How aliyun/ai-agent-handbook works
This repository is a handbook, not an executable Agent platform. Its reader workflow is to learn the architecture, study how to build and run Agents, apply governance and optimization practices, then consult enterprise cases. The graph reflects the documented lifecycle and recommended reading paths; it does not imply runtime calls between chapters.
Main entry points: 01-architecture Agent Architecture · 第 3 章 范式：Harness 的主流构建方式和责任边界.md Harness Design · 第 4 章 任务：编排、长程推进与协作流转.md Task Orchestration · 第 5 章 信息：上下文、状态与可复用能力资产.md Context and State
Components 27
Architecture and Build
- Agent Architecture
01-architecture - Harness Design
02-build/第 3 章 范式：Harness 的主流构建方式和责任边界.md - Task Orchestration
02-build/第 4 章 任务：编排、长程推进与协作流转.md - Context and State
02-build/第 5 章 信息：上下文、状态与可复用能力资产.md - Controlled Actions
02-build/第 6 章 行动：受控执行、验证反馈与交付准备.md
Runtime
- Runtime and Sandbox
03-run/第 7 章 Agent 运行时与沙箱.md - State and Assets
03-run/第 8 章 Agent 状态存储与语义资产.md - AI Gateway
03-run/第 9 章 AI 网关与统一流量治理.md - Async Workflows
03-run/第 10 章 Agent 异步任务与自动化流程.md - Multi-Agent Orchestration
03-run/第 11章 Multi-Agent 协作与编排.md - Distributed Communication
03-run/第 12 章 Agent 分布式通信.md
Governance
- Observability
04-governance/第 13 章 Agent 的可观测性.md - Agent Security
- Asset Management
04-governance/第 15 章 AI 资产的发现与管理.md - Behavior Validation
04-governance/第 16 章 Agent 行为生成与质量验证.md
Optimization
- Model Tuning
05-optimization/第 17 章 模型调优.md - Agent Improvement
05-optimization/第 18 章 Agent 调优总览.md - Trajectory Data
05-optimization/第 19 章 Agent 轨迹数据.md - Runtime Data Processing
05-optimization/第 20 章 Agent 运行时数据处理.md - Golden Datasets
05-optimization/第 21 章 Agent 黄金数据集.md - Badcase Optimization
05-optimization/第 22 章 Agent 优化：Badcase.md - Controlled Evolution
05-optimization/第 23 章 受控自进化.md - Edge Runtime
05-optimization/第 24 章 Agent 边缘运行时与全球优化.md
Cases and Outlook
- Enterprise Cases
06-case-study - Agentic OS Outlook
07-conclusion/第 30 章 从 Agentic Application 到 Agentic OS.md
Other components
- Reader
- Developer Survey
2026-agent-survey-report.md
Connections 26
- Reader → Developer Survey: surveys practice
- Reader → Agent Architecture: learns architecture
- Agent Architecture → Harness Design: frames building
- Harness Design → Task Orchestration: organizes work
- Task Orchestration → Context and State: uses information
- Context and State → Controlled Actions: enables actions
- Reader → Runtime and Sandbox: studies runtime
- Reader → Observability: studies governance
- Reader → Agent Improvement: studies improvement
- Reader → Enterprise Cases: consults cases
- Reader → Agentic OS Outlook: explores outlook
- Runtime and Sandbox → State and Assets: pairs with
- Reader → AI Gateway: studies traffic
- Reader → Async Workflows: studies automation
- Reader → Multi-Agent Orchestration: studies collaboration
- Multi-Agent Orchestration → Distributed Communication: needs communication
- Reader → Agent Security: studies safeguards
- Reader → Asset Management: studies discovery
- Reader → Behavior Validation: studies validation
- Agent Improvement → Trajectory Data: uses evidence
- Trajectory Data → Runtime Data Processing: informs processing
- Runtime Data Processing → Golden Datasets: prepares datasets
- Golden Datasets → Badcase Optimization: supports regression
- Badcase Optimization → Controlled Evolution: informs improvement
- Reader → Model Tuning: studies models
- Reader → Edge Runtime: studies deployment