# References and Acknowledgments

> The writing of this book would not have been possible without the wisdom of the community. Below are all the sources we cited, referenced, or drew inspiration from. Our deepest respect to every contributor.

---

## Community Contributors

In the wake of the Claude Code 2.1.88 public source release (March 31, 2026), the Chinese-speaking technical community ignited an unprecedented wave of in-depth analysis. The following community members are listed in order of first citation; each was directly quoted in this book or provided crucial inspiration for our analysis.

### Directly Cited Community Voices

| Contributor | Platform | Key Contribution | Cited In |
|-------------|----------|------------------|----------|
| **@idoubicc** | X/Twitter | open-agent-sdk open-source project (1,658 likes, 321K views); "Claude Code's backyard caught fire, so I asked CC to move out the furniture and build a new house—free for everyone" | Prologue |
| **@IceBearMiner** | X/Twitter | "The more time you spend vibe coding, the more you appreciate software engineering"; treatise on the importance of Harness Engineering | Prologue |
| **@wquguru** | X/Twitter | harness-books (1,662 likes, 238K views); "Sharing the same nouns across tools does not mean sharing the same skeleton underneath" | Prologue, Tool Runtime |
| **@anthropic_security_review** | X/Twitter | "23 checks is not paranoia—it's the minimum surface area coverage for a tool that runs arbitrary shell commands on a real OS." | BashTool Security Defenses |

### Community Members Whose In-Depth Analysis Provided Important References

| Contributor | Platform | Key Contribution | Impact |
|-------------|----------|------------------|--------|
| **@Barret_China** (Li Jing) | X/Twitter | Mapped Claude Code's memory system onto Tulving's 1972 three-memory framework; competitive comparison of CC vs LangMem/Mem0/Zep/OpenClaw | 694 likes, 66K views |
| **@tvytlx** (Xiao Tan) | X/Twitter | One of the earliest systematic deep-dive technical reports, dissecting 4,756 source files | 3,413 likes, 924K views |
| **@AlchainHust** (Huashu) | X/Twitter | *The Claude Code Orange Book*: a 75-page open-source practical handbook | 3,339 likes, 300K views |
| **@0xJoooe** | X/Twitter | Distilled six major prompt-engineering techniques from the Claude Code system prompt | 402 likes |
| **@servasyy_ai** | X/Twitter | Harness Engineering framework (proposed two days before the public source release); 15-chapter, 912-line complete source analysis (cache_edits mechanism, tri-channel telemetry trace chain, four-tier progressive Compact, Flush Gate state machine—the most comprehensive single technical analysis in the Chinese community); CC vs OpenClaw hooks philosophy comparison | 698 + 404 likes, 113K views |
| **@yq_acc** | X/Twitter | Systemic effects theory: "Remove any single layer, and the effectiveness of the remaining layers drops. That is the essence of fine architecture." | Bilingual analysis |
| **@troyhua** (Troy Hua) | X/Twitter | Seven-layer memory + Dreaming system video breakdown; founder of EverMind, CMU PhD in dialogue systems | 383 likes, 132K views |
| **@Pluvio9yte** | X/Twitter | *The End of Vibe Coding Is Actually Engineering* | 344 likes, 145K views |
| **@MaxForAI** | X/Twitter | *First-Tier AI Agent Engineering Architecture Through the Lens of the Community-Released Claude Code Source*, with 7 architectural diagrams | 399 likes, 138K views |
| **@blackanger** (AlexZ) | X/Twitter | *Harness Engineering: From Claude Code Source to Best Practices in AI Coding* online book | 331 likes, 148K views |
| **@jiayuan_jy** | X/Twitter | Multica multi-agent collaboration platform ("manage tasks like Linear, but AI agents are first-class citizens") | 950 likes, 310K views |
| **@GoSailGlobal** | X/Twitter | AgentSkillsHub Top 10 ranking statistics, quantitative evidence of the ecosystem explosion after the public source release | 51 likes |
| **@cryptonerdcn** (NerdC) | X/Twitter | Architecture and agent design analysis + hidden Easter eggs; claude-code-2188 source backup; Buddy Web Demo | 73 likes |
| **@boniusex** | X/Twitter | *Claude Code Source Exposed: The Most Valuable Feature 99% of Users Never Touch* | 62 likes |
| **@jesselaunz** | X/Twitter | PPT-style source analysis (generated by NotebookLM) | 47 likes |
| **@shuang** | X/Twitter | Connected Tulving's 50-year-old paper to AI agent memory design | Academic retrospective |
| **@fried_rice** | X/Twitter | First to discover and publicly report the source code circulating in the community | Event first report |

---

## Open Source Projects

The Claude Code public source release catalyzed a concentrated explosion of open-source activity. Within one week of the release, 6 of the top 10 trending GitHub Agent Skills projects were CC-related. The following projects are mentioned in this book or provided reference material for our analysis.

