. Create Directory Structure

cd /home/al/git/kubani
AGENT_NAME="my-agent"

mkdir -p agents/${AGENT_NAME}/{src/${AGENT_NAME//-/_},tests}

2. Create pyproject.toml

[project]
name = "${AGENT_NAME}"
version = "0.1.0"
description = "Description of what this agent does"
readme = "README.md"
requires-python = ">=3.11"
dependencies = [
    "strands-agents>=1.20.0",
    "temporalio>=1.7.0",
    "httpx>=0.27.0",
    "pydantic>=2.5.0",
    "openai>=1.0.0",
    "core-agents",
]

[project.optional-dependencies]
dev = [
    "pytest>=7.4.0",
    "pytest-asyncio>=0.23.0",
    "ruff>=0.8.0",
]

[build-system]
requires = ["hatchling"]
build-backend = "hatchling.build"

[tool.hatch.build.targets.wheel]
packages = ["src/${AGENT_NAME//-/_}"]

[tool.uv.sources]
core-agents = { path = "../core", editable = true }

3. Create Worker with AgentFactory

Create src/${AGENT_NAME//-/_}/worker.py:

from core_agents import AgentConfig, get_agent_factory
from core_agents.worker import AgentWorker, AgentWorkerConfig

def create_worker() -> AgentWorker:
    config = AgentWorkerConfig(
        task_queue="my-agent",
        name="my-agent",
        description="My agent description",
        workflows=[MyWorkflow],
        activities=[my_activity],
    )
    return AgentWorker(config)

def main() -> None:
    worker = create_worker()
    worker.run()

if __name__ == "__main__":
    main()

4. Create agent_info.py

from core_agents.communication import AgentCapability, AgentInfo

AGENT_INFO = AgentInfo(
    id="my-agent",
    name="My Agent",
    description="What it does",
    endpoint="my-agent.ai-agents.svc.cluster.local",
    capabilities=[
        AgentCapability(
            name="my-capability",
            description="What it does",
            tags=["my", "tags"],
        ),
    ],
)

5. Create GitOps Manifests

mkdir -p gitops/apps/ai-agents/${AGENT_NAME}

Create deployment.yaml, service.yaml, kustomization.yaml.

6. Register with Flux

Add to gitops/apps/ai-agents/kustomization.yaml:

resources:
  - ${AGENT_NAME}

Development Workflow

After creating the agent:

# Run locally with hot-reload
kubani-dev run my-agent --hot-reload

# Run tests
kubani-dev test my-agent

# Run evaluation
kubani-dev eval my-agent

# Build and deploy
kubani-dev build my-agent
kubani-dev deploy my-agent

Reference Agents

  • k8s-monitor: Full federated agent with Sentinel, Healer, Explorer
  • news-monitor: Temporal workflows with personalization
  • core: Shared library with all utilities