Plan kit · nutrition practice · open source

The complete plan for patients to follow the plan

Specification, 99 acceptance tests, guardrails and prompts for an agent to build, in your own environment, the patient care system of a nutrition practice: in-person and online scheduling with confirmation and follow-up, a meal plan with versions, a diary for meals, water and weight, opt-in reminders and a Telegram team desk. The bot never prescribes. The system is not implemented yet: this repository is the plan, and you run the implementation.

Start the guide View on GitHub
Atende Nutrição banner: in-person and online scheduling, meal plan, diary, opt-in reminders and a Telegram team desk
What it is

A kit to hand to the agent: from the contract to 99 green tests

There is no ready-made app here. There is the contract for what the app must do, the test suite that proves it, and the guardrails that stop the agent from cheating. The builder is Claude Code or Codex, running on your machine, on your subscription. Its sibling Atende Clínica, which this kit is based on, was built exactly this way and closed 93 of 93 tests.

In-person and online scheduling, versioned meal plan, diary of meals, water and weight, opt-in reminders, 192 and 188 alerts, health-data privacy and a Telegram team desk

🥗 The system the plan produces

Scheduling for first visits and follow-ups, in person or online (the team supplies the link), with no double booking. A meal plan entered by the nutritionist, with versions, and a bot that answers only what is in that patient's plan. Python 3 with the standard library only, plus SQLite, one practice per installation.

📦 The kit

A 21-section specification, 99 black-box tests, 19 proposed decisions, a map to Raio-X, prompts for /goal and for the headless loop, and the guardrails that freeze the contract. An adversarial validator went over everything and fixed 10 problems, 1 of which would have blocked execution.

📈 Source: Raio-X de Margem

No-shows, follow-ups that never happen and idle slots are the leaks that scheduling attacks. The system returns atendimentos_mes, falta_pct, ocupacao_pct, clientes_novos_mes and retorno_atual_pct for the Raio-X de Margem dashboard, which is the source of the business rules.

🚨 Alerts in two lists

A physical sign (chest pain, fainting, vomiting that will not stop, hypoglycemia) is answered with 192 and the emergency room. Eating disorders and self-harm (binge eating, purging, laxatives, suicidal thoughts) are answered with 188 (CVV) and the nutritionist. Both open a high-priority queue entry, even while a human chat is open. 192 and 188/CVV are Brazilian emergency numbers: in another country, the practice must swap in its local emergency contacts.

🔒 Health-data privacy (LGPD)

The plan, diary and weight are sensitive data: a patient can export everything, and deletion erases it (it does not merely anonymize). "PARAR" (stop) blocks everything automatic. Campaigns need specific consent. LGPD is the Brazilian data protection law: in another country, the practice must apply its local health-data law.

⚠️ CFN rules: not researched yet

Raio-X researched the medical, dental and physiotherapy councils (CFM, CFO and COFFITO), not the nutritionists' council (CFN). The campaign guardrail is only the common minimum (promising results, before-and-after). The owner must confirm the CFN rules before publishing any campaign.

How it works

From "I want to book" to the everyday diary

This is the flow of the planned system, what the agent will build. A dependency-free Python server, SQLite in a data folder and Docker for the VPS. Evolution and Telegram come in only through environment variables; without them the outbox is simulated and no network call leaves the machine.

Site chat or WhatsApp (Evolution)→ Book: in person or online→ Confirmation 1/2 and waitlist→ Meal plan sent by the team→ Diary and opt-in reminders→ Scheduled follow-up and monthly numbers for Raio-X

Patient

Through the chat or WhatsApp the patient books, confirms with 1 or 2, asks for the online visit link, checks meu plano (my plan) or o que posso comer no lanche da tarde (what can I eat at the afternoon snack), logs 2 copos (2 glasses), pulei o almoço (I skipped lunch) or peso 82,5 (weight 82.5), gets FAQ answers, asks for a human and sends "PARAR" when they want no more messages. The bot's chat commands are in Portuguese.

Team on Telegram

Notices of new and cancelled bookings, and the commands /agenda, /alertas, /fila, /responder and /encerrar, also as a direct reply to the patient's message.

/equipe page

With a token: schedule, patients, meal plan, diary and adherence, weight series, FAQ, records, blocked time and the human queue.

Meal plan

Meals by time of day, with options and notes, in versions (the latest is the active one). The team reviews it and sends it to the patient. Outside the plan, "can I eat chocolate?" becomes "I don't advise outside the plan" and goes to a human. There is no nutrition calculation in v1.

Diary, weight and goals

Glasses of water against a goal, meals done or skipped, a meal photo (metadata only, no image is downloaded) and a weight series with the change. The reply to a weight is neutral and non-judgmental: the system never comments on progress.

