The idea that runs through the course
Reverse engineering isn’t a product; it’s a discovery technique.
Videos about Claude Code "mods" showed Claude using Codex and several sessions working in parallel, almost entirely through reverse engineering. The idea is great; the approach breaks with the next update. Today, almost all of that has an official method: codex exec, claude --bg e claude agents, official Claude Code plugins.
Kit rule: move up the ladder and always use the most stable available route — API → MCP → CLI → SDK → computer use → local bridge → reverse engineering (lab only).
Who it’s for
Clara, 47 — clinic owner
The schedule is a spreadsheet. She wants the agent to tell her which time slots are available for Dr. Ana and Dr. Bruno without anyone opening the file—and without the agent booking anything on its own.
Sônia, 51 — accountant
The client's ERP has no API; it only exports CSVs. She wants the total sales per customer checked, without manually building a spreadsheet every month.
If you already use an AI chat, have heard of Claude Code and Codex, and want them to work with the systems you use every day, this course is for you. You don't need to be a programmer: each technical term is explained the first time it appears, and every command comes with what you should see on screen.
The 4 tracks
🧭 Discover
Why reverse engineering isn't a product, installing the kit, running the diagnostic, the path ladder, and the capabilities map.
🔌 Connect
Claude uses Codex, its first MCP bridge, the bridge to your system without an API, and the browser with policies.
⚙️ Route and execute
The right model for each task, a three-role team, a background team, and a long-running agent that finishes only with proof.
🛡️ Govern and learn
N0–N4 autonomy policy, the agent proposes and you approve, saves and dashboard, lab, and final project.
🏁 Final project: your bridge and your team
You choose a system from your work that doesn't have an API, fill in the CAPACIDADES.md, chooses the highest possible route, builds the bridge, runs a team of three roles with an "ask first" policy, and delivers the output from the verificar.mjs with all criteria marked OK, plus one lesson you approved yourself.
The deliverable is the folder runtime/ filled in, not a video or a presentation. Details in module 4.4.
What you need
On your machine
Node 18 or later and Claude Code and/or Codex signed in with a subscription. Linux or Mac; on Windows, use WSL.
Before, if you can
The courses Agentic OS (the concept of an agent operating system) and 7 Principles of agent management (autonomy N0–N4). They help, but aren’t required.
How to study here
Each topic has a button. Your progress appears here, in the learning path, and in the module.
"I have a question" saves the topic for you to review later.
Select a passage of text to highlight or add a note.
Progress, questions, and notes in one place. Export them to a file so you don't lose them.
Everything is saved only in your browser. Nothing is sent to any server.