One brand voice file + 3 commands. Claude researches, writes, generates the visual, and schedules. Nothing goes live without your "yes."

Adaptation of “Social Autopilot” for the INEMA stack: everything runs inside Claude Code, with providers, brands, and networks configurable in a single config.yaml.
One brand-voice.md by brand: who you are, your audience, your offerings, real phrases, true stories, and rules for each network. That’s what makes the post sound like you.
Publishing via Metricool (validated), Blotato, or manual copy. Image with flux2-klein or Magnific. Switching providers is one line in the config.
Every publication goes through the gate yes / draft / no. Nothing is scheduled without your explicit reply in the conversation.
The daily workflow of /ms-post. The weekly one (/ms-semana) does the same for 5–7 posts at once, with a retrospective based on analytics first.
Voice interview (one question at a time), active networks, and providers. Once per brand.
Today's post. Three questions, three options, a visual, and scheduling.
Last week's retrospective, a table of 5-7 posts to approve, scheduled queue. ~15 min.
A transcript, article, or URL becomes a standalone post for each network.
Only Claude Code is required. Without a publishing provider, the skills deliver ready-to-copy text.
Pro plan or higher. Open it inside the project folder: it reads the CLAUDE.md on its own.
# clone and open git clone git@github.com:inematds/ms-v7.git ~/projetos/ms-v7 cd ~/projetos/ms-v7 && claude
Schedule Instagram, TikTok, and more. Free plan: 20 posts/month, no LinkedIn/X.
claude mcp add --transport http metricool \ https://ai.metricool.com/mcp -s user # later: /mcp → metricool → Authenticate
MCP providers only accept direct URLs. The default is a GitHub Release; the skills upload automatically.
gh release upload v1 arquivo.png \
--repo inematds/midia --clobberEnter all commands below inside Claude Code, in the project folder. Questions are always asked in free text, one at a time.
Create brands/<slug>/brand-voice.md with an interview and fills in the config.yaml. If the brand is already described somewhere, say where: it pre-fills the details and only asks what’s missing.
/ms-config # voice → active networks → publishing provider → voice test
Answer with the network, topic, and goal. It researches what's working on the network, writes 3 options with a story or number from your file, and you choose.
/ms-post # 1. Network? → instagram # 2. Topic? → “why read the system prompt” (or “research what’s trending”) # 3. Goal? → followers
The visual comes from the configured image provider. Then it shows you the final post with the best time and waits for your response.
# Visual OK? → yes # Schedule? (yes / draft / no) sim # schedule with automatic publishing rascunho # stays in the provider’s planner; it doesn’t publish não # text only, ready to copy
Retrospective of the last 7 days via analytics, three questions for the week, a table of 5-7 posts for you to edit, and the entire queue scheduled with a single "yes."
/ms-semana # retro → 3 questions → table → “Edit anything or should I approve?” → queue → “Schedule the queue?”
Paste the transcript, file path, or URL. The only question: which networks. Each post stands on its own, never says “in this video.”
/ms-reaproveita https://youtube.com/watch?v=... # or an .md, .txt, article URL
The skills commit at the end of each post. You just push; the history stays in posts/.
git push # author inematds <inematds@gmail.com>
O config.yaml decides who does what. The brand-voice.md decides how it sounds.
# config.yaml (excerpt) marca_ativa: inema marcas: inema: redes: instagram: { conta: "@inemafuturos", ativa: true } tiktok: { conta: "@lnema.club", ativa: true } linkedin: { conta: null, ativa: false } publicacao: provedor: metricool provedores: imagem: { padrao: flux2-klein, fallback: magnific }

posts/<data>-<slug>/post.md with status, ID, and approved visual.Version 1.0.0. The manual workflow runs first; cron automation only comes after weeks without failures.