What Happened
Rebecca Bultsma published a guide arguing that most people using Claude Code are skipping a critical setup step: building personal context files. Her recommendation is to create five markdown files before touching any AI coding tool:
- Constitution-Context.md - Your core values and non-negotiables
- Work-Context.md - Current role, active projects, what's working and what's broken
- Goals-Context.md - Specific, measurable objectives with deadlines
- PersonalLife-Context.md - Daily rhythms, energy patterns, constraints
- Clients-Context.md - Information about stakeholders you serve (optional)
The suggested time investment is 2-3 hours upfront, with a claimed payoff of 5+ hours saved per week. Bultsma recommends using voice interviews with Claude or ChatGPT to generate the files rather than typing them from scratch. The files live in a desktop folder and get uploaded to any AI tool when starting conversations.
The core principle: "Build once. Use everywhere."
Why It Matters
If you've used Claude Code or any AI coding assistant for more than a week, you've felt this pain. You start a new conversation, and the AI has zero memory of your preferences, your project structure, or your coding style. You spend the first 10 minutes re-explaining context that should be obvious.
Claude Code already supports CLAUDE.md files for project-level instructions, and this approach extends that thinking to your personal workflow. The difference between a generic AI response and a useful one almost always comes down to context quality. Five markdown files won't solve every problem, but they address the most common one: the AI doesn't know who you are or what you care about.
The "build once, use everywhere" framing is the real insight. These files aren't locked to Claude Code. They work with ChatGPT Projects, Cursor, or any tool that accepts context documents. You're building a portable knowledge base about yourself.
Our Take
This is solid advice wrapped in slightly too much enthusiasm. The five-file structure is a good starting framework, but most practitioners will find they need something more targeted. A "Constitution" file sounds nice in theory, but what actually saves time is specific technical preferences: "I use TypeScript strict mode," "I prefer functional components," "Never suggest Redux."
The real value here is the habit, not the specific file structure. If you're not systematically feeding context to your AI tools, you're getting generic output and blaming the model for it. That said, Claude Code's CLAUDE.md already handles most of what's described here for coding contexts. Where personal context files add genuine value is for non-coding work: writing, planning, strategy.
If you're spending more than 30 seconds per conversation re-explaining who you are and what you want, build the files. Start with two - your work context and your preferences - and expand from there. Don't overthink the five-file taxonomy. Just write down what you keep repeating to AI tools, save it as markdown, and attach it every time.