### Projects Directly Spawned by the Public Source Release

| Project | Stars | Author | Description | Relationship to This Book |
|---------|-------|--------|-------------|---------------------------|
| **claw-code** | 121K+ | Sigrid Jin | Python/Rust rewrite of Claude Code; fastest-growing GitHub project in history (50K in 2 hours) | Prologue event narrative |
| **open-agent-sdk** | — | @idoubicc | MIT-licensed open-source SDK alternative using function calls | Prologue, system evaluation (evidence for process-architecture bottleneck) |
| **open-agent-sdk-typescript** | 1,822 | codeany-ai | TypeScript SDK with no CLI dependency | Prologue project statistics |
| **harness-books** | 1.1K | @wquguru | Books comparing CC vs Codex Harness | Prologue, competitive analysis reference |
| **ccunpacked.dev** | — | Community | English visual guided tour, 11-step Agent Loop interactive animation | Prologue, design reference |
| **byterover-cli** | 3,638 | campfirein | Agent memory layer with cross-session persistence | Prologue project statistics |
| **taches-cc-resources** | 1,731 | glittercowboy | Configuration collection + workflows + best practices | Prologue project statistics |
| **claude-reviews-claude** | 988 | openedclaude | 17-chapter self-review of Claude reading its own source | Prologue project statistics |
| **how-claude-code-works** | 808 | Windy3f3f3f3f | Architecture + Agent loop + tool mechanism analysis | Prologue project statistics |
| **claude-code-from-scratch** | 472 | Windy3f3f3f3f | ~1,300-line TypeScript hands-on tutorial | Prologue project statistics |
| **claude-code-prompts** | 322 | repowise-dev | Ready-to-use prompt templates | Community ecosystem |
| **Multica** | — | @jiayuan_jy | Multi-agent collaboration platform | Community ecosystem impact |
| **Claude Code 橙皮书** (*Orange Book*) | — | @AlchainHust (Huashu) | 75-page practical handbook, Feishu doc | Competitive reference |
| **驾驭工程** (*Harness Engineering*) | — | @blackanger (AlexZ) | Online book at zhanghandong.github.io | Competitive reference |

### Competing Projects Referenced in Analysis

| Project | Nature | Role in This Book |
|---------|--------|-------------------|
| **GitHub Copilot** / **Copilot Workspace** | Microsoft/GitHub AI coding assistant | Multi-chapter competitive comparison (multi-endpoint strategy, plan-driven tool invocation, cloud container model) |
| **Cursor** | AI-first code editor | Multi-chapter competitive comparison (RAG strategy, Background Agents, .cursorrules directive system) |
| **Windsurf** | Codeium AI editor | Competitive comparison (Cascade Engine context tree, VS Code extension ecosystem) |
| **Aider** | Open-source terminal AI coding assistant | Competitive comparison (AST-level Repo Map, ConversationSummaryMemory) |
| **Cline** | Open-source VS Code AI extension | Competitive comparison (extension ecosystem, interaction model) |
| **Continue** | Open-source AI coding assistant | Competitive comparison |
| **CodeX（OpenAI）** | OpenAI coding agent | Multi-chapter competitive comparison (Rust rewrite, parallel agent workflow, AGENTS.md, OS-level network isolation) |
| **OpenCode** | Open-source CLI AI tool (110K+ Stars) | Industry context |
| **Kimi Code** | Moonshot AI coding assistant | Competitive comparison (Agent Swarm architecture, up to 100 concurrent sub-agents) |
| **Devin** | Autonomous AI developer | Competitive comparison (Manage Devins, Human-in-the-loop) |
| **OpenClaw** (formerly Clawdbot) | Open-source fully autonomous agent | Competitive comparison (memory system philosophy: "Memory is not an asset; the ability to use memory correctly is") |
| **GLM / Z.ai / Z Code** | Zhipu AI coding platform | Competitive comparison (GLM-5.1 744B parameters, domestic Ascend 910B chips, private deployment) |
| **Google Antigravity** | Google's AI coding tool | Competitive comparison (Mission Control architecture, Artifacts mechanism, Allow/Deny List) |
| **ForgeCode** | AI coding tool | Terminal Bench benchmark comparison (defeated Claude Code with the same underlying model) |

---

## Academic and Industry Frameworks

This book cites multiple academic theories and industry frameworks, connecting Claude Code's engineering implementation to a broader knowledge system.

