Know what to post today, repeat what already worked, create a post in minutes, and deliver material to commenters. An agent can handle the routine left over.

This page brings together five pieces that already exist separately at INEMA (the free AI club at inema.club) and shows how they fit together. There are four free tools that run in your computer's terminal with help from Claude Code (an AI assistant that works there), and one short course. They're for people who create content (on their own or at a small company) and get lost between choosing a topic, producing, publishing, and replying. Each piece handles one step: finding today's topic, seeing what has already worked on your channels, building the carousel, delivering material to commenters, and handing the rest of the routine to an agent. Each has its own guide and works on its own. You can use one or all five together.
These are the four tools INEMA uses every day, and they work as shown here. The radar only covers AI topics. Generated text (headlines, ideas, guides, and replies) is in Portuguese, sometimes in the voice of someone who teaches AI. The carousel comes with INEMA's brand, colors, and cover art, and the lead-comenta page has INEMA in its title and footer. Changing the name, @, and hashtags is easy (in the carousel's PERFIL.md). Changing colors, language, and the rest means editing files, which Claude Code can help you with. If you create AI content in Portuguese, they work as shown. In the carousel, the final slide suggests an inema.club link as the free material (swap it for yours), and the cover carries the INEMA.CLUB brand; don't ask it to "post", because its automatic posting uses INEMA's account.
The problem isn't a lack of ideas. It's having to decide everything from scratch every day. Here, every tedious decision becomes a command.
Topic, idea, post, contact, and routine. Each tool does one thing and delivers a result you can check.
The tools read public sources (news sites and the public YouTube feed) and use Claude Code or Codex through your subscription.
Automation handles the preparation: topics, ideas, slides, and replies are ready. Nothing is posted or sent on its own; you review and decide what goes out.
The pieces follow the natural order of the work. One piece's output becomes the next one's input, but none depends on another to run.
Open each piece's guide to install and use it. All the guides are in Portuguese, English, and Spanish.
Reads around 20 AI news sources (company blogs, press, Hacker News, and Reddit) and the GitHub projects that gained the most stars that week. Groups the same story from different sources, scores it, checks whether the link opens, and returns the 6 hottest topics from the past 24 hours. If you ask, it suggests a headline, hook, and format in Portuguese.
Requirements: Python 3. Claude Code is optional (only for headlines, which are in Portuguese). The "Trending on GitHub" source is included only if gh is installed and logged in; without it, the radar still works with news sites.
Looks at your YouTube channels and shows which recent videos are performing well above each channel's usual level. It calculates how many views per hour each video gets and compares that with a typical video on the same channel. Videos that pass 2 times the usual level become "winners" and, if you ask, turn into 8 new video ideas.
Requirements: Python 3. Ideas use Claude Code through your subscription. It doesn't use the YouTube API.
A Claude Code skill that creates Instagram and TikTok carousels (the same images also work on Facebook). Give it a topic and get a cover, 5 to 10 slides with one idea each, a final slide saying "comment WORD," and 3 stories, all at the right size. Before showing you, it checks whether any text is cut off or too small.
Requirements: Linux or Mac (on Windows, use WSL, the Linux environment that runs inside Windows), Node.js 18 or newer, Python 3 with Pillow, Chromium, and Claude Code. For cover art, use Codex signed in to your ChatGPT account. The PERFIL.md file comes with INEMA's brand (name, @, sign-off, and hashtags) and Portuguese text: replace them with your own before making your first carousel.
Creates the "free material" page you promise when you say, "comment WORD and I'll send it to you." Give it a word, a topic, and a source (a page or a text). It writes a short 4 to 7-step guide using only what's in the source, creates the page, and prepares replies for the comment, direct message, and story.
Requirements: Python 3, Claude Code, gh, and a GitHub account. The page is hosted for free with your account's GitHub Pages. The guide and replies are in Portuguese, and the generated page has an INEMA.CLUB credit in the footer that the tool doesn't yet let you change.
Weekly reports, new contacts, invoices, new clients: tasks that take hours and don't require genius. In 8 lessons of about 15 minutes each (around 2 hours in total), you'll find the first task to delegate, write the agent's profile, review drafts before anything is sent, choose the right model for each task, and leave with a 30-day plan.
Requirements: Just read and practice. Exercises work in chat (ChatGPT, Claude, or Gemini) or in the terminal, with no paid API.
Each piece has its own requirements; check its card. This is the complete setup for all four tools on a Linux, Mac, or Windows computer with WSL (the Linux environment that runs inside Windows). Commands run in the terminal, the window where you type commands. Each tool's guide has more detailed installation steps.
Python runs radar-pautas, muse-canais, and lead-comenta. Git downloads the projects. gh is GitHub's command-line tool: it publishes the lead-comenta page and enables radar-pautas's "Trending on GitHub" source. If Ubuntu can't find the gh package, follow the instructions at cli.github.com.
# Ubuntu or WSL sudo apt install python3 git gh # Mac brew install python git gh
Node and Chromium (a browser that runs without a window) render the slides; Pillow (a Python image library) helps with checks. Node also installs Codex. On Ubuntu, the last command installs the libraries Chromium needs; without them, the carousel's setup.sh fails.
