adk-production
This skill should be used when the user asks about "deploying", "production", "Agent Engine", "Vertex AI", "Cloud Run", "GKE", "Kubernetes", "hosting", "scaling", "guardrails", "safety", "content filtering", "input validation", "output validation", "authentication", "OAuth", "API keys", "credentials", "security plugins", "testing agents", "evaluation", "evals", "benchmarks", "tracing", "Cloud Trace", "logging", "observability", "AgentOps", "LangSmith", "user simulation", or needs guidance on deploying ADK agents to production environments, implementing safety measures, access control, secure authentication, testing, debugging, monitoring, or evaluating ADK agent quality.
When & Why to Use This Skill
The ADK Production skill is a comprehensive guide designed to help developers transition AI agents from local development to enterprise-grade production environments. It focuses on deploying agents using Google Cloud services like Vertex AI Agent Engine, Cloud Run, and GKE, while emphasizing robust security through input/output guardrails and OAuth authentication. By integrating advanced observability tools and evaluation frameworks, it ensures that ADK agents are scalable, secure, and maintain high-quality performance in real-world applications.
Use Cases
- Deploying and scaling ADK agents on Google Cloud Platform (GCP) using managed services like Agent Engine or containerized environments like Cloud Run and GKE.
- Implementing multi-layered security measures, including content filtering guardrails and secure credential management, to protect agent interactions and sensitive data.
- Establishing production-grade observability and quality assurance through structured logging, Cloud Trace integration, and automated evaluation (evals) frameworks.
| name | ADK Production |
|---|---|
| description | This skill should be used when the user asks about "deploying", "production", "Agent Engine", "Vertex AI", "Cloud Run", "GKE", "Kubernetes", "hosting", "scaling", "guardrails", "safety", "content filtering", "input validation", "output validation", "authentication", "OAuth", "API keys", "credentials", "security plugins", "testing agents", "evaluation", "evals", "benchmarks", "tracing", "Cloud Trace", "logging", "observability", "AgentOps", "LangSmith", "user simulation", or needs guidance on deploying ADK agents to production environments, implementing safety measures, access control, secure authentication, testing, debugging, monitoring, or evaluating ADK agent quality. |
| version | 2.0.0 |
ADK Production
Complete guide for deploying, securing, testing, and monitoring ADK agents in production. Covers deployment platforms, security guardrails, authentication, testing frameworks, tracing, and observability.
When to Use
Deployment:
- Deploying agents to production
- Choosing between hosting options (Agent Engine, Cloud Run, GKE)
- Configuring auto-scaling
- Setting up CI/CD for agents
- Integrating with Vertex AI services
Security:
- Validating or filtering user input
- Filtering agent responses before delivery
- Implementing OAuth or API key authentication
- Creating reusable security plugins
- Blocking unsafe topics or content
Quality & Testing:
- Creating test suites for agents
- Evaluating agent responses against criteria
- Debugging execution with tracing
- Setting up production monitoring
- Automated testing with synthetic users
When NOT to Use
- Local development → Use
@adk-getting-startedinstead - Agent creation → Use
@adk-agentsinstead - Tool integration → Use
@adk-toolsinstead - General callbacks → Use
@adk-behaviorinstead - Multi-agent systems → Use
@adk-multi-agentinstead
Key Concepts
Deployment
Agent Engine is the recommended managed deployment. Auto-scales, integrates with Vertex AI services, no infrastructure management.
Cloud Run offers container control with serverless scaling. Build custom Docker images for more control over the runtime.
GKE (Kubernetes) provides enterprise-scale deployment. Full control over infrastructure, networking, and scaling policies.
Deployment CLI: adk deploy handles Agent Engine deployment. For Cloud Run/GKE, containerize with adk api_server.
Environment Configuration: Use environment variables for credentials. Never commit secrets to source control.
Security
Input Guardrails validate user input before processing. Use before_model_callback to block or modify unsafe requests.
Output Guardrails filter agent responses before returning to users. Use after_model_callback to redact PII, profanity, or sensitive data.
Authentication secures tool access. Configure OAuth credentials for Google APIs or custom authentication for third-party services.
Security Plugins bundle reusable security callbacks. Create plugins for logging, rate limiting, or content moderation.
Credential Management uses environment variables and secure storage. Never hardcode secrets in agent code.
Quality & Testing
Evaluations (Evals) test agent behavior against expected outputs. Define test cases with inputs and expected results, measure pass rates.
Tracing captures execution flow for debugging. Cloud Trace integration shows LLM calls, tool executions, and timing.
Logging provides structured event capture. Use LoggingPlugin for consistent log formatting and levels.
Observability integrates with third-party platforms (AgentOps, LangSmith) for production monitoring and analytics.
User Simulation automates testing with synthetic conversations. Generate diverse test scenarios without manual testing.
References
Detailed guides with code examples:
Deployment:
references/agent-engine.md- Vertex AI Agent Enginereferences/cloudrun.md- Cloud Run deploymentreferences/gke.md- Kubernetes deployment
Security:
references/guardrails.md- Input/output validationreferences/auth.md- Authentication patternsreferences/security-plugins.md- Reusable security bundles
Quality & Testing:
references/evals.md- Evaluation frameworkreferences/tracing.md- Cloud Trace integrationreferences/logging.md- Structured loggingreferences/observability.md- Third-party integrationsreferences/user-sim.md- Synthetic user testing