### Cognitive Science

| Framework | Originator | Year | Application in This Book |
|-----------|------------|------|--------------------------|
| **Three-memory system** (episodic / semantic / procedural) | Endel Tulving | 1972 | Mapped to Claude Code's memory architecture: jsonl conversation storage (episodic), extractMemories + autoDream (semantic), feedback memory type (procedural) |

### Industry Frameworks

| Framework | Originator | Application in This Book |
|-----------|------------|--------------------------|
| **Model + Runtime + Harness three-layer architecture** | Harrison Chase (LangChain founder) | Foundational framework for prologue and architecture analysis |
| **Three-stage evolution**: Prompt Engineering (2023-24) → Context Engineering (2025) → Harness Engineering (2026) | @servasyy_ai / community consensus | Prologue industry-context positioning |

### Classic Computer Science Concepts (Analogies and Mappings Throughout the Book)

The following classical concepts serve as theoretical anchors for explaining Claude Code's engineering decisions:

- **Operating Systems**: Copy-on-Write (COW), process scheduling, virtual file system, system calls, fork()
- **Distributed Systems**: distributed tracing (OpenTelemetry Span), circuit breaker pattern, eventual consistency, WAL (Write-Ahead Log)
- **Databases**: cache coherence, read-your-writes, append-only writes + offline compaction
- **Computer Networks**: protocol evolution and backward compatibility, ALPN negotiation
- **Security Engineering**: defense in depth, principle of least privilege, supply chain attacks
- **Software Engineering**: deprecation strategy, observability three pillars (Metrics/Tracing/Logging)

---

## Media and Platforms

The following media outlets and platforms provided reports and analytical articles that served as references for the industry background and event narrative in this book.

| Media/Platform | Language | Relevant Content |
|----------------|----------|----------------|
| **CyberNews** | English | First report on the Claude Code public source release |
| **DEV.to** | English | Harness Engineering framework article, source of the three-stage evolution narrative |
| **X / Twitter** | Chinese & English | Primary battlefield for community discussion; 22 directly related tweets with a combined ~3,350,000+ views |
| **GitHub** | English | Main host for community distribution of the source and derivative projects |
| **Zhihu** | Chinese | In-depth technical analysis articles (including four-layer compression + 950-line Python reproduction) |
| **Feishu / Lark** | Chinese | Hosting platform for long-form content such as the *Orange Book* |

---

## Tools and Technical References

The following tools and technologies are analyzed, compared, or mentioned as technical background in this book.

### Claude Code Core Dependency Stack

| Technology | Role | Coverage in This Book |
|------------|------|-----------------------|
| **TypeScript** | Primary language (1,884 files) | Throughout |
| **Node.js / Bun** | Runtime | Startup sequence, pre-import parallel optimization |
| **React / Ink** | Terminal UI framework | Terminal UI chapter (reconciler, virtual DOM) |
| **Yoga Layout Engine** | Layout engine (pure TypeScript port) | Terminal UI chapter (Flexbox character-grid implementation) |
| **Zod v4** | Schema validation | Settings system chapter (engineering decision analysis) |
| **Commander.js** | CLI framework | Startup sequence chapter |
| **chalk** | Terminal colors | Startup sequence |
| **marked** | Markdown parser | CLAUDE.md loading system (`@include` parsing) |

### Anthropic Proprietary Dependencies

| Package | Role | Analysis Status |
|---------|------|-----------------|
| **@anthropic-ai/sdk** | Primary LLM API calls | Analyzed |
| **@anthropic-ai/bedrock-sdk** / **@anthropic-ai/vertex-sdk** | AWS Bedrock / GCP Vertex access | Analyzed |
| **@anthropic-ai/claude-agent-sdk** | Sub-agent capabilities | Analyzed |
| **@anthropic-ai/sandbox-runtime** | Sandbox execution runtime (closed-source) | Analyzed (with explicit analysis-gap notation) |
| **@modelcontextprotocol/sdk** | MCP protocol implementation | Analyzed |

### Observability and Operations

| Technology | Role | Coverage in This Book |
|------------|------|-----------------------|
| **OpenTelemetry** | Distributed tracing, metrics, logs | Telemetry chapter (Span hierarchy, gRPC exporter) |
| **Datadog** | Event analysis backend | Telemetry chapter (client token, event routing) |
| **Honeycomb** | Tracing backend | Telemetry chapter (60KB truncation limit, OTLP protocol) |
| **GrowthBook** | Feature flags / A-B testing | Multiple chapters (`tengu_` prefix gating system) |

### Security-Related

| Technology | Role | Coverage in This Book |
|------------|------|-----------------------|
| **macOS seatbelt** / **sandbox-exec** | Native macOS sandbox | Sandbox system chapter |
| **bubblewrap (bwrap)** | Linux userspace sandbox | Sandbox system chapter |
| **gitleaks** | Secret scanning rules | Cross-cutting concerns (`secretScanner.ts`) |
| **@withfig/autocomplete** | Command completion library (Fig) | BashTool security defenses (command spec registration) |

