You collect videos from here and there and don’t know where to start or what comes next. The complete course with chapters puts everything in order, from basics to advanced topics, and each chapter has a free kit you can use the same day.

Claude Code is Anthropic's AI assistant that works with the files on your computer: it reads, writes, organizes, and runs tasks for you. This is a complete course with chapters, in a single free video that takes you from the Claude map to agents that prove what they do, work safely, and run on their own. Each chapter has a step-by-step guide and a ready-to-copy prompt. All you need to watch is a browser. The repository also includes the tool that assembled this video, so you can combine your own short videos into a long course.
The short video introducing the course is in production and will appear here as soon as it’s published. The full course is right below.
Watch it all at once or skip to the chapter you need: on YouTube, the video progress bar is already divided into chapters.
Watch on YouTube (video in Portuguese): youtube.com/watch?v=E5ZumaYjzT8
Click the time to open the course at that point. Each chapter has a complete guide to the topic (free) and a ready-to-copy prompt. The prompts for chapters 1, 2 and 5 go in the Claude chat (website or app); the others go in Claude Code, opened in your work folder, where the files it will read are.
Claude's guide for people who don't code: models, projects, memory, connectors, skills, Cowork, and Claude Code, and what each plan includes.
I work as <your profession>. Of Claude's pieces (projects, memory, connectors, skills, Cowork, Claude Code), which ones am I probably not using yet, and which one should I turn on first? Give me a 5-minute step by step to start with it.Documents, presentations, and design pieces made with your branding, from a Project containing the text "who I am" (most of it works even on the free plan).
Write me a "who I am" text to paste into a Claude Project: my business is <describe>, my audience is <describe>, my colors are <colors> and my tone of voice is <describe>. Then make a 5-slide presentation about <topic> using that text.What Claude Code does on your computer and the most useful new features for people who don't code: subagents, background tasks, dashboards, and more, with guidance on each one.
Look at this folder and tell me, in plain language, what is in it. Then suggest 3 repetitive tasks from my day that you could do in this folder, and do the first one only after I say ok.How to read the logs that Claude Code and Codex already save on your computer to see which project or routine used up your subscription's weekly limit.
Download the medidor-de-limite kit (https://github.com/inematds/medidor-de-limite), follow its README to run the meter on this machine and explain the result: which project or routine used the most of my limit in the last 7 days, and what I can merge or space out to use less. Do not delete or pause anything.Eight writing phrases that make AI show what it did (the file, the output, the test) instead of just saying “done,” in ChatGPT, Claude, or Gemini.
Before saying you are done, run the check and paste the last line of the output here. If you cannot prove it, say what is missing instead of saying "done".The request template that lets the agent say "done" only when it has proof, and the checker that scores your request from 0 to 8.
Rewrite this request in the prompt-with-proof format: goal, how you will prove you did it, what counts as done, and what to do if it is not possible. My request: <paste here>.The steps that help an agent stop repeating the same mistake: record the failure, separate a request error from a machine error, and turn the lesson into an automatic safeguard.
Look in this project for where I note the mistakes you made and the fixes (for example FALHAS.md or LESSONS.md). If there is none, create a FALHAS.md with one line per mistake I correct from now on. Then show me the mistakes that come back most often and, for each one, suggest an automatic lock (a test, a hook or a check) instead of one more written rule. Do not apply anything without my approval.The 4 rules for having the agent study an app and build its own version without making up anything it didn't see, with an evidence file that a checker fails when evidence is missing.
Study this app and write what it does in a proof file: for each feature, where you got it from (file, screen or output). Whatever you did not see, mark as "not proven" instead of guessing.Why the agent can read your passwords and how to make it work in a disposable sandbox, without a network connection or your secret files.
List which files in this folder look like secrets (.env, keys, tokens) and tell me how to run you here inside a disposable box, with no network and without those files. Do not open the contents of the secrets.The three-layer setup to keep from losing your code if your account is blocked: a copy with another company, dated snapshots on a drive of your own, and, if you want, your own server.
List my git repositories on this machine and build a three-layer backup plan: another hosting company, a dated snapshot on a disk of mine, and the command to check the copy is complete. Show the commands before running them.A bot dashboard with names, roles, memory, and scheduled tasks, powered by the Claude or Codex subscription you already pay for, with no API key.
I want a team of 3 bots for my work as <profession>: write each one's role in one sentence, each one's scheduled task, and what each one must never do without asking me.A single screen that shows what your automated routines did overnight, what failed, and what’s waiting for you, read aloud in Portuguese.
