How to use the Hybrid Search on AlloyDB agent skill

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

An agent skill is a folder of instructions and reference files that an AI agent loads when a task matches the skill's description. This skill, named google-cloud-solution-hybrid-search-alloydb, is published in the google/skills repository under Apache-2.0.

When installed, the agent follows a structured workflow for building a search system that combines vector search with keyword and attribute filtering, using AlloyDB and related Google Cloud services. It first asks detailed questions about your data, performance, security, cost and reliability needs, then produces an architecture, a validation plan and deployment guidance. Its advice is grounded in the bundled Google product mapping and design references and in Google's documentation.

• Phase 1: requirements discovery, including questions on functional needs and six non-functional categories (security, reliability, cost, operations, performance, sustainability)
• Phase 2: solution architecture based on the gathered requirements
• Phase 3: solution validation, with a validation plan and scripts
• Phase 4: packaging the output using the bundled template

How do you install the Hybrid Search on AlloyDB skill?

Download the ZIP file for the skill from the page you are reading. Then choose one of two routes.

For Claude Code, unzip it into your personal skills folder at ~/.claude/skills/, or into a project-specific .claude/skills/ folder inside your project. For the Claude apps, upload the ZIP under Settings → Capabilities → Skills.

• Personal install: unzip into ~/.claude/skills/
• Project install: unzip into the .claude/skills/ folder of your project
• Claude apps: upload the ZIP under Settings → Capabilities → Skills

How do you trigger the skill with a natural request?

You do not need a special command. Describe a hybrid search workload in plain language that mentions the kind of search you need. The skill's description lists the triggers it responds to: vector search combined with structured SQL filtering, faceted attributes, semantic reranking, in-database AI validation, or serverless hosting on transactional relational databases, analytical warehouses, or managed database engines.

Because the skill is interview-driven, expect the agent to ask questions before it recommends anything. Phase 1 is designed so that the agent does not propose architecture until it understands your requirements.

• "Help me design product search on Google Cloud that combines vector search with SQL filters on category, color and price."
• "I need an architecture for catalog search with semantic reranking on AlloyDB."
• "Plan a hybrid search system for our patent database, with reranking and recall testing."

What do you need before you start?

The skill assumes you have a working Google Cloud setup and the tools the agent will use to run commands. The source lists these requirements.

The workflow also refers to the Google Developer Knowledge MCP server for grounding. Plan for the agent to need network access to Google documentation and to run scripts during validation.

• A Google Cloud account and project
• The gcloud CLI
• REST API or MCP tooling for executing Google Cloud commands
• Access to the Google Developer Knowledge MCP server, referenced in the workflow for grounding
• Runs scripts: yes
• Uses network: yes

What should you check for safety?

Because this skill runs scripts and uses the network, review what the agent proposes before it executes commands against your Google Cloud project. The skill's attribution file requires the agent to tag every gcloud command with a metrics environment variable that identifies only the skill and its version. That variable is set inline on each command, and the file says not to use gcloud config set for it.

The skill also requires the agent to pause at points marked for approval. It must present completed tasks and wait for your explicit approval before continuing.

• Read the commands the agent plans to run, especially those that change cloud resources
• Expect the workflow to halt for your approval at several points
• The skill's design recommendations reference security controls such as CMEK and VPC Service Controls; confirm these match your own policies

What tips help you get useful output?

Give the agent as much requirement detail as you can in your first message. The skill is instructed to acknowledge requirements you already provided and not ask you to repeat them. Also, if you already know your constraints, such as latency targets or catalog size, include them up front.

The skill's output template covers sections such as an executive summary, functional and non-functional requirements, and current state. Ask for the output to follow that structure if you want a consistent document.

• Include catalog size, search modalities, facet attributes and latency targets in your opening message
• State whether the workload runs elsewhere today
• Ask for the final document to follow the bundled output template

When should you not use this skill?

The skill is not meant for simple keyword-only search, and its description says so directly. It is also not the right fit when a standalone non-relational vector database is required. If you want a quick answer without a multi-step interview, this workflow will feel slow because it halts for input at several points.

• Simple keyword-only search
• Projects that require a standalone non-relational vector database
• Quick answers without a requirements interview