Skip to content
Get packs

Guide · Agent skill

How to use the Agent Platform RAG Engine Management agent skill

Updated · written with AI from the project's own documentation and checked against our catalogue data

This guide shows you how to install and use the agent-platform-rag-engine-management skill, a set of instructions that lets an AI agent work with Google Cloud RAG Engine. RAG Engine stores your documents in named collections called corpora so a model can look up relevant passages before answering. It is aimed at developers and data or ML engineers who already have a Google Cloud project. By the end you will know how to install the skill, ask for the tasks it covers, and what it will and will not do without asking you first.

Agent Platform RAG Engine Managementgoogle/skills · Apache-2.0

Quick start

  1. 1Download the ZIP. Get the skill ZIP from the item page on this site.
  2. 2Unzip into a skills folder. Place the folder in ~/.claude/skills/ or in a project's .claude/skills/ folder.
  3. 3Or upload in the Claude apps. Go to Settings → Capabilities → Skills and upload the ZIP.
  4. 4Authenticate with gcloud. Run gcloud auth login and gcloud auth application-default login.
  5. 5Ask for a task in plain language. Name the project, region and corpus ID, and say whether you want to list, inspect, retrieve or answer.
  6. 6Confirm any generation. When asked, check the corpus ID, model and query shown, then reply Yes or No.

What an agent skill is and what this one does

An agent skill is a folder containing a SKILL.md file with instructions that an AI agent reads when a request matches the skill's description. This one is named agent-platform-rag-engine-management and is distributed under the Apache-2.0 licence.

Once installed, the agent knows how to list your corpora and the files inside them, inspect a corpus, retrieve matching text passages, and generate answers grounded in those passages. The skill tells the agent to use the vertexai Python SDK rather than raw REST calls or MCP tools.

  • List corpora and the files in each corpus (read-only).
  • Inspect a single corpus by its resource name (read-only).
  • Retrieve matching context passages for a query (read-only).
  • Generate an answer grounded in a corpus (uses model compute, so the agent asks first).

Install the skill from the ZIP

Download the ZIP for this skill from the item page. Unzip it so the skill folder ends up in ~/.claude/skills/ for your personal use, or in .claude/skills/ inside a project if you want it limited to that project.

Alternatively, you can upload the skill in the Claude apps. Open Settings, go to Capabilities, then Skills, and upload the ZIP there.

  • Personal install: place the unzipped folder in ~/.claude/skills/.
  • Project install: place it in the project's .claude/skills/ folder.
  • Claude apps: Settings → Capabilities → Skills, then upload the ZIP.

Set up Google Cloud authentication

The skill cannot do anything without access to a live RAG Engine corpus. You need a Google Cloud account, a project, and the gcloud CLI. The skill's own setup steps authenticate you with two commands.

You also need Python 3 and access to Vertex AI / Agent Platform RAG Engine in your target project.

  • Requires Python 3.
  • Requires a Google Cloud project with RAG Engine access.
  • The skill needs the google-cloud-aiplatform and google-genai packages; it asks the agent to install them if they are missing.
gcloud auth login gcloud auth application-default login

Trigger the skill with natural requests

You do not need a special command. Describe what you want in plain language and include the project ID, region and corpus ID where you have them. The agent will look for the resource names itself by listing corpora first, and only asks you if that discovery fails.

Good requests name the task: list, inspect, retrieve, or answer. Requests about ordinary databases, Google Workspace content, or other RAG products such as gRAG fall outside the skill.

  • "List the RAG Engine corpora in project my-project in us-central1."
  • "Show me the files in corpus abc in region us."
  • "Inspect corpus abc and tell me its display name."
  • "Search corpus abc for passages about our remote work policy."
  • "Answer my question about remote work using corpus abc."

Confirm before any generation runs

Listing, inspecting and retrieving are read-only and run without confirmation. Generating a grounded answer consumes model compute, so the agent must stop and show you the target corpus ID, target model and query text, then ask for a Yes or No. It will not run the generation in the same turn as that prompt.

Reply Yes only when the parameters shown match what you intended.

Safety notes: scripts and network access

This skill runs scripts and uses the network. The agent executes Python snippets that call Google Cloud APIs under your credentials, and it may install the two Python packages if they are missing. Read the commands before approving a run, and use an account and project you are comfortable querying.

  • Runs scripts: yes. Uses network: yes.
  • Read-only tiers run immediately; generation needs your explicit Yes.
  • The skill tells the agent not to create a virtual environment and to run snippets directly.

Tips for getting useful results

If a retrieval returns nothing, that is a valid result. Report the corpus and query used, and ask whether a broader search is worth trying. The skill notes that a larger similarity_top_k can be tried in the same script. Expect a deprecation warning about vertexai.preview.rag on stderr; the skill says it is expected.

  • Give the exact project, region and corpus ID when you have them to skip discovery.
  • Check file extensions in the file list to see which document types are indexed.
  • Use the grounded answer tier only when you need a generated answer, not just passages.

What people use it for

  • Check which PDF and text files are indexed in a RAG Engine corpus before a release.
  • Confirm a corpus's display name and resource name while debugging an app.
  • Retrieve passages on a policy question to see whether your documents cover it.
  • Generate a grounded answer from an existing corpus after you approve the parameters.

Frequently asked questions

How do I install the agent-platform-rag-engine-management skill?

Download the ZIP from the item page and unzip it into ~/.claude/skills/ for personal use or a project's .claude/skills/ folder. You can also upload the ZIP in the Claude apps under Settings → Capabilities → Skills.

What do I need to use this skill?

You need Python 3, a Google Cloud account with a project, the gcloud CLI for authentication, and access to Vertex AI / Agent Platform RAG Engine in that project. The skill asks the agent to install google-cloud-aiplatform and google-genai if they are missing.

Is this skill free?

The skill is licensed under Apache-2.0. The source does not state pricing for Google Cloud or RAG Engine, so check Google Cloud's own billing terms for your project.

Can this skill query my ordinary database?

No. The skill says standard database queries should go to SQL or Spanner skills instead. It also excludes Google Workspace RAG and other RAG products such as gRAG.

Why does the agent ask me to confirm before generating an answer?

Generating grounded content consumes model compute. The skill requires a Yes or No confirmation showing the corpus ID, model and query text before it runs, and it will not run in the same turn as that prompt.

What does the vertexai preview deprecation warning mean?

The skill states that a warning about vertexai.preview.rag printed on stderr is expected. It says to keep using these snippets rather than switching clients because of it.

New agent skills, repos and AI tools — every week

The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.

Our verdict

Use this skill if you are a developer or data engineer on Google Cloud who wants an agent to explore and query existing RAG Engine corpora. Pick something else if you need ordinary database queries, Google Workspace content, another RAG product, or if you have no Google Cloud project and credentials.

Download the skill