### Comparative Frameworks and Tools

| Framework/Tool | Role in This Book |
|----------------|-------------------|
| **LangChain** / **LangGraph** | Multi-chapter competitive comparison (AgentExecutor, Callbacks, ConversationSummaryMemory, graph-state-machine orchestration) |
| **LangSmith** | AI observability industry benchmark comparison (Run Tree concept) |
| **LlamaIndex** | Context compression competitive comparison |
| **LangMem** | Memory system competitive comparison (vector database approach) |
| **Mem0** | Memory system competitive comparison (semantic indexing approach) |
| **Zep** | Memory system competitive comparison (knowledge graph + vector hybrid) |
| **LiteLLM** / **Instructor** | Lightweight alternative mentions |
| **Vercel AI SDK** | Hooks system comparison (`onToken` / `onFinish`) |
| **CrewAI** | Multi-agent framework comparison |
| **AutoGen** | Multi-agent framework comparison ("heavy code" orchestration route) |
| **AutoGPT** | Memory strategy comparison ("store memory every step" approach) |
| **Semantic Kernel** | Plugin → Planner evolution comparison |
| **E2B** | Sandbox solution comparison (Code Interpreter SDK) |
| **Docker** / **Kubernetes** | Architecture analogies and security solution comparisons (multiple locations) |
| **MCP（Model Context Protocol）** | Core protocol for the extension ecosystem (77 mentions) |

---

## Acknowledgments

This book was born from an unexpected public source release, but it has come this far thanks to the collective intelligence of the entire community.

**Thank you to the Anthropic engineering team.** We hold your engineering capabilities in genuine respect. Claude Code is a beautifully designed system, and many decisions inspire deep admiration upon close analysis. The critiques in this book are equally sincere—pointing out design trade-offs and potential issues is not disparagement, but the proper attitude of engineering analysis.

**Thank you to the community analysts.** @tvytlx's 924K-view deep report, @AlchainHust's 75-page *Orange Book*, @Barret_China's Tulving framework mapping, @0xJoooe's six principles of prompt engineering, @servasyy_ai's Harness Engineering framework proposed just two days before the public source release, @yq_acc's systemic effects theory, @troyhua's seven-layer memory breakdown—your work has set the benchmark for in-depth analysis in the Chinese technical community. Without these pioneers' explorations, many chapters in this book would be far narrower in analytical scope.

**Thank you to the open-source builders.** @idoubicc turned the public source release into an open-source opportunity, @wquguru established a comparative framework with two books, @jiayuan_jy turned collaboration needs into a product, and ccunpacked.dev lowered the barrier to understanding with interactive animations. You have proven that a community source release can become a learning resource for the entire industry.

**Thank you to the industry thought leaders.** Harrison Chase's three-layer architecture framework provided our analytical coordinate system, and Endel Tulving's 50-year-old memory theory has found new life in the AI agent era. From operating systems and distributed systems to cognitive psychology—the accumulated wisdom of half a century of computer science is validated one by one in this new field.

**Thank you to every reader.** This book has 88 chapters and roughly 390,000 Chinese characters. The very fact that you have read this far is the best possible vote of confidence in deep technical content.

---

> If we have inadvertently omitted any contributor or source that deserves acknowledgment, please let us know and we will add it in a subsequent edition.

---

## Known Source Code Gaps

This book analyzes a source snapshot of Claude Code 2.1.88 recovered from `cli.js.map`. The following modules are referenced in the current snapshot but their host files are missing. This does not affect the analysis of the main behavior thread, but this book will not speculate about their internal implementations.

| Missing Module | Referenced From | Nature |
|----------------|-----------------|--------|
| `SendUserFileTool/` execution host directory | `tools.ts` require references, `ToolSearchTool/prompt.ts`, `Messages.tsx`, `conversationRecovery.ts` | Tool host cluster gap |
| `UserCrossSessionMessage` render component | `UserTextMessage.tsx` require reference | Render host gap |
| `peerSessions.js` full implementation | `SendMessageTool.ts` require reference | Send host gap |
| `src/buddy/observer.ts` (`fireCompanionObserver` definition) | `REPL.tsx` call, `AppStateStore.ts` comment pointer | Observer host gap |
| `setReplBridgeActive()` caller | `bootstrap/state.ts` definition exists, no caller found in full source | Capability gate writer gap |
| `@anthropic-ai/sandbox-runtime` | Multiple references | Closed-source sandbox runtime package |

> The common characteristic of these gaps is that the consumer side (reader/caller/renderer) of the system remains intact; only the executor side (writer/executor/host) is missing from the current snapshot. Behavioral semantics can be inferred from the consumer side, but the concrete implementation awaits official open-sourcing or confirmation in a subsequent version.