Agent Engineering Fundamentals · Course Guide
This course consists of 32 chapters covering six pillars for building production-grade AI applications. Every section is rigorously aligned with authoritative source material:
| Module | Chapters | Authoritative Source |
|---|---|---|
| Harness Engineering Fundamentals | 01–06 | walkinglabs/learn-harness-engineering、OpenAI、Anthropic official docs |
| Prompt Engineering | 07–15 | Prompt Engineering Guide |
| MCP Protocol | 16–21 | Anthropic official modelcontextprotocol |
| Agent Skills | 22–25 | Anthropic official anthropics/skills |
| Loop Engineering | 26–29 | cobusgreyling/loop-engineering |
| AGENTS.md & Capstone | 30–32 | agentsmd/agents.md |
What You Will Learn
- Harness Engineering Fundamentals (01–06): Why strong models still fail, the five-subsystem model, the repo-as-spec principle, long-task continuity, control mechanisms, and observability;
- Prompt Engineering (07–15): From basic prompt elements to Few-shot, Chain-of-Thought, ReAct, ToT, RAG augmentation, and adversarial attack defense;
- MCP Protocol (16–21): Model Context Protocol architecture, three primitives — Tools/Resources/Prompts, transport layer and lifecycle; build a Python SDK Server and connect to a client;
- Agent Skills (22–25): SKILL.md standard structure, progressive disclosure mechanism, writing high-quality skills, team governance, and a three-repo comparison of mattpocock/skills · obra/superpowers · OpenSpec;
- Loop Engineering (26–29): "Stop prompting, design loops" — five building blocks + Memory, seven production loop patterns, L1–L3 autonomy levels, Loop Ready scoring and failure modes;
- AGENTS.md & Capstone (30–32): The open standard "README for agents", nested priority rules, wiring Prompt/Harness/MCP/Skills/Loop into a complete workflow.
Official Reference Sources
This course is built strictly against the following authoritative references:
| Source | Link |
|---|---|
| www.promptingguide.ai | https://www.promptingguide.ai/ |
| walkinglabs.github.io | https://walkinglabs.github.io/learn-harness-engineering/en/ |
| github.com | https://github.com/modelcontextprotocol |
| github.com | https://github.com/anthropics/skills |
| github.com | https://github.com/cobusgreyling/loop-engineering |
| github.com | https://github.com/agentsmd/agents.md |
Content version baseline: Multi-source integrated
Prerequisites
- Basic Python proficiency; MCP hands-on chapters require Python 3.10+;
- An LLM API Key (pairs well with this site's FastAPI / Agno courses).
Relationship to Other Courses
This course covers the "principles and protocol layer." It complements the framework courses on this site: Agno and CrewAI teach you to build agents with frameworks; this course teaches the underlying prompt design, tool protocols, capability packaging, and runtime loops.
👉 Start with Chapter 1.