Opt-in reminders

The team sets up, at the visit, reminders for water and for the plan's meals, at the times the patient chose. There is a tolerance window: a late reminder does not pile up. parar lembretes (stop reminders) turns off reminders only.

Prerequisites

What you need

To run the plan you only need Python for the tests and an agent logged in on your subscription. Docker is only needed for the final check and the deploy; Evolution and Telegram are optional and yours.

Python 3.10+ and pytest

The tests use pytest (a development tool). The application itself will have no dependencies.

# check
python3 --version
python3 -m pip install pytest

Claude Code or Codex

On your subscription, no paid API. For the headless loop, clone execucao-longa into ~/projetos.

# check
codex --version   # or: claude --version

Docker, Evolution and Telegram

Docker does the final build and the VPS deploy. An Evolution instance and a Telegram bot are needed only if you want those channels; credentials live in .env.

# check
docker --version
User guide · step by step

From clone to a working system, in your own environment

The commands below are the repository's own. The cycle: answer the decisions, freeze the contract, run the agent until 99 passed and LIMITES OK, and check with the independent verification.

1

Clone the repository

There is no ./atende yet: the repository only has the plan. Check that the suite is collected without errors and that nothing passes yet (without the implementation the tests fail or error out, which is expected).

git clone https://github.com/inematds/atende-nutricao && cd atende-nutricao
python3 -m pytest -q --collect-only | tail -n 1   # 99 tests collected
python3 -m pytest -q | tail -n 1                  # 15 failed, 84 errors (0 passed)
2

Answer the open decisions

Read docs/DECISOES-ABERTAS.md: 19 default proposals already applied in the specification and the tests. Answering "ok em tudo" (all fine) unlocks the run. If you change anything marked ⚠ (for example the campaign guardrail, the alert words or the weight rules), edit docs/ESPECIFICACAO.md and the tests before freezing. For a real pilot, also replace exemplos/nutricao.json with your own hours, services, FAQ and sample meal plan. Also answer the CFN question, described in the notice box below. The docs are in Portuguese.

less docs/DECISOES-ABERTAS.md
3

Freeze the contract

With everything committed, the script records the hash of tests/, pytest.ini, the specification and the two verifiers, plus the base commit in hash-congelado.txt, and makes one commit. From then on any change to the tests is detected. If there are pending changes, the script stops and asks for a commit first.

git add -A && git commit -m "contrato v1"
bash longrun/2026-10-05-nutri-v1/congelar.sh
4

Run the agent (pick one path)

a) Headless loop with Codex (recommended). loop.env sets 20 cycles of 30 min, 8 GB per cycle, a stop after 3 cycles with no progress, model gpt-6-astra and CODEX_ARGS="-c sandbox_workspace_write.network_access=true": without it the Codex sandbox blocks the tests' local server.

~/projetos/execucao-longa/tools/loop-longrun.sh longrun/2026-10-05-nutri-v1

b) /goal in Claude Code. Open a new Claude Code session in the project folder and follow longrun/2026-10-05-nutri-v1/prompt-goal-claude.md: paste the /goal condition (the output must show 99 passed and LIMITES OK) and, as the first message, the block from prompt-goal-codex.md starting at RESULTADO:.

claude   # new session, inside atende-nutricao

c) Codex TUI. Paste the contents of longrun/2026-10-05-nutri-v1/prompt-goal-codex.md.

codex -c sandbox_workspace_write.network_access=true
5

Follow along

In the loop, loop.log shows each cycle and progress.md has one line per checkpoint. The loop's exit code tells you what happened: 0 done (the final test passed), 1 cycle limit reached, 2 stopped after 3 cycles with no progress, 3 another loop is already running in the folder. A test that conflicts with the specification goes to failures.md: it is a human gate, and the agent must not adjust a test or the specification.

tail -f longrun/2026-10-05-nutri-v1/loop.log
cat longrun/2026-10-05-nutri-v1/state.md
6

Check with the independent verification

When the agent says "done", run the three commands. The last one is something the agent has never seen: it looks for fixture values copied into the code, brings up a practice that never appeared (20-minute step, 40-minute visit, follow-up after 15 days, no online visits, a meal plan with differently named meals, another clock) and runs docker build with a healthcheck. Then open / and /equipe in the browser.

python3 -m pytest -q tests/                                         # 99 passed
bash longrun/2026-10-05-nutri-v1/verificar-limites.sh               # LIMITES OK
python3 longrun/2026-10-05-nutri-v1/verificar-independente.py      # INDEPENDENTE OK
7

Run your practice and deploy to the VPS

With the system implemented, copy the example, replace TROQUE-ESTE-TOKEN and start the server. The patient chat is at / and the team page at /equipe (header X-Token).

mkdir -p dados && cp exemplos/nutricao.json dados/nutricao.json
./atende serve --porta 8080 --dados dados

