🌅 Build: morning routine that improves itself

The morning routine is a scheduled task (the one that runs automatically at a set time, without you having to open anything — covered in depth in module 6.2) that runs every morning, for example at 7h. When it runs, the agent checks what matters for your day — important emails, calendar appointments, or a summary of news/data you're interested in — and leaves a summary ready for you to read as soon as you wake up, without having to hunt for information in three different places.

The interesting part isn’t just "running every morning" — any alarm clock can do that. What makes this routine a A real build is that it improves on its own: uses the status file (the same pattern as module 6.4, "the routine that learns") to remember what it showed yesterday and avoid repeating it, and to adjust what it prioritizes based on what you found useful.

Caption: the cycle closes in a loop — the routine doesn’t just run; it records what it did so the next run can start smarter, without repeating what you’ve already seen.

New here? O status file is a plain text file (format JSON — an organized way to store information in "key: value" pairs, easy for both you and the agent to read) that keeps track of what happened in recent runs. It's the routine's "short-term memory" — without it, every morning would be the first morning all over again.

Three things this file needs to store: last run (date and time, to know how much time has passed), what has already been shown (so it doesn’t repeat the same email or news story two days in a row), and learned preferences (what you tend to ignore and what you tend to pay attention to).

Crie o arquivo rotina-matinal-status.json na raiz deste projeto com esta estrutura inicial:

{
  "ultima_execucao": null,
  "mostrado_recentemente": [],
  "preferencias_aprendidas": {
    "prioriza": [""],
    "ignora": []
  },
  "fontes": [""]
}

Depois de criar, me explique em 2 frases o que cada campo vai guardar.

With the status file ready, the next step is the trigger: the scheduled task itself. It needs a time (e.g., 7 a.m.), a frequency (every day, or only weekdays), and a clear instruction for what to do when it triggers — including an explicit request to read and update the status file.

Agende uma tarefa recorrente todo dia às 07:00 ()
chamada "rotina-matinal". Quando ela rodar, ela deve:

1. Ler rotina-matinal-status.json
2. Checar 
3. Ignorar tudo que já esteja em "mostrado_recentemente"
4. Priorizar de acordo com "preferencias_aprendidas"
5. Escrever um resumo curto (até 8 linhas) em resumo-do-dia.md
6. Atualizar rotina-matinal-status.json com a nova execução

Antes de salvar a tarefa agendada, me mostre o texto final da instrução pra eu revisar.

New here? Restrictive permission (covered in depth in module 6.6) means giving the agent only the access needed for the task — in this case, reading email/calendar and writing to the summary file, but never delete anything. A routine that runs on its own every day needs this safeguard: if one day it "misunderstands" what to do, the worst-case scenario is a bad summary — never a lost file.

A notification it’s how you find out that the routine ran—without having to open the terminal every morning. It could be an email, a message on your phone, or simply the file resumo-do-dia.md waiting for you in the folder — the important thing is that the notification reaches you in a way you'll actually see.

Ajuste a tarefa agendada "rotina-matinal" para rodar com permissões restritivas:

- Permitido: ler e-mail/agenda, escrever em resumo-do-dia.md e rotina-matinal-status.json
- Proibido: apagar ou mover qualquer arquivo, enviar e-mail/mensagem em meu nome

Ao final da execução, envie uma notificação para 
com o título "Resumo de hoje pronto" e o conteúdo de resumo-do-dia.md.
Se não houver nada relevante, envie mesmo assim avisando "sem novidades hoje".

Don’t wait until tomorrow morning to find out if it worked. Ask the agent to run the task now, once, manually — most scheduling tools have an “run now” option for testing without waiting for the scheduled time. That’s more useful than any theory: you’ll see the actual result in minutes.

Caption: each day, the routine gets a little more precise — not because the model “got smarter,” but because the status file has accumulated more signals about what you use and what you ignore.

“Learning” here isn’t magic—it’s the same cycle as Module 6.4: every time it runs, the agent checks the sources and also looks at whether you engaged with what it showed you yesterday (e.g., you never open the news section, but always read the agenda section). Based on that, it updates preferencias_aprendidas — add to "prioritize" the things you use; add to "ignore" the things you never touch. In practice, after a week the summary gets shorter and more relevant, without you needing to reconfigure anything.

Atualize a instrução da tarefa "rotina-matinal" para incluir esta regra de aprendizado:

Antes de montar o resumo de hoje, leia "preferencias_aprendidas" em rotina-matinal-status.json.
Depois de montar, se uma seção (ex.: ) não recebeu
nenhuma interação nos últimos 5 dias, adicione essa seção em "ignora" e reduza o espaço dela
no próximo resumo. Se uma seção for consistentemente a mais lida, adicione em "prioriza" e
mostre-a primeiro. Sempre registre essa decisão em "mostrado_recentemente" com a data.