List the automatic routines on this machine (timers, cron, queues) and tell me, for each one, when it last ran, whether it worked and where its log is. Read only: do not run or stop anything.Our what's trending radar tracked 320 videos across 7 AI topics (November 12, 2026 run) and compared each video with its channel’s usual performance. Very short and very long videos went further than the middle.
Median of how many times its channel's usual each video did, by length (20, 48, 31 and 33 videos). Source: INEMA's what's trending radar, November 12, 2026.
From here on, this is the technical part: combining your short videos into one long video with chapters. You not needed from this section to watch the course.
O montar_curso.py it's the same tool that put together the course above. You give it the list of videos and titles; it returns the long-form video, chapters, and description.
3 seconds of a black screen with "CHAPTER N OF T" and the title in yellow and white before each video.
Each video goes through audio normalization so no one has to adjust the volume between chapters.
Generates the "00:00 Title" list with the actual timestamps from the final video and a description ready to paste.
Every video is converted to the same size (1920×1080 by default, with borders if the aspect ratio differs), 30 frames per second, and audio at −16 LUFS. The chapter starts on the card itself. At the end, the tool checks that the MP4 duration matches the sum of the parts (within 1 second) and only then reports success.
Everything runs in the terminal, your computer's command window: on a Mac, open the Terminal app; on Linux, press Ctrl+Alt+T. The commands are for Mac and Linux; on Windows, use WSL (Ubuntu inside Windows) and follow the Linux steps.
Check with the command below. If you don't have it, download it from python.org.
python3 --versionThe program that joins and converts the videos. Ubuntu/WSL: sudo apt install ffmpeg. Mac: brew install ffmpeg.
ffmpeg -versionWith git, the command below. Without git: on the kit's GitHub page, click "Code" → "Download ZIP" and unzip it.
git clone https://github.com/inematds/claude-code-completo cd claude-code-completo
Pillow draws the chapter cards. Install it in a separate environment (venv), which works even on systems that block pip directly. Inside the kit folder:
python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt # next time, just run source .venv/bin/activate
Create three test videos and build a 45-second course with them, to see the result before using your own:
python3 ferramentas/videos_de_teste.py python3 ferramentas/montar_curso.py exemplos/meu-curso.json --saida saida/curso.mp4
Copy the example into the kit folder and open the copy in a text editor (TextEdit, Notepad or VS Code). Replace the title, the descriptions and, in each chapter, the file and the title. Use the full path of each video (e.g. ~/Videos/lesson1.mp4): the short paths in the example (videos/aula1.mp4) only work inside the exemplos/ folder. Keep the field names as they are (titulo, capitulos, arquivo, in Portuguese): the tool only reads those.
cp exemplos/meu-curso.json meu-curso.json # {"titulo": "...", "descricao_topo": "...", "descricao_fim": "...", # "capitulos": [{"arquivo": "~/Videos/lesson1.mp4", "titulo": "First steps"}, ...]}
Without spending time rendering, see the estimated chapters and YouTube warnings (at least 3 chapters, each 10 seconds or longer):
python3 ferramentas/montar_curso.py meu-curso.json --saida saida/meu-curso.mp4 --so-capitulos
A 30-minute course takes about half an hour on an ordinary computer. The output includes three files:
python3 ferramentas/montar_curso.py meu-curso.json --saida saida/meu-curso.mp4 # saida/meu-curso.mp4 · saida/meu-curso-capitulos.txt · saida/meu-curso-descricao.txt
Upload the MP4 to YouTube and paste the contents of -descricao.txt in the description. YouTube turns lines like "00:00 Title" into chapters on the video progress bar.
cat saida/meu-curso-descricao.txt
Change the card duration, size, or frame rate; and run the kit's own tests (they require pytest):
python3 ferramentas/montar_curso.py meu-curso.json --saida saida/meu-curso.mp4 --cartao-s 4 --largura 1280 --altura 720 pip install pytest python3 -m pytest -q # 10 tests with synthetic videos (9 + 1 skipped without PyYAML)
The course list above is at curso/curso.json (the original videos are not in the repository). This was the output:
00:00 As 25 peças do Claude 03:37 O Claude como escritório 06:13 13 novidades do Claude Code 09:21 Quem comeu o seu limite 11:53 As 8 frases que pedem prova 14:13 Prompt com prova 16:31 O agente que aprende com o erro 19:23 Refazer um app pedindo prova 22:06 A caixa do agente 24:18 Seus repositórios a salvo 27:44 Time de bots pela assinatura 30:37 O painel dos seus robôs