skyming/awesome-ai-agent

What it solves

This project is a curated directory of AI agent resources. It solves the problem of scattered information about AI agents by gathering links to articles, papers, frameworks, engineering guides, evaluation benchmarks, and open-source projects into one organized list. It helps developers and researchers quickly find what they need to understand, build, or improve AI agents.

How it works

The repository is a single README file that groups resources into themed sections. Each section focuses on a different aspect of AI agents: concepts and background reading, framework comparisons and engineering deep dives, harness design (the “scaffolding” around models), perception and memory systems, RAG (retrieval-augmented generation) practice, evaluation and observability, tooling and MCP (Model Context Protocol), benchmark suites, self-evolution techniques, safety practices, and curated open-source projects. Every entry includes a link and a short description in Chinese, so you can scan the list and jump to the resources that interest you. There is no code or installation; the value is in the curation and the accompanying summaries.

Who it’s for

This is for developers, engineers, researchers, and technical managers who are learning about or building AI agents. It is also useful for people comparing agent frameworks, looking for evaluation tools, or exploring real-world implementation patterns from companies like Uber, Anthropic, and Alibaba. Since the descriptions are in Chinese, it is especially helpful for Chinese-speaking readers, but the linked resources are mostly in English.

Highlights

• Organized into 13+ thematic sections, covering everything from core agent concepts to production-grade infrastructure.
• Includes a deep framework comparison across six categories (LangChain, AutoGen, CrewAI, etc.) with a selection decision matrix.
• Features recent practical articles from major companies (Uber, Anthropic, Alibaba) on harness engineering and long-running agents.
• Covers memory systems, RAG upgrades, and context-engineering techniques with hands-on examples.
• Provides evaluation benchmark overviews (GAIA, AgentBench, VBench, ClawBench) and observability tool comparisons (LangSmith vs Langfuse vs Phoenix).
• Lists open-source agent harnesses like Claude Code, OpenAI Codex, DeepSeek-Harness, and many more.