Building Agent Systems

A comprehensive reference for building production AI agent systems, covering the full stack from single-agent patterns like ReAct and plan-execute to multi-agent coordination with swarm intelligence and file locking. The routing table is particularly useful: it gives you clear decision rules for when to use single vs. multi-agent architectures, with concrete thresholds like "3+ independent files" or "2+ parallel flows" to trigger TeamCreate mode. Also includes RAG pipeline design (chunking strategies, hybrid retrieval, reranking), LLM security hardening against prompt injection and jailbreaks, and evaluation frameworks using RAGAS and LLM-as-Judge. The "scale up is hard, scale down is easy" principle for choosing multi-agent upfront is solid practical advice. Treats prompts as version-controlled code and mandates eval sets before production, which is the right forcing function.

GitHub stars: 224
last updated: Jun 10, 2026

View on GitHub

Install

$ npx -y skills add telagod/code-abyss --skill building-agent-systems --agent claude-code

Installs into .claude/skills of the current project.

Files

丹鼎秘典 · Agent / LLM 工程

判断先于执行：决定「是否做 / 选什么 / 如何取舍」（栈、方案、架构、权衡）前，先读领域判断内核 skills/_kernel/ml/SKILL.md——它管 judgment，本秘典管 execution；冲突时以内核判断为准。

单 Agent 是器，多 Agent 是阵。先选规模，再选模式。

路由

意图	加载	核心
单 Agent 开发（工具调用、ReAct）	agent-dev	ReAct / Plan-Execute / Reflection
多 Agent 协同（>=3 文件 or >=2 并行）	multi-agent-coordination	蚁群仿生、文件锁、依赖图
多 Agent 协议细节（消息素、收阵报告）	multi-agent-protocol	Codex 原生协议、角色定义
LLM 安全（注入、越狱、输出过滤）	llm-security	OWASP LLM Top 10 视角
RAG 系统（向量、检索、重排）	rag-system	Chunking / 混合检索 / Cohere rerank
Prompt + 评估	prompt-and-eval	Few-shot / CoT / RAGAS / LLM-as-Judge

规模决策

单步任务（一文件、一查询）         → 直接执行（不需要 Agent 框架）
多步任务（计划 + 工具）             → 单 Agent (ReAct)
复杂任务（>5 步、需反思）           → 单 Agent (Plan-Execute / Reflection)
独立并行任务（>=3 文件、>=2 流）    → 多 Agent (TeamCreate)
跨域协作（角色明确）                → 多 Agent (角色分工)

犹豫时优先 TeamCreate — 串行降级容易，并行升级难。