Skip to content

Learning Map

This tutorial follows the order of real agent engineering work: build the smallest useful capability, then add requirements, context, tools, evaluation, and release gates to the same quality loop.

Beginner path

  1. Quick Start - Run a minimal example first.
  2. What is Skill - Understand reusable capability boundaries.
  3. MCP Introduction - Learn how agents connect to external context.
  4. OpenSpec Core Concepts - Define intent before implementation.
  5. Harness Testing Infrastructure - Protect AI output quality with tests.

Engineering path

  1. Skill Practice - Package a task into a reusable capability.
  2. Build an MCP Server - Expose tools, resources, and safety boundaries.
  3. Complete Workflow - Move from proposal to implementation to archive.
  4. Evaluation and Quality - Turn acceptance criteria into regression checks.
  5. Deployment and Safety - Handle permissions, secrets, sandboxing, and rollback.

Architecture path

  1. Context Engineering - Design context windows, memory, and information priority.
  2. Workflow Orchestration - Connect Skills, MCP, OpenSpec, and Harnesses.
  3. Retrieval and Knowledge - Bring docs and codebase knowledge into the agent.
  4. Best Practices - Avoid unmaintainable prompt and tool design.
  5. Contributing - Extend the tutorial with the same quality bar.

Module standard

StandardMeaning
ConceptExplain the problem and vocabulary
OperationProvide executable steps or minimal configuration
PracticeInclude a task learners can self-check
TroubleshootingName common failure modes and diagnosis steps
Next stepLink to the next chapter or example

Start now

Begin with Quick Start. If you already know the basics, jump to MCP Introduction or Evaluation and Quality.

基于 MIT 许可发布