本课程面向开发者与自动化构建者，系统教授基于Codex Desktop的AI代理循环工程，涵盖验证优先设计、多智能体协作、工作树及安全配置，打造可审计的AI生产力系统。
原始标题：Loop Engineering with Codex Desktop: AI Agent Workflows
本课程是一门面向开发人员、技术运营及自动化构建者的 AI 代理（Agentic AI）循环工程进阶实战课，核心教导如何基于 Codex Desktop 平台构建高可靠性、非单发式（Non-one-shot）的自主 AI 智能体工作流。课程彻底摆脱了“拼运气”的盲目提示词调优，专注于将模糊的需求转化为具备明确目标契约、上下文工程、约束边界和可衡量完成标准的闭环重复循环。学员将系统掌握“验证优先”的设计思维，学习如何让代理在执行任务时自动检测回归、收集证据并触发自我修复循环，同时深入攻克多智能体协作的高阶架构——包括利用工作树（Worktrees）、状态隔离、子代理（Subagents）分工与多剂监督，并安全地配置工具权限、提示词注入防护及 MCP 外部上下文边界，旨在为团队打造一套稳定、可审计且可扩展的 AI 生产力操作系统。
Published 7/2026
Created by The AI Orchestrator
MP4 | Video: h264, 1920×1080 | Audio: AAC, 44.1 KHz, 2 Ch
Level: Intermediate | Genre: eLearning | Language: English | Duration: 18 Lectures ( 2h 7m ) | Size: 999.3 MB
B
Build reliable AI agent workflows with planning, context, verification, tools, automations, worktrees, and subagents
What you’ll learn
⚡ Design reliable AI-agent loops with clear goals, context, constraints, and done-when criteria.
⚡ Use Codex Desktop planning, agent instructions, and reusable skills to keep work consistent and auditable.
⚡ Build verification-first workflows that test outputs, catch regressions, and repair failures with evidence.
⚡ Apply safe tool, permission, MCP, and prompt-injection boundaries when agents use external context.
Requirements
❗ Codex Desktop experience is required
❗ Basic comfort using a computer, browser, terminal, or code editor is helpful but not mandatory.
❗ Bring a real workflow, repository, or automation task you want to improve with AI-agent loops.
Description
This course contains the use of artificial intelligence.
Loop Engineering with Codex Desktop is a practical course for building reliable AI agent workflows instead of relying on one shot prompts. You will learn how to turn a vague request into a repeatable loop with a clear goal, useful context, visible constraints, verification steps, and a measurable done when condition.
The course starts with small loops and then builds toward professional agent workflows. You will practice planning before execution, giving Codex Desktop the right project context, using agent instruction files, setting boundaries for tools and permissions, and checking results with evidence instead of assumptions. Each lesson connects a concrete problem to a loop pattern you can reuse on real work.
We cover goal contracts, context engineering, verification first thinking, repair loops, review and evaluator loops, regression checks, prompt injection awareness, stop conditions, MCP and external context boundaries, reusable skills, native automations, scheduled loops, worktrees, handoffs, state isolation, subagents, and multi agent supervision. The capstone ties these ideas together by showing how to build, fail, repair, and prove a workflow instead of hoping the first attempt is correct.
This course is designed for developers, technical founders, operators, automation builders, product builders, and power users who want dependable AI assisted execution on real projects. You do not need an advanced machine learning background. You need basic computer literacy, curiosity, and a willingness to work step by step.
By the end, you will know how to run Codex Desktop in a more professional way: safer permissions, clearer prompts, stronger evidence, cleaner handoffs, better review habits, and workflows that are easier to repeat, explain, repair, and scale. The goal is not only to get one impressive output. The goal is to build a dependable operating system for AI agent work.
Who this course is for
⭐ Developers, technical operators, and automation builders who want reliable AI-agent workflows.
⭐ Product, ops, and engineering leads evaluating Codex Desktop for repeatable team processes.
⭐ AI power users who want structured planning, verification, repair, and safety practices instead of one-shot prompts.
此处内容需要权限查看
会员免费查看