The deploy is yours, with your credentials. The agent itself writes the deploy README as part of the goal: bring up docker compose (port only on 127.0.0.1:8080, behind an HTTPS proxy), the Evolution webhook at /webhook/evolution/<secret> (event MESSAGES_UPSERT), the Telegram webhook with setWebhook and secret_token, and the daily backup (./atende backup in cron).

cp .env.exemplo .env && chmod 600 .env   # fill in EVOLUTION_*, TELEGRAM_*, WEBHOOK_SEGREDO
docker compose up -d --build

⚠️ Before publishing any campaign: the CFN

The advertising rules for nutritionists, set by the Brazilian Federal Council of Nutritionists (CFN), have not been researched in this kit yet. The campaign guardrail (section 11 of the specification) is the minimum common to the medical, dental and physiotherapy councils: it blocks promessa_resultado (guaranteed, guaranteed result, 100%) and antes_depois (before and after). No CFN article is cited, on purpose. The practice owner confirms the CFN rules (price, promotions, testimonials, before and after) before publishing any campaign; if there is an extra rule, it enters as a new code and changes the contract (decision 4 in docs/DECISOES-ABERTAS.md).

What's in the kit

Contract, tests and guardrails, with the real numbers

Everything the agent needs to build, and everything that stops it from pretending it built.

99 acceptance tests

Black-box, over HTTP and the command line, in 11 files.

test_agenda.py                16
test_integracoes.py           15
test_conversa.py              13
test_cadastros.py             10
test_diario.py                10
test_lembretes.py              9
test_plano.py                  8
test_basico.py                 6
test_campanhas.py              5
test_docker.py                 4
test_lgpd_raiox.py             3

A 21-section specification

docs/ESPECIFICACAO.md: execution, nutricao.json, scheduling rules, authentication, HTTP API, conversation, waitlist, confirmation and reminders, meal plan, diary and weight, campaigns, export to Raio-X, LGPD, what stays out of v1, pages, records, schedule management, database, Evolution, Telegram and Docker. Nutrition calculation, photo analysis, payments, insurance billing and multiple practices are explicitly out.

Anti-shortcut guardrails

A frozen hash of tests/, pytest.ini, the specification and the two verifiers. File scope: the agent only touches the code and its own records. No external dependency, no API key, no external URL, no media download and no 0.0.0.0 in the code. verificar-limites.sh checks all of that and prints LIMITES OK.

Independent verification

Level 4, hidden from the agent: it looks for fixture values in the code, brings up a practice it has never seen, with every argument explicit, and runs docker build and a container healthcheck. Only that makes "99 passed" prove general logic.

19 decisions already proposed

Language, channels, the campaign guardrail, what the bot answers, alert signs, neutral weight replies, opt-in reminders, the meal plan, photos as metadata only, online visits, follow-ups, LGPD, Raio-X and one practice per installation. Only the items marked ⚠ change the contract.

Validated by an adversarial agent

A validator that had not seen the planning checked tests against the specification, redid the arithmetic and tried to get around the guardrails. It found and fixed 10 problems: 1 blocked execution, 6 got in the way and 3 were cosmetic. The full account is in docs/VALIDACAO.md and each fix has a line in FALHAS.md.

The business rules come from Raio-X

No-shows, follow-ups that never return and idle slots: docs/MAPA-RAIO-X.md links each leak in the clinica package to a v1 piece or to the reason it stays out. More in the Raio-X de Margem guide.

Brazilian rules built in

LGPD for health data (confirmation, follow-up, reminders and the plan as health care; campaigns only with consent; export and erase) and Brazil's emergency contacts: 192 and 188 (CVV). All of these are Brazilian: in another country, the practice must swap in its own emergency contacts and privacy law. The CFN rules are still to be confirmed, as in the notice box at the end of the user guide.

Roadmap

Where it stands and where it is going

The plan is ready and validated. The implementation does not exist yet: it is born when you run the /goal or the loop, using the execucao-longa method. Atende Clínica followed the same path and closed 93 of 93.

Plan ✅
Specification, tests and guardrails validatedA 21-section contract, 99 acceptance tests, proposed decisions, a map to Raio-X and an adversarial validation completed on 05/10/2026.
Build
You run the /goal or the loopAnswer the decisions, freeze the contract and let the agent work until 99 passed and LIMITES OK, then the independent verification. Today the system is not implemented.
Pilot
In a real practiceSet up real hours, services, FAQ and sample meal plan, confirm the alert words and the CFN rules with the nutritionist, publish on the VPS and track no-shows, occupancy and follow-up in the Raio-X Recovery Dashboard.
Later
Outside v1Nutrition calculation, photo analysis, payments, insurance billing, multiple practices, an LLM in the replies and a UI in EN/ES are left for a later version, behind a human gate.