Further Reading Map
This map turns a curated resource list into a reading path for this tutorial. These links are optional; the code in this repo remains self-contained.
Core Harness Design
| Read This | Why It Helps | Related Chapters |
|---|---|---|
| Anthropic: Building Effective Agents | Agent loop, tool use, workflows vs agents. Read this first. | s01, s02, s24 |
| Anthropic: Effective harnesses for long-running agents | Long-running sessions, persistence, recovery. | s06, s07, s09, s14 |
| Anthropic: Harness design for long-running application development | Coding-agent harness concerns. | s04, s07, s13, s23 |
| OpenAI: A practical guide to building agents | Agent components and product framing. | s01, s02, s24 |
Context Engineering
| Read This | Why It Helps | Related Chapters |
|---|---|---|
| Anthropic: Effective Context Engineering for AI Agents | Context selection, isolation, compaction. | s10-s15 |
| Claude Code best practices | Practical coding-agent workflows. | s04, s07, s16 |
| Lilian Weng: LLM Powered Autonomous Agents | Planning, memory, reflection overview. | s07, s10-s14 |
Tools, Skills, and Protocols
| Read This | Why It Helps | Related Chapters |
|---|---|---|
| Anthropic: Writing effective tools for agents | Tool schema design and affordances. | s02, s03, s17 |
| DeepSeek API docs | DeepSeek key setup and compatible API entrypoints used by the learning path. | README, provider appendix |
| DeepSeek Anthropic-compatible guide | Lets the early Anthropic-style lesson scripts run with a DeepSeek key. | s01-s24 |
| OpenAI: Responses API overview | Direct model interface for stateful interactions, tools, function calling, and built-in capabilities. | s01, s02, s24 |
| OpenAI: Function calling | Tool-call flow and function_call_output shape used by the OpenAI provider adapter. | s02, provider appendix |
| OpenAI Agents SDK | Higher-level runtime for managed turns, tools, guardrails, handoffs, sessions, and tracing. Useful as a contrast to owning the loop. | provider appendix, future |
| Model Context Protocol | External tool connector protocol. | s17 |
| Agent Client Protocol | UI/client to agent communication model. | s07, s24 |
| Claude Code Skills / Agent Skills | Skill packaging patterns. | s16 |
| datawhale: 如何写出好的 Skill | Practical Chinese guide to writing skills. | s16 |
| Skill survey: Evaluation and Evolution | 4 evolution paradigms, 6 benchmark families, SKILL-INJECT attack taxonomy. Local digest: skill-evolution-and-evaluation.md | s16, s17, s23 |
Memory Systems
| Read This | Why It Helps | Related Chapters |
|---|---|---|
| mem0 | Memory extraction and retrieval patterns. | s10-s12 |
| Letta (MemGPT) | Stateful agents and memory management. | s10-s14 |
| Generative Agents paper | Memory, reflection, planning loop. | s10-s14 |
Multi-Agent Systems
| Read This | Why It Helps | Related Chapters |
|---|---|---|
| Anthropic: How we built our multi-agent research system | Context isolation and subagent coordination. | s07, s08, s14 |
| Cognition: Don't Build Multi-Agents | A useful skeptical counterweight. | s07, s18 |
| AutoGen / CrewAI | Alternative multi-agent abstractions. | s07, s18 |
Classic Papers (chronological)
| Paper | Why It Matters Here | Related Chapters |
|---|---|---|
| ReAct (2022) | The reasoning + acting loop s01 implements. | s01 |
| Toolformer (2023) | Learning when to call tools. | s02, s03 |
| Reflexion (2023) | Self-reflection from execution feedback. | s09, s14 |
| Generative Agents (2023) | Memory / reflection / planning. | s10-s12 |
| Voyager (2023) | Skill library + lifelong learning — the ancestor of skill self-evolution. | s16 |
| AgentBench (2023) / WebArena (2023) / SWE-bench (2023) | How agent capability is measured. | s20, s24 |
Evals and Observability
| Read This | Why It Helps | Related Chapters |
|---|---|---|
| Anthropic: Demystifying Evals for AI Agents | How to test agent behavior. | s20, s23, s24 |
| Langfuse / Phoenix / Helicone | Tracing and observability. | s09, s21, s23 |
| SWE-bench / τ-bench / WebArena / AgentBench / GAIA | Must-know agent benchmarks. | s20, s24 |
Similar Build-From-Scratch Tutorials
| Project | Why It Helps | Related Chapters |
|---|---|---|
| shareAI-lab/learn-claude-code | Gradual Claude Code-style rebuild. | Whole repo |
| shareAI-lab/claw0 | Agent loop to always-on gateway, ~10 sections. | s01-s24 |
| datawhalechina/hello-agents | Chinese agent fundamentals. | s01-s04 |
| walkinglabs/learn-harness-engineering | Harness-oriented learning path. | s05-s24 |
| HKUDS/nanobot | Ultra-light personal agent baseline (Python). | s01-s04, s10 |
| HKUDS/OpenHarness | Open harness reference direction. | s06-s24 |
| lukilabs/craft-agents-oss | Full-stack agent app architecture sample. | s05-s07, s24 |
How To Use This Map
Do not read everything first. The best order is:
- Build chapters s01-s04.
- Read the core harness design pieces.
- Build s05-s09.
- Read context engineering and memory resources.
- Build s10-s18.
- Read tools/protocols and multi-agent pieces.
- Build s19-s24.
- Read evals and observability resources.