hardness1020/awesome-agent-architecture — reverse-engineered prompt

Build me a repo that teaches how modern AI agents are actually put together around the model. I want it to feel like a clean study guide plus runnable examples, starting with the basic agent loop and then adding tool calling, permissions and sandboxing, hooks, planning and todos, subagents, skills, context handling, memory, prompt assembly, error recovery, tasks, background work, scheduling, worktree isolation, coordination between agents, protocols, autonomy, plugins and channels, observability, and finally loop engineering.

Please organize it as numbered sections with a short writeup for each one, and for the practical sections include small Python demos that build on the previous section so someone can compare versions and learn one mechanism at a time. I also want a top level README that explains the harness idea in plain English, shows how the loop works, compares a couple of real systems like Claude Code and Hermes Agent, and includes simple setup instructions with an API key so I can run the demos. Look up current docs online if you need to.