Agent Platform RAG Engine Manager
Manage Vertex AI RAG corpora and retrieve grounded context with Python SDKs.
The skill separates read-only operations from compute-consuming generation and requires explicit confirmation with exact parameters before generation. However, ADC login, unpinned dependencies, and limited disclosure of RAG data flow, sensitive-data handling, least privilege, and recovery reduce the score.
The workflow and examples cover the main listing, inspection, retrieval, and generation paths, and placeholder replacement is specified clearly. There is no error handling, abnormal-input feedback, compatibility validation, or reproducible test evidence, so the static score remains conservative.
The intended scenarios, task types, and non-fit products are reasonably clear. Trigger boundaries, input/output contracts, Chinese-language support, and mainland-China reachability are not addressed, while core functionality depends on Google Cloud services.
The documentation is readable and layered, with safety tiers, setup, a decision tree, examples, installation guidance, support information, and repository-level licensing. It lacks skill versioning, a changelog, explicit maintenance ownership, pinned dependency versions, and troubleshooting guidance.
The examples directly cover common RAG Engine management and retrieval tasks and could complete the core workflow. There are no committed execution results, complete output contracts, or comparative-benefit evidence, and SDK/API compatibility is unverified; the score stays within the static-review cap.
Concrete SDK calls, parameters, and resource-name formats provide limited auditability. The supplied files contain no skill-specific test suite, CI coverage, execution records, or independent corroboration, so evidence remains thin.
- Before execution, confirm the project, region, corpus, and query, and verify that ADC permissions are least-privilege.
- Pin and validate google-cloud-aiplatform, google-genai, and vertexai API versions, and add explicit exception handling and failure feedback.
- For sensitive corpora, clarify transmission, storage, and model-processing boundaries; mainland-China users should separately verify network reachability to the required Google Cloud services.
What does this skill do, and when should you use it?
This skill supports corpus management and retrieval for Google Agent Platform RAG Engine. It lists corpora and files, inspects a corpus, retrieves relevant contexts, and generates content grounded in a selected corpus. Operations must use the Vertex AI Python SDK rather than raw REST calls or MCP tools. Content generation consumes compute resources and requires confirmation of the exact corpus, model, and query parameters.
It initializes a Google Cloud project and region with the vertexai Python SDK, lists RAG corpora and their files, retrieves a corpus's resource name and display name, calls rag.retrieval_query to return relevant context and source URIs, and configures a Vertex RAG Store through the google-genai client for Gemini-based grounded generation.
- A Google Cloud user needs to discover RAG corpus and file resource names before running a query.
- A developer needs to inspect the resource name and display name of an Agent Platform RAG Engine corpus.
- An application needs to retrieve relevant context and source URIs for a natural-language query.
- A team needs Gemini to produce an answer grounded in a specified RAG Engine corpus.
- A user wants to inspect file display names and identify PDF or plain-text files by extension.
What are this skill's strengths and limitations?
- Covers corpus and file discovery, corpus inspection, context retrieval, and corpus-grounded generation.
- Provides examples using the Vertex AI Python SDK and Google GenAI SDK.
- Shows both automatic and manual pagination approaches.
- Defines a clear confirmation boundary between read-only retrieval and resource-consuming generation.
- Scoped to Agent Platform RAG Engine; it is not for standard database queries, Google Workspace RAG, or gRAG.
- Requires Google Cloud authentication, ADC, a Python environment, and two Google SDK packages.
- Grounded generation consumes compute resources and must wait for interactive confirmation.
- The source provides no test suite, detailed error-handling guidance, pricing information, or cross-platform validation.
How do you install this skill?
Install the repository's skill collection with npx skills add google/skills. Select Agent Platform RAG Engine Management during installation; its path is skills/cloud/agent-platform-rag-engine-management. The README does not document a separate package installation method for this directory.
How do you use this skill?
Run gcloud auth login and gcloud auth application-default login. Create and activate an environment with python3 -m venv ~/rag_agent_venv and source ~/rag_agent_venv/bin/activate, then install pip install google-cloud-aiplatform google-genai. Use the Python SDKs with a project ID, region, and corpus ID; if resource details are missing, list corpora and files first. Example request: "List the RAG corpora in this project and retrieve context about product policies." Activate the virtual environment before every Python execution. Before grounded generation, confirm the target corpus ID, model, and query text.
How does this skill compare with similar options?
The skill explicitly directs standard database queries to SQL/Spanner skills and excludes Google Workspace RAG and gRAG. It is therefore a fit for Agent Platform RAG Engine users, not a general database or cross-product RAG solution.