You build a hierarchy of CLAUDE.md
files for your codebase. Claude Code uses these files to understand your architecture, terminology, and business rules. You work with Claude Code to build them; no script auto-generates them. Expect to invest 2-4 hours initially, plus refinement over 1-2 weeks.
Best for: Production codebases (50-5000+ files) with limited documentation, teams using AI assistants.
- Not auto-generated. You and Claude Code build these files together through interview and validation. No script generates documentation from your code.
- Not automatic. Plan to spend 2-4 hours on initial build, plus ongoing maintenance.
- Not for new projects. Use this on brownfield codebases where AI assistants lack context.
- Root CLAUDE.md - Navigation hub with decision trees ("What are you trying to do?"), critical patterns, gotchas
- Reference docs (docs/)
- GLOSSARY.md - UI ↔ code ↔ database terminology mappings
- ARCHITECTURE.md - Tech stack, data patterns, design decisions
- BUSINESS_CONTEXT.md - Workflows, entities, business rules
- Subdirectory CLAUDE.md files - Scoped context (app/controllers/CLAUDE.md, app/views/CLAUDE.md)
- All in git - Maintained alongside your code
Best for: First-time users, single project
curl -sSL https://raw.githubusercontent.com/RossH3/context-tree/main/install.sh | bash
Downloads four guide files to project root:
- CONTEXT_TREE_BUILDER.md
- CONTEXT_TREE_PRINCIPLES.md
- CONTEXT_TREE_DISCOVERY.md
- CONTEXT_TREE_QUICK_START.md
Usage:
claude
# Tell Claude: "Help me build a context tree using CONTEXT_TREE_BUILDER.md"
Best for: Multiple projects, systematic quality enforcement
cd ~/your-project
curl -sSL https://raw.githubusercontent.com/RossH3/context-tree/main/install-advanced.sh | bash
Installs to .claude/
:
- Context Tree Maintenance skill (automatic discovery)
- Slash commands (
/audit-context
,/discover-codebase
,/capture-insight
) - Session hooks
Usage:
cd ~/your-project
claude
/discover-codebase
# Or: "Use the Context Tree Maintenance skill"
Start with Basic. Upgrade anytime:
cd ~/your-project
curl -sSL https://raw.githubusercontent.com/RossH3/context-tree/main/install-advanced.sh | bash
Your context tree files remain unchanged.
Feature | Basic | Advanced |
---|---|---|
Guides | ✅ | ✅ |
Builder workflow | Manual | Skill-guided |
Reusable across projects | Copy files | Simple curl install |
Slash commands | ❌ | ✅ |
Validation checklists | Manual | Built-in |
Quality enforcement | Self-discipline | Guardrails |
Signal over Noise:
- Document what code structure cannot reveal
- Capture terminology mismatches (UI vs code vs database)
- Focus on gotchas, business rules, architectural decisions
- Remove generic framework explanations
Verify Against Code:
- Check architectural claims against actual implementation
- Never document from other documentation
- Test effectiveness with real tasks
Single Source of Truth:
- Each architectural fact has one authoritative location
- Other documents reference, never duplicate
See CONTEXT_TREE_PRINCIPLES.md for complete principles.
- CONTEXT_TREE_QUICK_START.md - Getting started
- CONTEXT_TREE_BUILDER.md - Interview guide and templates
- CONTEXT_TREE_PRINCIPLES.md - Design principles
- CONTEXT_TREE_DISCOVERY.md - Automated analysis commands
- skills/README.md - How skills work (Advanced)
How long does this take? 2-4 hours for initial context tree, then refinement over 1-2 weeks of real usage.
Do I document everything? No. Focus on signal-to-noise ratio. Document only what code structure cannot reveal.
Can I do this in multiple sessions? Yes. Commit after each session. Next session: "Continue building the context tree."
Basic or Advanced?
- Basic: First context tree, learning the methodology
- Advanced: Multiple projects, systematic quality enforcement
- Issues: GitHub Issues
- Discussions: GitHub Discussions
MIT License - See LICENSE file
- Concept: Ross Hanahan
- Inspired by: Jesse Vincent's Superpowers skills system
- Persuasion principles: Robert Cialdini's research on commitment and authority
Build context trees. Make Claude Code effective. Ship faster.