Wiki Index

A knowledge base built and maintained by LLM, following the LLM Wiki pattern by Andrej Karpathy.

Topics

AI Agent Architecture

The full landscape of AI agent design — from foundational patterns and scaling laws to enterprise deployment and security. 24 sources, 16 concepts, 5 entities, 4 synthesis pages.

Enterprise Agent China

Private deployment and domestic ecosystem for AI agents in Chinese enterprises. Covers platform integration (WeChat, DingTalk), domestic technology stack (Qianwen, ERNIE, Ascend NPU), skill lifecycle management, and compliance requirements (MLPS 2.0, PIPL). 14 sources, 8 concepts, 5 entities, 3 synthesis pages.

Concepts:

• Harness - Controlled boundary between reasoning and execution
• Multi-Agent Architectures - Five topology patterns
• Agent Scaling Laws - Empirical performance laws
• Cognitive Depth Adaptation - Dynamic reasoning allocation
• Planning - Task decomposition and dynamic replanning
• Agent Memory - Memory as harness function
• RAG - Retrieval-augmented generation for static knowledge
• Skills - Fundamental capability units
• Skill Lifecycle - Create, evaluate, connect, evolve
• Self-Evolving Agents - Learning from deployment
• Reflection - Metacognitive self-critique and improvement
• Sandbox Architectures - Isolation patterns
• Agent Security - Attack vectors and defenses
• Observability - Production monitoring and tracing
• Agentic Problem Frames - Engineering framework
• AI Infrastructure Stack - Layered architecture

Entities:

• Anthropic - Claude, agentskills.io
• LangChain - LangGraph, Deep Agents
• NVIDIA - OpenShell, GPU ecosystem
• SkillNet - Skill ontology platform
• Memento-Skills - Self-evolving framework

Synthesis:

• Capability vs Reliability: The 2026 Inflection
• Skills Landscape Comparison
• Enterprise Agent Adoption: Reality vs Hype
• RAG vs Agent Memory: Architectural Boundaries

How It Works

1. Ingest - Feed sources, LLM builds structured wiki pages
2. Query - Ask questions, answers get filed back into the wiki
3. Lint - Health-check for contradictions, gaps, and stale content

Recent Activity

See log for full history.