# Ubuntu or WSL sudo snap install node --classic sudo apt install python3-pil npx playwright install chromium sudo npx playwright install-deps chromium # Mac brew install node brew install pillow brew install --cask google-chrome # on Mac, the carousel uses Google Chrome
On Ubuntu, npx may ask "Ok to proceed?": answer y.
Claude Code is Anthropic's assistant that works in the terminal; here it writes headlines, ideas, slides, and text. Install it from the official site at claude.com/claude-code. You need a paid Claude plan (Pro or higher) and must sign in once. Codex is OpenAI's equivalent: it only creates the carousel cover art through your ChatGPT account. None of the tools use an API key.
# opens and asks you to sign in the first time claude sudo npm install -g @openai/codex # optional, only for the cover (on Mac, omit sudo) codex login
The commands below are the first ones in each guide. Each piece's guide explains the options and settings.
The radar shows the 6 topics without saving anything. Choose one.
git clone https://github.com/inematds/radar-pautas ~/projetos/radar-pautas cd ~/projetos/radar-pautas python3 radar.py --dry # 6 topics, nothing saved python3 radar.py --angulos # saves + headline, hook, and format (Claude Code)
With --angulos, the result is saved to ~/projetos/output/radar-pautas/<date>/radar.html; open that file in your browser. The suggested format may be a carousel or an INEMA video format; outside INEMA, use just the headline and hook.
Download muse-canais and find your channel ID (it starts with UC), replacing the address with your channel's:
git clone https://github.com/inematds/muse-canais ~/projetos/muse-canais cd ~/projetos/muse-canais python3 -c "import muse; print(muse.resolver_id('https://www.youtube.com/@mychannel'))"
The canais.json file comes with INEMA's channels. Open it in an editor (nano canais.json), delete everything, and leave only your channel in this format (save with Ctrl+O, Enter, and Ctrl+X):
{"mychannel": {"nome": "My channel", "url": "https://www.youtube.com/@mychannel", "id": "UC..."}}
Then run muse-canais. The report is saved to ~/projetos/output/muse-canais/<date>/muse.md.
python3 muse.py --ideias # report + 8 video ideas
If no recent video passes 2 times the channel's usual level (or has fewer than 50 views), the report won't have winners or ideas. That's normal: run it again after publishing more.
Install the skill once (a skill is a package of instructions and scripts that teaches Claude Code to do a task). Then open Claude Code and ask: "make a carousel about <chosen topic>". It shows you the finished images to approve; you post them yourself, by hand. Before the first one, replace INEMA's brand with yours in PERFIL.md. You can also ask Claude Code: "fill out PERFIL.md with me".
git clone https://github.com/inematds/carrossel-inema ~/projetos/carrossel-inema cd ~/projetos/carrossel-inema bash setup.sh # must end with SELFTEST PASS mkdir -p ~/.claude/skills ln -s "$PWD" ~/.claude/skills/carrossel-inema # connects the skill to Claude Code
Before posting "comment WORD," the material page needs to be live in your account on GitHub. Otherwise, replies will include a link that isn't yours. You only need to do this once: sign in to GitHub, create your copy, enable GitHub Pages (GitHub's free service for hosting pages), and save your address. Replace YOUR-USERNAME with your username, which gh api user displays. The --scopes workflow option gives the login permission to publish the page automatically; the lead-comenta guide explains each login prompt.
git clone https://github.com/inematds/lead-comenta ~/projetos/lead-comenta cd ~/projetos/lead-comenta gh auth login --scopes workflow git config --global user.name "Your Name" git config --global user.email "you@example.com" gh api user -q .login # your username gh repo create lead-comenta --public --source . --remote meu git push meu main gh api -X POST repos/YOUR-USERNAME/lead-comenta/pages -f build_type=workflow
Save your address so it's available in every new terminal. On Ubuntu or WSL:
echo 'export LEAD_BASE_URL=https://YOUR-USERNAME.github.io/lead-comenta/paginas' >> ~/.bashrc source ~/.bashrc
On Mac:
echo 'export LEAD_BASE_URL=https://YOUR-USERNAME.github.io/lead-comenta/paginas' >> ~/.zshrc source ~/.zshrc
Your copy comes with the example word AGENTE, which points to INEMA's address; always use a word of your own. For each post (replace TIPS, the topic, and the source with your own):
python3 lead.py nova TIPS --tema "6 tedious tasks an agent can handle" --fonte https://inematds.github.io/empresa-de-ia/guia/en/ python3 lead.py resposta TIPS # text for the comment, direct message, and story git add -A && git commit -m "TIPS guide" && git push meu main # publishes the page
The page takes a few minutes to appear. Open the link shown by lead.py resposta before posting the invitation.
Reports, new contacts, invoices, new clients. The course shows how to set up the agent's profile and review its work before sending. The 30-day plan in the banner is the course's final lesson.
# in your browser https://inematds.github.io/trabalho-chato-ao-agente-v72/en/
Trying to set everything up on day one puts you back where you started. One piece per week works better.