skyming/awesome-ai-agent

Purpose and Entry Point

This is a Markdown-only curated knowledge repository for AI-agent architecture, frameworks, practices, papers, tools, benchmarks, and open-source projects. The primary navigation and entry point is README.md; it links readers to a small set of locally maintained deep-dive documents and many external resources. There is no executable application, runtime, API, build pipeline, or deployed infrastructure.

Top-Level Information Architecture

README.md organizes the resource catalog into major domains: agent concepts, frameworks, harness engineering, memory, RAG, observability, tools, benchmarks, self-improvement, safety, projects, and resource hubs. It is the canonical index and integration boundary: local documents expand selected topics, while most references resolve to external articles, papers, documentation, and GitHub repositories.

Agent Concepts and Research Foundation

Core agent theory is collected through introductory and research links, with a local translation/analysis document for key reasoning paradigms:

• README.md

• key-concepts/page_tran.md

This section frames agents around LLM reasoning, planning, memory, reflection, action/tool use, and autonomous execution. The repository itself does not implement these stages; it documents their conceptual evolution and source material.

Framework Evaluation and Selection

Framework comparison is centralized in:

• frameworks/frameworks-eval.md

• README.md

The evaluation covers general-purpose frameworks, multi-agent systems, autonomous agents, RAG/data frameworks, platform tools, and domain-specific projects. This is the repository’s main decision-support artifact: readers enter from the index, compare architectural capabilities and use cases, then follow framework-specific external documentation.

Harness and Long-Running Agent Operations

Harness engineering is a major operational theme in:

• README.md

The catalog emphasizes the control plane around an agent loop: context assembly, tool orchestration, permissions, state persistence, subagents, recovery, safeguards, feedback loops, and skill injection. Linked external materials describe concrete implementations from Anthropic, OpenAI, and industry teams; no local runtime or harness code exists.

Memory, Retrieval, and Knowledge Boundaries

Memory and RAG are documented as separate but connected capabilities:

• README.md

Memory resources cover session logs, long-term facts, retrieval, context compression, and persistence. RAG resources cover indexing, keyword/semantic retrieval, graph-based retrieval, and answer synthesis. The boundary is conceptual: this repository points to external systems such as OpenClaw, MemGPT, LightRAG, and GraphRAG rather than hosting indexes, vector stores, embeddings, or document-processing services.

Ecosystem, Evaluation, and Safety

The index curates external infrastructure and standards for production agent systems:

• README.md

Material external services and ecosystems include LLM providers/models, MCP tool servers, OpenTelemetry, LangSmith, Langfuse, Phoenix, browser automation, Docker sandboxes, GitHub projects, and paper archives. Evaluation references cover task benchmarks and safety testing; observability references cover traces, token/cost metrics, latency, and tool failures. These are informational dependencies, not configured repository integrations.

Supplementary Topic Collections

Additional local/top-level documents provide focused reading paths:

• hot-product-tracking/README.md

• the-evolution-of-human-economic-dev.md

hot-product-tracking/README.md appears to be a separately scoped collection, while the-evolution-of-human-economic-dev.md is an independent essay/topic document. Neither is referenced by the supplied root README content, so they should be treated as adjacent content rather than part of the root catalog’s primary navigation unless explicit links are added.