Appearance
Sources Registry
Last reviewed: 2026-08-28
版本敏感事实优先使用协议规范、官方 SDK 文档和厂商 Engineering 文档。社区资料只作为补充。模块编号按 canonical knowledge ownership 分组；不表达 13 周课程顺序。
First-party Engineering Case Study — OpenAI Codex
OpenAI 官方 coding-agent 实现，适合作为 Agent loop、Tool Runtime、Session/Context、Compaction、Approval/Sandbox 与 public SDK boundary 的一手工程案例。实现细节是 version-sensitive engineering evidence，不等于协议规范；固定阅读版本
rust-v0.150.1（稳定版 0.150.1，2026-08-27）。所有结论必须区分 Source Fact / Design Inference / Product-specific Choice，且固定 tag 时不得引用仅main存在的路径。仓库 Apache-2.0；模型权重与服务端模型实现不在该开源范围内。Reviewed: 2026-08-28.
Engineering Case Study — Grok Bot 0.18 Reconstruction
仅作为工程案例层（Case Study），优先级低于 Specification / RFC / First-party Engineering。该仓库明确声明自己是对公开 Grok Bot 0.18.0 的非官方 source-oriented reconstruction，并包含 inference router、MCP routing、local Docker sandbox 等扩展；不得把其中的命名、协议版本或模块边界当成规范事实。Reviewed: 2026-08-26.
- pandafeeeder/grok-bot-0.18-reconstructed
- Upstream README
- Provenance / evidence rule
- Security notes
- Source map
01. Agent / Harness Foundations
- Anthropic — Building effective agents
- Anthropic — Scaling Managed Agents: Decoupling the brain from the hands
- OpenAI Agents SDK — Agents
- OpenAI Agents SDK — Agent orchestration
02. Model Layer & Structured Interaction
- OpenAI Agents SDK — Models
- OpenAI Agents SDK — Running agents
- OpenAI Agents SDK — Testing
- OpenAI Agents SDK — Results
03. Python Engineering for Harness
04. Prompt & Context Engineering
- Anthropic — Effective context engineering for AI agents
- Anthropic — Advanced tool use
- Anthropic — Effective harnesses for long-running agents
05. Retrieval & RAG
- OpenAI API — Vector stores
- LangChain — Retrieval
- LangGraph — Agentic RAG
- LangChain — Cross encoder reranker
06. Tool Engineering
- Anthropic — Writing effective tools for AI agents
- Anthropic — Advanced tool use
- OpenAI Agents SDK — Tools
07. Planning & Orchestration
- Anthropic — Building effective agents
- OpenAI Agents SDK — Agent orchestration
- OpenAI Agents SDK — Handoffs
08. Async & Concurrency
- Python 3.14 — Coroutines and Tasks
- Python — asyncio queues
- Python — asyncio synchronization primitives
09. Execution Runtime
10. State / Session / Memory / Durable State
11. Reliability & Human Control
12. HTTP & Streaming for Harness Services
13. Database & Durable Storage
14. Redis / Cache / Queue / Worker
15. MCP
2026-08-24 核验：
2026-07-28Core 已正式发布；Roots / Sampling / Logging 和 legacy HTTP+SSE 进入 deprecation window。Tasks 属于独立 Extension，当前 specification 页面仍为 Draft。
- MCP — 2026-07-28 Specification
- MCP — 2026-07-28 release notes
- MCP TypeScript SDK — Supporting protocol revision 2026-07-28
- MCP TypeScript SDK — Protocol versions
- Model Context Protocol
16. Workspace & Coding-Agent Engineering
- OpenAI Agents SDK — Sandbox agents concepts
- OpenAI Agents SDK — Sandbox agents quickstart
- Anthropic — Effective harnesses for long-running agents
17. Security & Sandbox
- Anthropic — How we contain Claude across products
- OWASP — Top 10 for Large Language Model Applications
- OpenAI Agents SDK — Sandbox concepts
18. Observability, Testing & Evaluation
2026-08-24 核验：GenAI semantic conventions 已迁移到独立
open-telemetry/semantic-conventions-genai仓库；agent/framework spans、plan 与 tool execution 目前仍是 Development，必须经版本化 mapper 接入。
- OpenAI Agents SDK — Tracing
- OpenAI Agents SDK — Testing
- OpenTelemetry GenAI — GenAI spans
- OpenTelemetry GenAI — Agent / framework spans
- LangGraph — Persistence
19. Performance, Cost & Capacity
- OpenAI Agents SDK — Running agents
- OpenAI Agents SDK — Tracing
- OpenTelemetry GenAI — GenAI spans
- OpenTelemetry GenAI — Agent / framework spans