Building agent systems

Skill telagod/code-abyss/skills/building-agent-systems

AI agent and LLM system engineering reference covering single-agent dev (ReAct, tool calling, plan-execute), multi-agent coordination (swarm, role decomposition, file locking), LLM security (prompt injection, jailbreak defense, output filtering), RAG architecture (chunking, hybrid retrieval, rerank), and prompt engineering / evaluation (RAGAS, LLM-as-Judge). Use when building AI agents, designing RAG pipelines, orchestrating multi-agent workflows, hardening LLM apps, or writing prompts.From its SKILL.md

Install

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

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

SKILL.md

3.8 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

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。

通用原则

Prompt 即代码须版控 | 输入输出皆验证 | 成本效果平衡 | 持续评估迭代 | 安全边界明确

跨场景铁律

Prompt 版控 — Prompt 是代码，必须 Git；变更要走 review
I/O 验证 — 输入侧防注入，输出侧防 hallucination 落地（结构化 schema、引用追溯）
评估前置 — 上线前必有 eval set；RAGAS / LLM-as-Judge 至少二选一
成本观测 — token / latency / 失败率必埋点；预算阈值自动告警
降级路径 — 多 Agent 失败 → 单 Agent；单 Agent 失败 → 直接回答 + 标记 [unverified]

多 Agent 启用判据

信号	启用 TeamCreate
涉及 ≥3 独立文件	✅
需 ≥2 并行流	✅
总步骤 >10	✅
用户明确要求	✅
单一探索任务	❌（用 explorer 或单 Agent）
单文件改动	❌（用 worker 或直接执行）
单步任务	❌（直接执行）

详细生命周期、文件锁规则、依赖感知、过载保护、降级链：multi-agent-coordination.md

与其他 skill 联动

涉及部署 → provisioning-infrastructure（Vector DB、模型服务）
涉及前端 → applying-ui-design-system（Chat UI / Agent 状态可视化）
涉及安全审计 → securing-systems（LLM AppSec 子域）
涉及评估自动化 → automating-devops（CI 中跑 eval）

What ships with it: 6 files

49.3 KB alongside SKILL.md

references/

Gives 0 of the 12 instructions most agent orchestration skills give in ~1.2k tokens

Counted across 848 of the 1,300 authors here whose files we hold, read 2026-09-06

Dispatch one agent per independent problem domainin 56 of 848, across 42 files
Run full test suite after integrationin 55 of 848, across 42 files
Verify fixes do not conflictin 40 of 848, across 32 files
Review each summary when agents returnin 40 of 848, across 31 files
Write a handoff document summarising the current conversationin 30 of 848, across 25 files
Reference existing artifacts by path or URLin 26 of 848, across 24 files
Give each agent a specific scopein 19 of 848, across 10 files
Give each agent a clear goalin 19 of 848, across 10 files
Include a suggested skills section in the documentin 18 of 848, across 16 files
Tailor the doc to the user argumentsin 18 of 848, across 15 files
Issue all subagent dispatches in the same responsein 17 of 848, across 11 files
Use git worktrees for isolationin 17 of 848, across 8 files

Said here and by no other author read

Read the kernel before making decisions
Version control all prompts
Establish evaluation sets before launch
Monitor costs and latency
Implement degradation paths

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.