/build-full-stack

Arguments: $ARGUMENTS

Goal: Complete end-to-end Google ADK project setup from initialization through deployment

Core Principles:

• Analyze project requirements FIRST before building
• Execute sequentially - wait for each command to complete before proceeding
• Track progress with TodoWrite
• Validate at checkpoints
• Each phase builds on previous output

Phase 0: Project Analysis and Requirements Gathering Goal: Analyze existing project files to understand what needs to be built

Actions:

Read Project Configuration Files:

1. Read .claude/project.json: !{Read .claude/project.json} Extract:

   • Tech stack (languages, frameworks)
   • Infrastructure components
   • Deployment targets
   • Project name and description

2. Read features.json: !{Read features.json} Extract:

   • Feature list with IDs (F001, F002, etc.)
   • Feature descriptions and requirements
   • Dependencies between features
   • Implementation status

3. Read execution.json (if exists): !{bash test -f execution.json && cat execution.json} Extract:

   • What's already been executed
   • What's pending
   • What failed and needs retry

4. Scan specs/ directory: !{bash find specs/ -name "*.md" -type f 2>/dev/null | head -20} Extract:

   • Feature specifications
   • Technical requirements
   • API contracts
   • Architecture decisions

Analyze Requirements:

Parse project.json to determine:

• Language: Python/TypeScript/Go/Java
• Agent types needed: LLM/Custom/Workflow
• Tools required: Gemini API, Google Cloud, Custom
• Streaming needed: Yes/No (check for voice/video features)
• A2A needed: Yes/No (check for multi-agent features)
• Observability: Development/Production
• Deployment target: Vertex AI/Cloud Run/GKE/Local

Parse features.json to determine:

• Which features require agents
• Which features need streaming
• Which features need A2A protocol
• Which features are already implemented

Create Customized Build Plan:

Based on analysis, determine execution order:

• Skip init if SDK already configured
• Only add components required by specs
• Prioritize features with dependencies first
• Use language-specific configurations
• Target deployment platform from project.json

Store analysis results:

• $LANGUAGE (from project.json)
• $AGENT_TYPES (from features.json)
• $TOOLS_NEEDED (from specs)
• $FEATURES_TO_BUILD (from features.json + execution.json)
• $DEPLOYMENT_TARGET (from project.json)

Phase 1: Initialize Progress Tracking Goal: Create todo list for all workflow phases

Actions:

TodoWrite with todos:

1. "Initialize Google ADK project" (pending)
2. "Add agents (LLM/custom/workflow)" (pending)
3. "Integrate Gemini API and Google Cloud tools" (pending)
4. "Configure bidirectional streaming" (pending)
5. "Set up A2A protocol" (pending)
6. "Add observability (logging/tracing/analytics)" (pending)
7. "Configure evaluation and testing" (pending)
8. "Deploy to cloud (Vertex AI/Cloud Run/GKE)" (pending)

Phase 2: Sequential Command Execution Goal: Execute all setup commands in order

Actions:

Mark todo #1 as in_progress. Run /google-adk:init $ARGUMENTS WAIT for completion. !{bash test -f pyproject.toml || test -f package.json} Mark todo #1 as completed.

Mark todo #2 as in_progress. Run /google-adk:add-agent WAIT for completion. Mark todo #2 as completed.

Mark todo #3 as in_progress. Run /google-adk:add-tools WAIT for completion. Mark todo #3 as completed.

Mark todo #4 as in_progress. Run /google-adk:add-streaming WAIT for completion. Mark todo #4 as completed.

Mark todo #5 as in_progress. Run /google-adk:add-a2a WAIT for completion. Mark todo #5 as completed.

Mark todo #6 as in_progress. Run /google-adk:add-observability WAIT for completion. Mark todo #6 as completed.

Mark todo #7 as in_progress. Run /google-adk:add-evaluation WAIT for completion. Mark todo #7 as completed.

Mark todo #8 as in_progress. Run /google-adk:deploy WAIT for completion. Mark todo #8 as completed.

Phase 3: Summary Goal: Display comprehensive setup summary

Actions:

Display summary:

Google ADK Full Stack Setup Complete

Components Configured:

• Project initialized with Google ADK SDK
• Agents: LLM agents, custom agents, workflow agents
• Tools: Gemini API, Google Cloud tools integrated
• Streaming: Bidirectional streaming configured
• A2A Protocol: Agent-to-Agent communication enabled
• Observability: Logging, tracing, analytics active
• Evaluation: Testing and metrics configured
• Deployment: Live on Vertex AI/Cloud Run/GKE

Next Steps:

1. Test agent interactions
2. Monitor observability dashboards
3. Run evaluation test suite
4. Access deployment endpoints

Mark all todos as completed.