Put Codex to work for you
Projects, rules, the agent loop, permissions, skills, subagents, and scheduled tasks.
The problem
The model that wrote the plan tends to approve its own plan.
Using a single AI leaves gaps that it cannot see itself. The problems below show up for people who already work with agents, and the training addresses each one.
- The plan looks good because the same session that wrote it also reviewed it.
- Switching assistants forces you to explain the whole project all over again.
- The project rules sit in a CLAUDE.md that Codex does not read.
- /goal pursues the target with no limit and spends time and tokens beyond what was planned.
- Every task goes to the same model, even when another one would do it better or more cheaply.
After the training
By the end of the 6 courses, you will be able to:
Separate the brain from the model
CLAUDE.md becoming AGENTS.md, a portable core, skills with polyskill, and readback proven in a fresh session.
Switch assistants without losing the thread
Handoff at the end of a session and prime at the start of the next: the next agent reads and checks the state before continuing.
Have Codex critique Claude's plan
The claudex loop: stop hook, state in YAML, three personas, and calibrated rounds until LGTM.
Build plugins using this method
SPEC first, incremental prompts, state machine, stop hook with a loop, and a security pass.
Split the work across models
The 70/20/10 rule, task routing, cross-review, and metrics to tell whether the split pays off.
6 courses in order · 126 lessons and modules
Six courses, from the first command in Codex to automatic debate.
The order follows the path of the area: understand Codex, separate the project from the model, build with agents, and automate the debate between the two. If you already use Codex, you can start at course 2.
- 1
Codexv6
Codex: the 18 essential concepts
Projects, rules, the agent loop, permissions, skills, subagents, and scheduled tasks, in 15-minute lessons for beginners.
18 lessonsOpen course 1 - 2
AgnosticPT · EN · ES
Claude → Codex
Why separate the brain from the model, CLAUDE.md → AGENTS.md, a portable core, skills with polyskill, readback, and the daily handoff and prime cycle.
18 modulesOpen course 2 - 3
Building
Master Codex
The 4 faces of Codex, AGENTS.md, plan mode, worktrees, and a complete product built with agents, through to Codex Cloud, @codex review, and automations.
24 modulesOpen course 3 - 4
Debate
iClaudeX
claudex under the hood: Claude writes the plan, Codex critiques it with three personas, calibrated rounds, /claudex review, and return metrics.
24 modulesOpen course 4 - 5
Plugin
MakeClaudeX
From prompt to production plugin: hooks, state, SPEC first, fail-open patterns, and a complete plugin assembled in five prompts.
24 modulesOpen course 5 - 6
Multimodel
DeepClaudeX
The 70/20/10 rule across GPT, Claude, and DeepSeek, task routing, cross-review, and hands-on projects, from a landing page to a microservice, with a comparative study.
18 modulesOpen course 6
Who it's for
Made for you if
- You have Claude and Codex and use each one in isolation, with neither checking the other.
- You want to switch assistants in the middle of a project without explaining everything again.
- You want a plan reviewed by another model before writing the first line.
- You want to know which model does which part of the work, with testing instead of guesswork.
Not for you yet if
- You want to migrate from one to the other, or stay agnostic for good: the path is the Claude → Codex area.
- You have never opened a terminal: start with the From Chat to Agent area.
Included in the training
The kits that connect the two come included.
Use Both, agente-claude-codex, claudex, and the Codex Cheat Sheet are open on GitHub, along with the area's videos. You apply them to your own project while you study.
Projects and kits
Model: one context, several platformsagente-claude-codex kitagente-claude-codex · GitHubUse Both Kitclaudex · GitHubCodex Cheat SheetTopic mapClaude → Codex areaLong Execution Area
Videos
Why this training
What changes compared to switching between two tabs.
One does it, the other checks it
The second assistant receives the same briefing and the real artifact, never a vague summary. Debate lasts at most two rounds; the tests decide.
Everything through the subscription
codex exec, claude -p, the Codex plugin, and claudex run through each assistant's subscription, with no API key.
A common link in Markdown
AGENTS.md, tasks, and handoffs live in the project. Any runtime reads them, and the work is not tied to a single model.
From beginner to advanced
Course 1 has 15-minute lessons for people who are just starting; course 6 splits the work across three models.
Questions
Before you start.
Do I need both subscriptions?
For courses 1 and 3, only the Codex one. To use both together, you need Claude Code and Codex CLI.
Do I need to follow the order?
The suggested order is 1 → 6. If you already use Codex, you can start with course 2; if you only want the automatic plan debate, you can jump to course 4.
What is the difference between claudex and Use Both?
claudex is software: it automates the debate over a plan in rounds. Use Both is a method for everything else: reviewing a diff, a goal with a stopping condition, and handoff. They complement each other.
How long does it take?
There are 126 lessons and modules across 6 courses: 18 lessons in course 1 and, in the others, 18, 24, 24, 24 and 18 modules.
What is the best first step?
One small task: write the briefing and the acceptance criterion, ask one AI for the plan and the other for the critique, on the same file. Then do a handoff before closing the session.
Your next plan can arrive reviewed by another AI before the first line of code.
Start with course 1. In 18 short lessons you master the concepts that make Codex work for you.