Agent framework lists need comparable receipts

Martim Santos' ai-agents-frameworks repo collects runnable Python examples across AG2, Agno, AutoGen, Claude Agent SDK, CrewAI, Google ADK, LangChain, LangGraph, LlamaIndex, Microsoft Agent Framework, OpenAI Agents SDK, Pydantic AI, smolagents, and Strands Agents SDK. Each framework folder has its own README, uv setup, and examples for tools, structured output, memory, streaming, observability, RAG, MCP/A2A, or multi-agent flows. Why it matters: Agent-framework comparisons are usually too abstract. This repo is useful because it exposes defaults through code: how each stack handles tool calls, memory, RAG, observability, handoffs, workflows, and failure-prone setup. The comparison lab also records response time, token usage, RAG/API misses, and prompt-dependence caveats.

Try this: Do not ask which framework is best. Pick one real task, run it in two or three folders, and compare receipts: trace, token usage, memory behavior, retry/failure mode, observability hooks, and whether a verifier can sit outside the agent.