adk-infra-expert
关于
This Claude Skill provisions production Vertex AI ADK Agent Engine infrastructure using Terraform. It handles core components including Agent Engine runtime, Code Execution Sandbox, Memory Bank, VPC-SC, and IAM security. Use it when you need to deploy or manage secure multi-agent infrastructure for ADK production environments.
快速安装
Claude Code
推荐/plugin add https://github.com/jeremylongshore/claude-code-plugins-plusgit clone https://github.com/jeremylongshore/claude-code-plugins-plus.git ~/.claude/skills/adk-infra-expert在 Claude Code 中复制并粘贴此命令以安装该技能
技能文档
What This Skill Does
Expert in provisioning production Vertex AI ADK infrastructure with Agent Engine, Code Execution Sandbox (14-day state), Memory Bank, VPC Service Controls, and enterprise security.
When This Skill Activates
Triggers: "adk terraform deployment", "agent engine infrastructure", "provision adk agent", "vertex ai agent terraform", "code execution sandbox terraform"
Core Terraform Modules
Agent Engine Deployment
resource "google_vertex_ai_agent_runtime" "adk_agent" {
project = var.project_id
location = var.region
display_name = "adk-production-agent"
agent_config {
model = "gemini-2.5-flash"
code_execution {
enabled = true
state_ttl_days = 14
sandbox_type = "SECURE_ISOLATED"
}
memory_bank {
enabled = true
}
tools = [
{
code_execution = {}
},
{
memory_bank = {}
}
]
}
vpc_config {
vpc_network = google_compute_network.agent_vpc.id
private_service_connect {
enabled = true
}
}
}
VPC Service Controls
resource "google_access_context_manager_service_perimeter" "adk_perimeter" {
parent = "accessPolicies/${var.access_policy_id}"
name = "accessPolicies/${var.access_policy_id}/servicePerimeters/adk_perimeter"
title = "ADK Agent Engine Perimeter"
status {
restricted_services = [
"aiplatform.googleapis.com",
"run.googleapis.com"
]
vpc_accessible_services {
enable_restriction = true
allowed_services = [
"aiplatform.googleapis.com"
]
}
}
}
IAM for Native Agent Identity
resource "google_project_iam_member" "agent_identity" {
project = var.project_id
role = "roles/aiplatform.agentUser"
member = "serviceAccount:${google_service_account.adk_agent.email}"
}
resource "google_service_account" "adk_agent" {
account_id = "adk-agent-sa"
display_name = "ADK Agent Service Account"
}
# Least privilege for Code Execution
resource "google_project_iam_member" "code_exec_permissions" {
for_each = toset([
"roles/compute.viewer",
"roles/container.viewer",
"roles/run.viewer"
])
project = var.project_id
role = each.key
member = "serviceAccount:${google_service_account.adk_agent.email}"
}
Tool Permissions
Read, Write, Edit, Grep, Glob, Bash - Enterprise infrastructure provisioning
References
GitHub 仓库
相关推荐技能
sglang
元SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。
evaluating-llms-harness
测试该Skill通过60+个学术基准测试(如MMLU、GSM8K等)评估大语言模型质量,适用于模型对比、学术研究及训练进度追踪。它支持HuggingFace、vLLM和API接口,被EleutherAI等行业领先机构广泛采用。开发者可通过简单命令行快速对模型进行多任务批量评估。
langchain
元LangChain是一个用于构建LLM应用程序的框架,支持智能体、链和RAG应用开发。它提供多模型提供商支持、500+工具集成、记忆管理和向量检索等核心功能。开发者可用它快速构建聊天机器人、问答系统和自主代理,适用于从原型验证到生产部署的全流程。
llamaguard
其他LlamaGuard是Meta推出的7-8B参数内容审核模型,专门用于过滤LLM的输入和输出内容。它能检测六大安全风险类别(暴力/仇恨、性内容、武器、违禁品、自残、犯罪计划),准确率达94-95%。开发者可通过HuggingFace、vLLM或Sagemaker快速部署,并能与NeMo Guardrails集成实现自动化安全防护。
