Canvas, prompts, agent instructions, workflows, integration and governance checklists, metric sheets, and a three-phase plan—ready to copy and use with Copilot, Copilot Studio, Power Automate, and Power BI. Plus two more tools for people building INEMA v5 courses.
Grew out of the course Copilot + Agents for Business (INEMA.CLUB PRO, 9 lessons). The course teaches; the kit provides what you can copy. The tools are what proved useful from producing the course itself.
Markdown, in Portuguese, no code. Automation canvas with agents, everyday prompts, data prompts, agent instructions, flow anatomy, the four-choice rule, seven governance checks, indicator card, three-phase plan, the five steps.
Lesson-by-lesson checker for the formato-curso v5 format: estimated vs. stated time, jargon sentinels, figure coverage, gate counts, recap ↔ card deduplication, emoji, and double exclamation marks. Python 3 only, no dependencies.
Headless engine test (Playwright): router, timing, figures on mobile, quiz, practice, progress, panels, theme, persistence, and landing page. Ends with ERROS: nenhum or point out what failed.
The suggested order for a company’s first agent project. Each step has a kit file; the prompts (day-to-day and data) are included from day one.
Place the work on the five steps and fill in the nine-field canvas with real names.
Agent in Copilot Studio (instructions + knowledge), workflow in Power Automate (approval, logging, failure alert), integration through the best option.
Three indicators with a target, owner, and automatic source; Copilot explains the variance; the alert notifies the manager.
Seven checks before moving beyond the draft, and a three-phase plan with a definition of done.
Copy the block, replace what’s between < >. The examples (occupational safety technician, purchasing analyst, HR manager, finance coordinator) are fictional.
User → augmented → workflow designer → agent builder → automator. The Person → Agent → Systems mental model and the filter for the first three tasks.
Nine fields on one page, a filled-in example (payment reconciliation), and the agent / flow / screen / indicator breakdown.
Search across sources, long email → decided / pending / who hasn’t responded, meeting → decisions, owners, and deadlines, document and presentation.
Excel, Analyst, and Researcher: “what changed and why,” a testable hypothesis across files, research inside + outside, the three two-minute checks.
Template for role · scope · rules · tone in Copilot Studio, a six-step creation checklist, and weekly maintenance.
Approval, expiring document, and incident called up by the agent—always with logging and a failure alert.
Connector → workflow → screen automation → Computer Use; end-to-end vendor registration; case with answer key.
Seven checks before publishing, where a person is required, an argument for leadership, and a mini-case on excessive sharing.
Indicator card, a request for Copilot to explain a deviation, a minimum dashboard in five steps, a database by area, questions a manager should be able to ask.
Prototype → execution → scale with a clear definition of done, the four labs, department-specific cases, and the company roadmap in nine phases.
The kit is text: it only requires Copilot enabled by your company. The tools require Python 3 and, to test the engine, Node with Playwright.
Microsoft 365 Copilot on your work account. Copilot Studio and Power Automate for templates 04 and 05 (if you don’t have access, the templates become the specification IT will need).
# nothing to install git clone https://github.com/inematds/copilot-agentes-kit
Standard Python 3. No extra packages.
python3 --version # 3.8 or higher
Node 18+ and Playwright with Chromium (or Playwright already installed somewhere else, via PLAYWRIGHT_DIR).
npm i -D playwright npx playwright install chromium
Steps 1 to 4 are for those who will use the kit at work. Steps 5 to 7 are for those who create INEMA v5 courses.
kit/Ten Markdown files. Read 10 first (five minutes) to place your work on the five steps.
git clone https://github.com/inematds/copilot-agentes-kit cd copilot-agentes-kit/kit && ls # 01 … 10
Open 02-prompts-copilot-dia-a-dia.md, copy the meeting block, replace what’s between < > and paste it into Copilot in Teams. A rule that applies to all of them: a subject with a proper name + time period + what you want to see at the end.
Use a reunião <nome e data>. Faça, nesta ordem:
1. Resumo em até 10 linhas. 2. Decisões, sem "talvez".
3. Para cada decisão: responsável (um nome), tarefa e prazo.
4. Pendências sem responsável. 5. Mensagem de acompanhamento cobrando com prazo.
01-canvas-automacao-com-agentes.md: nine fields, twelve minutes, real names. Mark decisions that go through the risk checklist (file 07) with ✱. “I don’t know yet” is fine in the Integration field—the answer comes later in file 06.
CANVAS — <nome do processo>
Objetivo · Entradas · Conhecimento · Decisões (✱ humanas) · Ações
Integração · Validação · Indicadores
Draft agent using the template in file 04 (test with three questions, one outside the scope). Approval workflow using file 05 (log + failure alert). Three metrics using file 08. Before publishing, complete the seven checks in file 07. Present it using the three-phase plan in file 09.
# suggested order
10 → 01 → 04 → 05 → 06 → 08 → 07 → 09
Write each lesson in a file (parts/NN-aula.html) and run the checker when you finish each one. It prints the estimated time, the jargon sentinel with context, the gate counts, and the recap ↔ cards anti-duplication check. Only then concatenate.
python3 tools/check-v5.py parts/01-aula.html # words=1880 quiz=1 practice~10min → estimated 20min | data-time=20 # steps=5 figs=5 … manifest=1 MISSING=0 # DUPLICATION ≥6: … ← if it appears, rewrite the card cat parts/*.html > curso.html
Opens curso.html e landing.html in a headless Chromium browser, dismisses the onboarding, navigates, clicks the quiz and practice, opens the panels, switches the theme, reloads, and tests on mobile. Takes screenshots in the specified folder.
node tools/browser-test.cjs /caminho/do/curso shots # … statcards: 0/9 lessons completed ~203min remaining # … mobile body scrollWidth ok: true # ERRORS: none # Playwright somewhere else? PLAYWRIGHT_DIR=/x/node_modules/playwright node tools/browser-test.cjs …
Cover via the skill capa-inema, index.html with a redirect to the landing page, .nojekyll, push, and GitHub Pages. Catalog via the skill atualiza-portal.
node ~/.claude/skills/capa-inema/assets/gerar-capa.cjs --repo . --title "Título" --cat "Categoria" git push && gh api -X POST repos/inematds/<repo>/pages -f 'source[branch]=main' -f 'source[path]=/'
Both tools ran on the original course: nine lessons, 100% manifest compliance, zero console errors.
Version 1 is what's in the repo today. The next phases depend on real-world use.