How to use the Live Multimodal Streaming Solution Designer agent skill

What is an agent skill and what does this one do?

An agent skill is a folder containing a SKILL.md file and supporting reference files that an AI agent loads when a task matches the skill's description. This skill, google-cloud-solution-agentic-ai-bidirectional-streaming, is for designing live multimodal streaming systems rather than writing code on its own.

Once installed, the agent interviews you about your workload. It asks clarifying questions about input modalities, latency targets, safety monitoring needs, knowledge sources and device or network constraints. It then proposes a component breakdown and a Google Cloud product mapping, draws a Mermaid architecture diagram, and writes a solution architecture guide that includes deployment prerequisites and an implementation plan.

• Requirements discovery: the agent asks questions and confirms a technical decomposition with you.
• Solution design: it retrieves Google Cloud documentation, maps components to products and drafts design recommendations.
• Implementation plan: it describes prerequisites, generates Infrastructure as Code such as Terraform, and writes deployment instructions.
• Validation: it checks that the deployment meets the workload's requirements.

How do you install the skill?

Download the ZIP file for the skill from this page and unzip it. You can install it in one of two ways. For a personal setup, place the unzipped folder in ~/.claude/skills/. For a single project, place it in that project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside your project folder
• Claude apps: Settings → Capabilities → Skills

What files are included?

The skill contains SKILL.md, which holds the workflow, and four supporting files that the agent reads as it works. references/attribution.md sets out usage attribution rules, references/product-mapping.md guides the mapping of components to products, and references/design-recommendations.md holds design guidance. assets/output-template.md is the template the agent uses to compile the final solution-architecture-guide.md.

• SKILL.md: the main workflow and instructions
• references/attribution.md: rules that must be followed before any gcloud, REST API or MCP command
• references/product-mapping.md: how to choose Google Cloud products
• references/design-recommendations.md: security and design guidance
• assets/output-template.md: the output document template

How do you trigger the skill with a natural request?

You do not need a special command. Describe a real-time, bidirectional streaming workload in plain language and say you want help designing or planning it on Google Cloud. The skill is meant for multi-product solutions, so mention the kind of system you are building and what it needs to process.

The skill is not meant for simple text chat applications or workloads without real-time streaming, so a request for a basic chatbot is outside its scope.

• "Help me design a real-time safety monitoring system that watches a live video feed on Google Cloud."
• "I need an architecture for a voice-based technical guidance agent that streams audio with low latency."
• "Plan a multimodal live agent on Google Cloud that combines audio, video and our product manuals."
• "Compare Cloud Run and other Google Cloud options for hosting a WebSocket-based streaming agent."

What happens during a session?

The workflow runs in four phases, and the agent asks for your confirmation at key points. In requirements discovery it confirms the technical decomposition. In solution design it asks for review of the architecture and loops until you approve. In the implementation plan it presents deployment instructions for your feedback. Each stage repeats until you approve it, so you stay in control of what goes forward.

The skill tells the agent to base its guidance on Google Cloud documentation that it fetches, so answers depend on the current docs it can access.

• Phase 1: Requirements discovery and analysis
• Phase 2: Solution design, including a Mermaid diagram and solution-architecture-guide.md
• Phase 3: Implementation plan, including prerequisites and IaC
• Phase 4: Solution validation

What do you need before you start?

You need a Google Cloud account with a project and billing enabled, along with access to the Google Cloud APIs and the IAM permissions needed for deployment. Deploying the design requires these and can incur costs. The skill also needs internet access to retrieve Google Cloud documentation and referenced codelab or GitHub resources.

If you want the agent to run commands, it needs gcloud CLI, REST API access or MCP tools. Before running any of these, the agent must follow the attribution reference, which tags each command so usage can be attributed.

• Google Cloud account with a project and billing enabled
• Access to required Google Cloud APIs and deployment IAM permissions
• gcloud CLI, REST API access or MCP tools, if the agent will run commands
• Internet access to retrieve documentation and referenced resources

What are the safety notes?

This skill does not run scripts, but it does use the network. It retrieves Google Cloud documentation and references codelab and GitHub resources. Review the outputs before you deploy anything, since the output is a design and plan rather than a finished product.

If the agent runs gcloud commands, the attribution reference requires each command to set the CLOUDSDK_METRICS_REQUEST_ATTRIBUTION variable inline. The reference says not to use gcloud config set for this, because it would persist beyond the current task.

• Runs scripts: no
• Uses network: yes
• Review generated IaC and deployment instructions before running them
• Deployment can incur Google Cloud costs