Stats
Actions
Tags
Help us improve
Share bugs, ideas, or general feedback.
From code-abyss
AI/LLM knowledge reference covering agent development, RAG systems, prompt engineering, LLM security, and evaluation. Use when building AI agents or working with LLMs.
npx claudepluginhub telagod/code-abyss --plugin code-abyssHow this skill is triggered — by the user, by Claude, or both
Slash command
/code-abyss:building-ai-systemsThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
| 意图 | 秘典 | 核心 |
Builds production-ready LLM applications, advanced RAG systems, and AI agents with vector search, multimodal AI, agent orchestration, and enterprise integrations. Use for LLM features, chatbots, or AI apps.
Builds production-ready LLM apps, advanced RAG systems, and AI agents with vector search, multimodal AI, LangChain orchestration, and enterprise integrations.
Assesses AI system type (RAG, agents, prompts, LLM evaluation) and applies patterns for hallucination guards, context budgets, injection defenses, and cost tracking.
Share bugs, ideas, or general feedback.
| 意图 | 秘典 | 核心 |
|---|---|---|
| Agent 开发 | agent-dev | 多 Agent 编排、工具调用、ReAct 循环 |
| LLM 安全 | llm-security | Prompt 注入、越狱防护、输出过滤 |
| RAG 系统 | rag-system | 向量库、检索策略、重排序、混合检索 |
| Prompt/评估 | prompt-and-eval | Few-shot、CoT、RAGAS、LLM-as-Judge |
文档 → Chunking(递归/语义) → Embedding → 向量库(Pinecone/Qdrant/pgvector)
查询 → Query 改写(HyDE/多查询) → 混合检索(向量+BM25) → Rerank(Cohere/cross-encoder) → LLM 生成
| 决策点 | 选项 | 判据 |
|---|---|---|
| Chunk 策略 | 固定/递归/语义 | 结构化文档→递归；长文→语义 |
| 检索方式 | 纯向量/混合/知识图谱 | 通用→混合；关系密集→图谱 |
| 向量库 | pgvector/Qdrant/Pinecone | 已有 PG→pgvector；大规模→托管 |
| 模式 | 结构 | 适用 |
|---|---|---|
| ReAct | Think→Act→Observe 循环 | 通用工具调用 |
| Plan-Execute | 先规划再逐步执行 | 复杂多步任务 |
| Multi-Agent | 角色分工+消息传递 | 大型协作任务 |
| Reflection | 生成→自评→修正 | 代码/文本质量提升 |
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