Agent coursepublicCodex
Working With Coding Agents
A practical course for builders who want coding agents to become reliable collaborators, not unpredictable code generators. Learn how to brief agents, provide context, choose local or background workflows, split work, review diffs, use instructions and automation safely, and avoid common failure modes.
This course was created with an agent. Human Handbook has checked its structure, but has not independently reviewed its content or facts.
Covers
- commissioning coding agents with clear goals and constraints
- context files such as AGENTS.md and CLAUDE.md
- local, IDE, web, cloud, and background task choices
- task splitting, worktrees, subagents, and parallel review
- diff review, tests, hooks, routines, and safety gates
Doesn’t cover
- model benchmarking
- prompt jailbreaks or unsafe automation
- vendor pricing optimization
- full CI/CD platform setup
- secrets management implementation
Lesson 3
CodexContext That Scales
Use project instructions and memory without turning them into clutter.
Put durable rules where the harness reads them
Different products use different instruction systems. Codex reads AGENTS.md files before work. Claude Code uses CLAUDE.md, rules, and memory mechanisms. The shared practice is simple: keep durable project norms in files, and keep task-specific instructions in the prompt.