AgentSkillsCN

retail-agent-learner

学习ADK核心概念，深入了解零售代理的运作流程。当您想探究代理的工作原理、ADK究竟是什么，或是希望理解其架构设计、状态流转，以及各代理的具体职能时，不妨从这里入手。

SKILL.md
--- frontmatter
name: retail-agent-learner
description: Learn ADK concepts and understand the retail agent pipeline. Use when asking how the agent works, what ADK is, wanting to understand the architecture, state flow, or individual agent purposes.

Retail Agent Learner

Pipeline Overview

This is an 8-agent pipeline for retail site selection analysis:

code
User Query
    ↓
IntakeAgent ────────────→ Extract location + business type
    ↓
MarketResearchAgent ────→ Google Search for demographics
    ↓
CompetitorMappingAgent ─→ Google Maps Places API
    ↓
GapAnalysisAgent ───────→ Python code execution (pandas)
    ↓
StrategyAdvisorAgent ───→ Extended thinking synthesis
    ↓
ParallelAgent ──────────→ Concurrent artifact generation
    ├── ReportGenerator ─→ HTML report
    ├── InfographicAgent → Image generation
    └── AudioOverview ───→ TTS podcast audio

What Each Agent Does

AgentPurposeKey Tool/Feature
IntakeAgentParse user requestAgentTool pattern
MarketResearchAgentFind market datagoogle_search built-in
CompetitorMappingAgentMap competitorsGoogle Maps API
GapAnalysisAgentCalculate viabilityCode execution (pandas)
StrategyAdvisorAgentSynthesize recommendationsExtended thinking
ReportGeneratorCreate HTML reportArtifact generation
InfographicAgentGenerate visualImage generation
AudioOverviewCreate podcast audioTTS multi-speaker

Key ADK Concepts

Agents

  • •LlmAgent: Single LLM call with tools and instructions
  • •SequentialAgent: Runs sub-agents one after another
  • •ParallelAgent: Runs sub-agents concurrently

State Flow

Data passes between agents via session state:

code
IntakeAgent → state["target_location"], state["business_type"]
MarketResearchAgent → state["market_research_findings"]
CompetitorMappingAgent → state["competitor_analysis"]
GapAnalysisAgent → state["gap_analysis"]
StrategyAdvisorAgent → state["strategic_report"]

Tools

Functions that agents can call to perform actions. Access state via ToolContext.

Callbacks

Lifecycle hooks: before_agent_callback and after_agent_callback.

Learning Path

Start with the 9-part tutorial series:

  1. •Part 1: Setup + First Agent
  2. •Part 2: IntakeAgent
  3. •Part 3: MarketResearchAgent
  4. •Part 4: CompetitorMappingAgent
  5. •Part 5: GapAnalysisAgent (code execution)
  6. •Part 6: StrategyAdvisorAgent (extended thinking)
  7. •Part 7: ArtifactGeneration (parallel outputs)
  8. •Part 8: Testing
  9. •Part 9: Production Deployment

[See references/architecture.md for detailed data flow] [See references/adk-concepts.md for ADK deep dive]