✅ Capstone handoff checklist
New here? Capstone is the final project in a course, the thing that brings together everything you learned in a single deliverable. Here, the capstone is your executive assistant — the folder structure, context files, skills, MCPs, and routines you built throughout Track 7.
The deliverable isn’t "I finished watching the modules." It’s a set of verifiable criteria — things you can point to and show, not describe vaguely. Each item below corresponds to a previous module in this track.
📋 Capstone evaluation criteria
- 1.Complete folder structure — the assistant’s “home” (module 7.1) exists and is organized; it isn’t a loose folder of random files.
- 2.A lean CLAUDE.md that points the way — it doesn't try to explain everything; it points you to where to find each thing (context files, skills, routines).
- 3.Context files filled with real, specific information — your actual business, your team, your quirks (module 7.2), never a generic placeholder.
- 4.At least one skill working end to end — you asked, it ran, the result came out right (module 7.3).
- 5.At least one connected MCP — the assistant has an extra "hand" beyond the plain terminal (module 7.3, and all of Track 2).
- 6.Some scheduled routine running — it doesn't need to be complex, but it does need to run on its own (module 7.3, building on Track 6).
- 7.Backup configured — if the disk dies tomorrow, the assistant survives (module 7.5).
💡 Practical Tip
Print (or copy) these 7 items into a simple list and check them off as you review your project. You don’t need a score; you need an honest “yes” or “no” on each line — that’s what topic 2 automates.
🔍 Ask Claude Code to evaluate itself
The most honest way to check the 7 criteria in topic 1 is to ask the assistant itself to read your folder and compare it with the list. It has no ego or desire to "pass" — it just looks at what’s there and points out what’s missing.
Você é um avaliador externo do meu assistente executivo. Leia a
estrutura de pastas deste projeto (comece pelo CLAUDE.md) e responda,
item por item, com SIM ou NÃO + evidência concreta (arquivo/caminho):
1. Existe estrutura de pastas organizada (não solta)?
2. O CLAUDE.md é enxuto e aponta pra onde encontrar cada coisa?
3. Os arquivos de contexto têm informação real e específica de
<seu-negócio>, não texto genérico de exemplo?
4. Existe pelo menos uma skill funcionando de ponta a ponta?
5. Existe pelo menos um MCP conectado e configurado?
6. Existe alguma rotina agendada configurada?
7. Existe backup/versionamento configurado (git ou equivalente)?
No fim, liste os itens que faltam e sugira, para cada um, o menor
próximo passo possível para resolver.Objective: get an objective, item-by-item report with evidence, not a vague opinion. How to verify: check that each "YES" came with a real file path you can open — if the answer is generic and doesn’t point to a file, ask it to review the folder more carefully.
🔍 Inside
- Why it works: The agent has read access to your project — it can open every file mentioned and check whether the content is real or a placeholder, which a static checklist can’t do.
- Repeat later: this self-assessment isn’t just for the deliverable — run it again in 30 days to see whether the assistant has grown (this connects directly to module 7.4, the pace of growth).
🆚 The real difference from a regular chatbot
New here? One chatbot typical (ChatGPT in a new tab, for example) is a blank slate every time you open it. What you built in this track isn't that — it's an assistant with memory and hands. The difference is concrete, across three dimensions.
Caption: the three stacked axes on the left show a typical chatbot starting from scratch in each area; on the right, the assistant you built carries context, performs real actions, and keeps memory between sessions — the green glow marks the real difference.
Project context
The assistant knows who you are, your business, and your team—because it read the context files you filled out in module 7.2. A new chatbot without those files starts from scratch every time.
Real actions
The assistant DOES things—sends messages, runs tasks, edits files—using connected skills and MCPs (module 7.3). It doesn’t just respond with text: the result leaves the screen and enters the real world.
Knowledge that persists
What was decided yesterday still applies today—without you having to explain it again. That’s the “home” (7.1) + the backup (7.5) working together: the assistant’s state survives when you close the session.
⚠️ Common mistakes when handing it off
After watching many people complete this capstone, two mistakes keep coming up. One is about what you deliver; the other is about when you consider the work finished.
✓ Delivery with real evidence
- ✓Screenshot of the actual folder structure, with file names visible
- ✓Excerpt from a context file filled in with your specific information
- ✓Log or short video of the skill running from start to finish
✗ Vague deliverable, no concrete examples
- ✗"I configured everything correctly" without showing anything
- ✗Context file still contains the template’s sample text
- ✗Claims the skill "works" but has never actually run it
✓ Understand that it’s a living project
- ✓Delivers today, but already knows they’ll come back in 2 weeks to add something
- ✓Treat the self-assessment in topic 2 as a routine, not a one-time event
✗ Thinking it's finished
- ✗Delivers the capstone and never opens the folder again
- ✗Treat "passed the checklist" as the finish line, not the starting point
🎓 What you built in 7 days
Before looking ahead, it’s worth pausing to see the whole journey. Seven tracks, each adding a new capability, leading to the assistant you have running today.
Installed Claude Code, created the project folder, wrote the first CLAUDE.md, and learned the security rules.
Understood the real execution cycle (Write-Act-Test), where it fails, and built the first real automation.
Gave the agent “hands”: installed and authenticated MCP servers and learned the cheat sheet pattern.
Learned to teach the agent a repeatable process, with progressive loading and the skill itself as the trigger.
Moved the assistant off the laptop: pipeline, secrets, retries, and observability in production.
Gave the agent the ability to see its own work—screenshot loop, visual reference, published landing page.
Taught the assistant to act without you asking: scheduling, loops, routines that learn, notifications.
You brought all of this together in a single executive assistant: home, life, hands, growth, backup, and now the final deliverable.
🎯 Main Concept
No single track turns an assistant into an executive assistant. It’s the sum — environment + framework + tools + processes + production + interface + time — that turns a chatbot into a digital person who works for you.
🛤️ Paths to follow after the course
Finishing the course doesn’t mean the project is finished—it’s when the assistant truly comes into existence and can grow on its own, at your pace. Four concrete paths to take from here.
Add more skills, little by little
Don’t try to create 10 skills in your first week. Follow the growth pace from Module 7.4: create a new skill whenever you notice a repetitive process for the third time.
Explore more MCPs
Each new MCP is an extra “hand.” Go back to Track 2 whenever you need to connect a new tool to the assistant.
Deepen the automations in Track 6
Scheduled routines and loops are the difference between an assistant that waits for you to call and one that has already worked by the time you wake up.
Stay close to the community
O INEMA.CLUB keeps publishing new courses and learning paths — it’s where you can keep learning after this capstone.
⚠️ Attention
Don’t try to solve everything at once. The assistant you built over these 7 days is already useful today—the next step is to use it for real, not pile on more features before running what you already have.
Quick check (optional): what actually sets your executive assistant apart from a typical chatbot?