1. Requirements discovery. Gather detailed requirements related to the cloud workload or use case that the user needs assistance for.
2. Solution architecture. Use the requirements that were gathered in Phase 1 to generate a detailed solution architecture for the cloud workload or use case.
3. Solution validation. Create a plan to validate the generated solution, generate validation instructions and scripts, and run the validation.
4. Solution packaging and presentation. Consolidate the generated content and present the solution.

• Strict phase separation: During Phase 1 (Requirements discovery), when you ask the user clarifying questions, DON'T recommend, propose, or outline any architectural designs, cloud services, or component mappings. This prevents premature architecture commitments or hallucinations before the full scope is understood.
• Halting for approval: For any step where you are instructed to "obtain approval before proceeding", you MUST stop executing, present the completed tasks to the user, and wait for their explicit approval. You MUST NOT proceed to execute any subsequent tasks or generate any further guidance in that response.
• Ground all generated content: For all tasks across all phases, you MUST first look in the following resources:
  • Product Mapping,
  • Design Recommendations for the required guidance. If the guidance does not provide the required information, you MUST ground the generated content by using the following resources:
  • Google Developer Knowledge MCP server: https://developers.google.com/knowledge/mcp.md.txt
    • Server: https://developerknowledge.googleapis.com/mcp
      • Tools:

        • developerknowledge:search_documents

        • developerknowledge:get_documents

        • developerknowledge:answer_query

  • Relevant skills from https://github.com/google/skills
  • Official Google Cloud documentation in Related Guidance

<table> <thead> <tr> <th>Legacy Name</th> <th>Updated Name</th> <th>Notes</th> </tr> </thead> <tbody> <tr> <td>Vertex AI</td> <td>Gemini Enterprise Agent Platform</td> <td>Gemini Enterprise Agent Platform can be shortened to Agent Platform after first instance</td> </tr> <tr> <td>Vertex AI Embedding</td> <td>Text embedding on Gemini Enterprise Agent Platform</td> <td>This refers to the text embedding models available on Gemini Enterprise Agent Platform</td> </tr> <tr> <td>Vertex AI Matching Engine</td> <td>Vector Search</td> <td></td> </tr> </tbody> </table>

• Step 1: Ask the user to describe the functional requirements of the workload, including catalog dataset details (e.g., e-commerce apparel, retail products, patent database), search modalities (natural language text, visual search, attribute filters), metadata attributes for faceted filtering (e.g.,

  category

  ,

  sub_category

  ,

  color

  ,

  gender

  ,

  price

  ), and quality checks (reranking, LLM validation).
• Step 2: You MUST explicitly ask the user to describe ALL of the following six categories of non-functional requirements. You need this information because each category represents a critical architectural pillar, and neglecting any of them can result in a solution that is insecure, unreliable, or inefficient (do NOT omit any of them):
  • Security, privacy, and compliance: E.g., private VPC endpoints, Private Service Connect, Direct VPC Egress, and access control.
  • Reliability: E.g., high availability, failover, disaster recovery goals (RTO/RPO), regional vs multi-region AlloyDB topology.
  • Cost: E.g., budget constraints for compute, database instances, and Gemini Enterprise Agent Platform API calls.
  • Operational excellence: E.g., monitoring, logging, dashboards, and automated deployment.
  • Performance: E.g., target P95 query latency (e.g., < 100ms), vector search recall target (e.g., > 95%), catalog item scale, and QPS expectations.
  • Sustainability: E.g., carbon footprint, low-carbon region selection.
• Step 3: Ask the user whether the workload currently runs on other cloud providers or on-premises.
  • If the user's answer is "yes", then ask the user to describe the architecture of the current deployment.
  • If the user's answer is "no", then proceed to the next step.
• Step 4: Ask the user to describe dependencies, if any, on other workloads, products, or tools (e.g., existing inventory databases, ERP systems, application runtime languages like Java or Python).
• Step 5: Review the input that the user has provided so far, and check whether there are any ambiguities, conflicts, or contradictions in the functional requirements, non-functional requirements, and dependencies. You MUST compare all requirements against each other to identify any conflicts.
  If you identify any ambiguities, conflicts, or contradictions in the requirements that the user has provided, you MUST do the following for each ambiguity, conflict, or contradiction:
  • Identify exactly where each contradiction lies and explain to the user why the requirements are incompatible and cannot be simultaneously satisfied. Do NOT treat fundamental contradictions as design choice questions (e.g., asking how to implement or configure a conflicting requirement).
  • Ask the user to clarify their trade-off preferences to resolve the contradiction.
  • If the user delegates the choice to you (e.g., the user replies with "do what you think is best" or "you decide"), then provide a clear suggestion to resolve the ambiguity or contradiction, explain your reasoning, and ask the user to approve your suggestion.
  Critical: Until all the ambiguities and contradictions that you identify are resolved according to the preceding guidance, you must NOT recommend or generate any architecture design or Google Cloud product recommendations.
• Step 6: Summarize the functional and non-functional requirements provided by the user into a consolidated requirements summary.
• Step 7: Present the generated requirements summary to the user and obtain approval (the user MUST explicitly say "yes" or "I approve") before proceeding to Phase 2.