agentic-infra

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Agentic Infra Skill

Agentic 基础设施技能集

Quick Start

快速开始

Design use case: "Help me design a customer support agent" → Consult

taxonomy.yaml

to identify relevant concerns → load the relevant

patterns/*.md

cards → use

templates/adr.md

Review use case: "Review this agentic architecture ADR" → Run

scripts/lint_agentic_arch.py <adr.md>

→ cross-reference findings against

checklists/review.md

→ produce

templates/review-report.md

设计用例："帮我设计一个客户支持Agent" → 查阅

taxonomy.yaml

确定相关关注点 → 加载对应的

patterns/*.md

模式卡片 → 使用

templates/adr.md

模板

审查用例："审查这份Agentic架构的ADR文档" → 运行

scripts/lint_agentic_arch.py <adr.md>

→ 对照

checklists/review.md

交叉验证检查结果 → 生成

templates/review-report.md

报告

Taxonomy

分类体系

See

taxonomy.yaml

for the full concern → patterns mapping across 8 concern areas:

reliability

safety

memory

tool-use

orchestration

observability

eval

cost

查看

taxonomy.yaml

获取涵盖8个关注点领域的完整「关注点→模式」映射：

reliability

（可靠性）、

safety

（安全性）、

memory

（内存管理）、

tool-use

（工具调用）、

orchestration

（编排调度）、

observability

（可观测性）、

eval

（性能评估）、

cost

（成本控制）

Pattern Library

模式库

All 21 patterns have full pattern cards in

patterns/

. Each card specifies:

Intent, context, forces/tradeoffs
Solution with code evidence
Failure modes and instrumentation
Related patterns and PDF page citations

#	Pattern	File
1	Prompt Chaining	patterns/01-prompt-chaining.md
2	Routing	patterns/02-routing.md
3	Parallelization	patterns/03-parallelization.md
4	Reflection	patterns/04-reflection.md
5	Tool Use / Function Calling	patterns/05-tool-use.md
6	Planning	patterns/06-planning.md
7	Multi-Agent Collaboration	patterns/07-multi-agent-collaboration.md
8	Memory Management	patterns/08-memory-management.md
9	Learning & Adaptation	patterns/09-learning-adaptation.md
10	Model Context Protocol (MCP)	patterns/10-mcp.md
11	Goal Setting & Monitoring	patterns/11-goal-setting-monitoring.md
12	Exception Handling & Recovery	patterns/12-exception-handling-recovery.md
13	Human-in-the-Loop (HITL)	patterns/13-hitl.md
14	RAG	patterns/14-rag.md
15	A2A Inter-Agent Communication	patterns/15-a2a-communication.md
16	Resource-Aware Optimization	patterns/16-resource-aware-optimization.md
17	Reasoning Techniques	patterns/17-reasoning-techniques.md
18	Guardrails / Safety	patterns/18-guardrails-safety.md
19	Evaluation & Monitoring	patterns/19-evaluation-monitoring.md
20	Prioritization	patterns/20-prioritization.md
21	Exploration & Discovery	patterns/21-exploration-discovery.md

所有21种模式在

patterns/

目录下都有完整的模式卡片。每张卡片包含：

意图、应用场景、约束与权衡
附带代码示例的解决方案
故障模式与监控手段
相关模式及PDF文档页码引用

序号	模式名称	文件路径
1	Prompt Chaining	patterns/01-prompt-chaining.md
2	Routing	patterns/02-routing.md
3	Parallelization	patterns/03-parallelization.md
4	Reflection	patterns/04-reflection.md
5	Tool Use / Function Calling	patterns/05-tool-use.md
6	Planning	patterns/06-planning.md
7	Multi-Agent Collaboration	patterns/07-multi-agent-collaboration.md
8	Memory Management	patterns/08-memory-management.md
9	Learning & Adaptation	patterns/09-learning-adaptation.md
10	Model Context Protocol (MCP)	patterns/10-mcp.md
11	Goal Setting & Monitoring	patterns/11-goal-setting-monitoring.md
12	Exception Handling & Recovery	patterns/12-exception-handling-recovery.md
13	Human-in-the-Loop (HITL)	patterns/13-hitl.md
14	RAG	patterns/14-rag.md
15	A2A Inter-Agent Communication	patterns/15-a2a-communication.md
16	Resource-Aware Optimization	patterns/16-resource-aware-optimization.md
17	Reasoning Techniques	patterns/17-reasoning-techniques.md
18	Guardrails / Safety	patterns/18-guardrails-safety.md
19	Evaluation & Monitoring	patterns/19-evaluation-monitoring.md
20	Prioritization	patterns/20-prioritization.md
21	Exploration & Discovery	patterns/21-exploration-discovery.md

Review Workflow

审查工作流

Run lint script:

python3 scripts/lint_agentic_arch.py <path/to/adr.md>

Inspect JSON findings (each finding includes:
```
severity
```
,
```
rule
```
,
```
pattern_ref
```
,
```
evidence_page
```
)
Walk
```
checklists/review.md
```
for any gaps not caught by lint
Fill in
```
templates/review-report.md
```
with findings

运行lint脚本：

python3 scripts/lint_agentic_arch.py <path/to/adr.md>

检查JSON格式的检查结果（每个结果包含：
```
severity
```
（严重程度）、
```
rule
```
（规则）、
```
pattern_ref
```
（模式引用）、
```
evidence_page
```
（证据页码））
对照
```
checklists/review.md
```
检查是否有lint未检测到的漏洞
将检查结果填入
```
templates/review-report.md
```
模板

Design Workflow

设计工作流

Identify system concerns from
```
taxonomy.yaml
```
Load the pattern cards for those concerns
Choose patterns considering forces/tradeoffs in each card
Record decisions in
```
templates/adr.md
```
Self-check against
```
checklists/review.md
```
before finalizing

从
```
taxonomy.yaml
```
中确定系统关注点
加载对应关注点的模式卡片
结合每张卡片中的约束与权衡选择合适的模式
将决策记录到
```
templates/adr.md
```
模板中
最终定稿前对照
```
checklists/review.md
```
进行自我检查

Examples

示例

Design scenario:
```
examples/design-example.md
```
Review scenario:
```
examples/review-example.md
```

设计场景示例：
```
examples/design-example.md
```
审查场景示例：
```
examples/review-example.md
```

Evidence Policy

证据规范

Every rule in this skill cites a PDF page. Format:

(p. N)

. Never state a rule without citation.

本技能集中的每条规则都标注了PDF文档的页码，格式为

(p. N)

。所有规则必须附带引用，不得无依据陈述。