Resource library

Vibe coding means building software by describing what you want in natural language, letting an AI coding assistant draft the code, testing the result, and refining it through conversation. You are not handing over clinical judgment. You are using AI as a fast junior developer while you remain the clinical architect.

Interventional pulmonology has problems that are obvious to physicians but too narrow for many commercial roadmaps: registry capture from procedure notes, board-style teaching tools, procedure simulators, anatomy modules, literature assistants, and workflows that understand EBUS-TBNA, cryobiopsy, navigational bronchoscopy, pleural procedures, valves, rigid bronchoscopy, and stent surveillance.

1. 1Describe the clinical problemState the user, setting, input, output, and success criteria. "Build an EBUS station quiz for fellows" is better than "make an app."
2. 2Plan before codingUse a chat assistant to turn the idea into requirements, a feature list, a data model, and first-version scope.
3. 3Generate small piecesAsk for one component, one script, or one route at a time. Large prompts are fine for planning; small prompts are better for reliable implementation.
4. 4Run and testUse the IDE terminal. Copy exact error messages back to the assistant. Ask for reproducible checks, not just another attempt.
5. 5Refine with domain expertiseCorrect medical language, workflows, and edge cases. Ask AI to critique the product, not just write code.
6. 6Commit, review, and shareUse GitHub as project memory. Save checkpoints, create issues, get review, and deploy only when the risk level is appropriate.

Do not start by learning every tool. Pick the stack that matches the current goal.

Three tiers of AI coding tools

An IDE is an integrated development environment: a workspace where you view files, edit code, run commands, manage Git, and debug errors. Think of it as the software workbench that keeps the editor, terminal, source control, and AI assistant in one place.

The five IDE areas beginners should learn

1. 1File explorerShows project folders and files. Ask AI to explain the folder structure.
2. 2EditorWhere files open. Comments can become prompts, such as "// Build a quiz component that loads questions from quizData.json".
3. 3TerminalWhere commands run, such as npm run dev, python app.py, or git status.
4. 4Source control panelShows what changed. Review this before every commit.
5. 5AI chat / agent panelWhere you describe tasks, review errors, and request scoped changes.

Git tracks changes. GitHub stores those changes online, supports collaboration, and can host simple websites. Beginners can start with GitHub Desktop before relying on terminal commands.

The beginner GitHub workflow

1. 1Create a private repositoryUse a short name, add a README, and add a license only when you are ready to share.
2. 2Clone it locallyUse GitHub Desktop or the terminal. Open the folder in VS Code, Cursor, or Windsurf.
3. 3Build one featureAdd one page, one parser, one component, or one test.
4. 4Review changed filesRead the diff. Ask the AI to explain anything you do not understand.
5. 5Commit with a useful messageExample: Add EBUS station quiz with JSON question data.
6. 6Push to GitHubNow the checkpoint exists online and can be reviewed, deployed, or continued elsewhere.

# See what changed
git status

# Save all changed files into the next checkpoint
git add .

# Create the checkpoint
git commit -m "Add EBUS station quiz"

# Upload to GitHub
git push

# Create a safe branch for a new feature
git checkout -b add-tnm-module

Agents are different from chatbots. A chatbot answers. An agent can inspect files, edit files, run commands, and sometimes create pull requests. That power is why agents are useful and why guardrails matter.

The safest agent workflow

1. 1Start from a clean Git stateRun git status. Commit or stash current work.
2. 2Create a branchExample: git checkout -b add-tnm-staging-module.
3. 3Ask for a plan before editsTell the agent: Inspect the repo and propose a plan. Do not edit yet.
4. 4Approve a small scopePrefer one feature. Avoid rebuilding the entire app in one pass.
5. 5Run tests or buildRequire npm run build, pytest, or whatever applies.
6. 6Review the diff and summaryAsk for changed files, risks, assumptions, and manual test steps.

You are working in my GitHub repo for an educational interventional pulmonology app.

Goal: Add a TNM staging teaching module for pulmonary fellows.

Rules:
- Do not use or create any patient-identifiable data.
- Do not make changes until you inspect the repo and propose a short plan.
- Keep the first version small: reference cards, a simple staging builder, and 3 cases.
- Use existing styling and navigation patterns.
- Add or update tests if the project already has tests.
- Run the build command and report whether it passes.
- At the end, summarize changed files, assumptions, risks, and manual test steps.

Start by inspecting README.md, package.json, and the src folder. Then propose a plan.

The goal is not perfect wording. The goal is enough context for the assistant to make good decisions.

I am a pulmonary/critical care physician building a beginner project with AI coding tools.

Clinical problem:
[Describe the problem]

Users:
[Medical students / fellows / research coordinator / attending / quality team]

Setting:
[Education only / research data / clinical workflow prototype]

Inputs:
[CSV, text notes, images, REDCap export, synthetic cases]

Outputs:
[Dashboard, quiz, parser output, report, PDF, table]

Constraints:
- Use de-identified or synthetic data only.
- Beginner-friendly stack.
- First version should be small and buildable in one weekend.

Please create:
1. A one-paragraph project summary.
2. MVP feature list.
3. Future feature backlog.
4. Recommended tech stack.
5. Data model.
6. Step-by-step build plan.
7. Risks and safety review checklist.

I am getting this error while running my app:

[Paste exact error]

Context:
- Tool/editor: [VS Code / Cursor / Claude Code / Codex]
- Command I ran: [npm run dev / python app.py / pytest]
- File I think is involved: [filename]
- What I expected: [expected behavior]
- What happened: [actual behavior]

Please:
1. Explain the error in beginner-friendly terms.
2. Identify the likely cause.
3. Give the smallest safe fix.
4. Tell me exactly which file to edit.
5. Suggest a test to confirm it is fixed.

This is a visual/spatial problem and text iteration is not converging.

Please build a temporary workbench page that lets me manually adjust the values.
Requirements:
- Show the target image or object.
- Let me drag, resize, and nudge the hotspots, labels, or points.
- Show live coordinates and dimensions.
- Include a Copy JSON button.
- Store the temporary page in a clearly named route: /debug-hotspot-editor.
- Do not integrate into production yet.

After I paste the final JSON, help me wire it into the production component and remove the debug page.

Build a React + Vite teaching app for EBUS nodal stations.

Audience: pulmonary fellows.
Purpose: rapid visual review before simulation lab.

Features:
1. Landing page with learning objectives.
2. Interactive station map with clickable stations 2R, 2L, 4R, 4L, 7, 10R/L, 11R/L.
3. Flashcard mode.
4. 15-question quiz with immediate feedback.
5. Case mode with 3 realistic staging cases.
6. Content stored in JSON files for easy editing.
7. Mobile-responsive layout.
8. Accessibility: keyboard navigation and alt text.
9. Disclaimer: educational use only.

Use synthetic cases only. Create the folder structure and starter components.

Healthcare software has a different risk profile than a hobby app. Treat safety as a design requirement, not a final checklist.

Pre-share checklist

• No patient names, dates, MRNs, accession numbers, locations, faces, voice, or identifiers.
• No secrets in code, screenshots, logs, notebooks, or GitHub history.
• README explains purpose, limitations, install steps, and data policy.
• Clinical content reviewed by a domain expert.
• Educational disclaimer or research disclaimer included.
• Build or test command passes.
• Repo visibility set intentionally: private for early work, public only when safe.
• Institutional IT, IRB, compliance, or CMIO involved when appropriate.

.env
.env.*
*.key
*.pem
*.csv
*.xlsx
*.dcm
*.nii
*.nii.gz
patient_data/
exports/
logs/
node_modules/
__pycache__/
.DS_Store