🏆 Mastery, Ethics & Best Practices
You've learned the tools. Now learn how to use them responsibly, efficiently, and wisely — and how to keep growing as the tools evolve.
In this module
Habits of mastery
The researchers who get the most out of Claude Code share these habits:
✅ Masters do this
- Always verify Claude's outputs before using them
- Maintain good CLAUDE.md files for every project
- Ask Claude to critique its own work
- Use Git so every change is reversible
- Build a personal library of custom skills
- Think in terms of workflows, not one-off questions
- Update their approach as Claude Code improves
❌ Beginners do this
- Accept outputs without checking
- Re-explain the project every session
- Ask vague, one-line questions
- Work without version control
- Repeat the same prompts manually each time
- Treat Claude as a search engine, not a collaborator
- Stick to what they learned in month one
The mastery mindset
"The question is not 'can I trust Claude?' The question is 'have I set up Claude to succeed?' A well-briefed Claude with a good CLAUDE.md, in a well-organised project, with a researcher who reviews outputs critically — that's a powerful combination."
Academic integrity and disclosure
The four questions every researcher should ask
| Question | Guidance |
|---|---|
| Did Claude help write any part of this paper? | Disclose in acknowledgements or per journal policy. Most journals now require disclosure of AI writing assistance. |
| Did Claude run or write any analysis code? | Review all code carefully. You are responsible for its correctness. Share the code so others can verify it. |
| Are all citations accurate? | Never trust Claude's citations without verifying them. Claude can hallucinate paper details. Always check against the actual source. |
| Could this qualify as research misconduct? | Check your institution's AI policy and relevant journal policies. Policies are evolving rapidly — what's acceptable today may change. |
Sample disclosure language
When disclosure is appropriate, here is example language adapted from various journal policies:
"The authors used Claude Code (Anthropic) to assist with [data cleaning / coding analysis scripts / drafting portions of this manuscript]. All outputs were reviewed and verified by the authors, who take full responsibility for the accuracy and integrity of this work."
Data privacy and security
- PII (personally identifiable information): real names, national IDs, phone numbers, exact addresses
- Health data: medical records, HIV status, mental health information
- Financial records: individual bank records, tax filings
- Protected data under research ethics approval: data you agreed to keep confidential
Safe alternatives: Anonymise data first, work with aggregated statistics, or use synthetic data with the same structure.
To work safely with sensitive data:
# GOOD: Share only the structure, not the content
> Here is the structure of my dataset (not the actual data):
hh_id (int), village_id (int), treatment (0/1),
income (float), age (int), education_years (int)
Write a data cleaning script assuming this structure.
# GOOD: Anonymise before sharing
> I've replaced all real names with codes (HH001, HH002...).
Now help me analyse this anonymised dataset.AI access and global equity
Claude Code is potentially transformative for researchers in low- and middle-income countries who may not have access to large research labs, expensive software, or many RAs. But access isn't equal — here are some considerations:
Options: (1) Use the free tier of Claude.ai for lighter tasks. (2) Apply for Anthropic's research access programs. (3) Use the cheapest models (Haiku) for batch tasks. (4) Share costs with colleagues by pooling an API key. (5) Some universities now provide institutional Claude access.
Claude Code works on any internet connection — it sends text, not large files. Even a slow mobile data connection (1–2 Mbps) is sufficient. Once Claude has read your files, subsequent interactions are fast. Use /compact to keep conversations short and reduce data usage.
Claude was trained primarily on English-language, Western-context text. Be aware that: (1) Its knowledge of local context (specific country institutions, local languages, informal economies) may be limited. (2) It may default to Western methodological conventions. (3) For local-language data, accuracy may be lower. Always apply your domain expertise to supplement Claude's knowledge.
Keeping up as Claude Code evolves
Claude Code is developing rapidly. Here's how to stay current:
- Official docs: docs.anthropic.com/en/docs/claude-code — updated frequently
- Release notes: Follow Anthropic on their blog for new features
- Community: The Awesome Econ AI list for economist-specific resources
- This course: Check the GitHub repo for updates — we'll incorporate new features as they're released
What's next?
Your next steps
- This week: Set up Claude Code on your current research project, write your CLAUDE.md
- This month: Use it daily for at least one task. Build your first custom skill.
- Ongoing: Contribute back — share your CLAUDE.md templates, skills, and workflows with the community via this course's GitHub
Contribute to this course
This course is open source. If you've built useful CLAUDE.md templates, custom skills, or workflows — please contribute them:
# Fork the repo, add your materials, submit a pull request
git clone https://github.com/toyibaremu/claude-learn
# Add your CLAUDE.md template to: examples/claude-md-templates/
# Add your custom skills to: examples/skills/Final Capstone: Your AI-powered research project
Apply everything from this course to a complete mini research project:
- Set up a new project folder with proper structure (Module 9)
- Write a thorough CLAUDE.md (Module 3)
- Use Claude to clean and analyse a dataset (Module 4)
- Generate LaTeX tables and draft a results paragraph (Module 5)
- Create at least one custom skill for a task you'll repeat (Module 6)
- Add a Git repository and commit your work (Module 11)
- Write a one-page reflection: what worked, what surprised you, what would you do differently?
Share your capstone project in the GitHub Discussions — we'd love to see what you built.