N2 — Advanced Operator ▾
Advanced Operator
Playbook
Claude Code as a professional working system. 10 modules to build, debug, refactor, automate, document, and collaborate without losing control of the project.
What you'll learn
Diagram — How an advanced operator works
Goal
What you want to build or improve
Project architecture
Structure, folders, conventions
Context + advanced CLAUDE.md
Rules, stack, constraints, flow
Git + debugging + refactors
Small, reversible, auditable changes
Automations + MCPs
Tools connected with judgment
Deploy + documentation
Ship, version, transfer
Scalable system
Another operator can continue without friction
Complete Level 1 to unlock
The Advanced Operator Playbook unlocks when you complete Operator Foundations. Level 1 gives you the foundation this level assumes you already have.
Go to Level 1 — Operator Foundations →Project architecture
Goal
Structure a project so Claude can understand it and modify it without breaking it.
Resultado
Clean repo, clear README, coherent folders, and initial rules documented.
mapa-arquitectura.md ADVANCED INSIGHT
Organize the structure as if Claude were going to read the repo cold. If the folder is confusing to you, it will produce misaligned code from the very first message.
Advanced CLAUDE.md
Goal
Create deep operational rules for a complex project.
Resultado
Claude understands style, limits, stack, commands, security, and the way you work.
CLAUDE.md v2 ADVANCED INSIGHT
A CLAUDE.md v2 isn't longer than v1 — it's denser. One specific rule ("don't add dependencies without justification") is worth more than three generic ones ("be careful").
Context control
Goal
Learn what to give Claude, what not to give it, and when to summarize.
Resultado
Fewer wasted tokens, more precise answers, more efficient sessions.
brief-proyecto.md ADVANCED INSIGHT
When the context grows too large, don't delete it: summarize it. A 20-line brief-sesión.md is more useful than 5,000 tokens of disordered history.
Professional Git workflow
Goal
Work with commits, branches, and diffs without chaos or lost work.
Resultado
Small, reversible, reviewable changes. A history that tells a story.
git-workflow.md documentado ADVANCED INSIGHT
Make small, frequent commits. A commit that describes "add footer" is useful. A commit with 12 changed files and the message "improvements" is a time bomb.
Assisted debugging
Goal
Use Claude to find bugs without blindly relying on its suggestions.
Resultado
Isolated bugs with clear hypotheses and small, reversible fixes.
bug-report + solución aplicada ADVANCED INSIGHT
Before asking Claude to "fix this," ask it to explain what it thinks is happening. If it can't explain it, the proposed fix may be plausible but wrong.
Safe refactors
Goal
Improve code without breaking existing functionality or introducing new debt.
Resultado
Phased refactors with the build passing after each change.
before/after documentado con diff ADVANCED INSIGHT
Refactor one thing at a time. Verify the build. Then the next thing. Confidence in a refactor comes from green builds, not from the feeling that it "should work."
Automations and MCPs
Goal
Understand when to connect external tools and when it's over-engineering.
Resultado
A map of useful automations with a justification for each one.
diagrama-automatizacion.md ADVANCED INSIGHT
Connecting an MCP is easy. Knowing what it runs on your behalf and when not to use it is not. Don't add tools when you don't understand what they do in the background.
Testing and validation
Goal
Don't just trust that it "seems to work" before shipping.
Resultado
A manual testing checklist run before every deploy.
qa-checklist.md ADVANCED INSIGHT
A QA checklist isn't bureaucracy: it's externalized memory. Without it, there's always something you forgot to check on deploy day.
Deploy and maintenance
Goal
Ship without breaking production and keep the improvement cycle active.
Resultado
A deploy → review → fix → version flow, documented and executable.
sop-deploy.md ADVANCED INSIGHT
Document the first deploy while you run it. The second time takes half as long. The tenth, minutes. Living documentation is the asset that pays off the most.
Operational documentation
Goal
Turn your own work into a system reusable by other people or agents.
Resultado
Processes documented in enough detail to delegate.
sop-proceso.md ADVANCED INSIGHT
The delegable-SOP test: can another person (or Claude) run it without asking you anything? If the answer is "no," the SOP is incomplete.
Four complete scenarios with Claude Code as a production system.
Build a complete landing page
From brief to deploy. Hero, copy, structure, components, basic SEO, and push.
Audit an existing system
Read code, document what it does, identify risks, and propose an improvement plan.
Turn a manual process into a system
Map the steps, identify what can be automated, design the flow, and document a SOP.
Prepare a repo for collaboration
CLAUDE.md, README, .gitignore, folder structure, and contribution rules.
- × Asking for too much at once — Claude loses context and produces code that breaks other parts.
- × Not isolating changes — a refactor that touches 5 files at once is impossible to review.
- × Not using commits — if something breaks, you can't go back to any earlier stable point.
- × Not reading diffs — approving changes you don't understand guarantees silent bugs.
- × Pushing secrets — an API key on GitHub can create debt within hours.
- × Using AI without validating — Claude's output can be plausible but wrong.
- × Letting Claude redo everything — ask for surgical changes, not massive rewrites.
Act as a senior software architect. Analyze the structure of this project: [paste the file tree here or describe the folders] Evaluate: 1. Does the folder organization make sense for this type of project? 2. Which files are redundant, missing, or in the wrong place? 3. Which naming conventions could be improved? 4. How would you organize this so Claude can navigate it without ambiguity? Proposal: suggest an alternative structure if the current one has problems. Be specific. Don't generalize.
Act as an instructions architect for Claude Code. This is the project context: [describe the project, its stack and its goals] Create an advanced CLAUDE.md that includes: 1. Project description and purpose in a single line 2. Technical stack with exact versions 3. Critical commands (build, test, lint, deploy) 4. Architecture rules: what to do and what never to do 5. Naming conventions and file organization 6. Expected workflow: how Claude should propose changes 7. Security rules: which files or variables never to touch 8. Success signals: how to know the work is done well 60 lines max. Dense. No filler.
Act as a refactoring engineer. I want to improve this code fragment: [paste the code] Before editing anything, I need a refactor plan that includes: 1. What problems the current code has (specific, not vague) 2. What you would change and why 3. Risks of each proposed change 4. In what order to make the changes so the code is never left broken 5. How to verify that each change hasn't introduced regressions Don't start writing code until I approve the plan.
Verify that you've completed all 10 modules before marking the level:
- ✓ Architecture map of a real project documented
- ✓ Advanced CLAUDE.md with the 8 operational sections
- ✓ Project brief of 20 lines max created
- ✓ Git flow with branch, semantic commit, and merge documented
- ✓ Assisted debugging carried out on a real bug
- ✓ Limited refactor with before/after documented
- ✓ Automation diagram not yet implemented
- ✓ QA checklist with at least 10 validation items
- ✓ Deploy SOP documented and run once
- ✓ Delegable process SOP of 1 page max
LEVEL 2 COMPLETED
Level 3 unlocked: ChatGPT → Claude Migration System