skills/claude-md-authoring/SKILL.md
Use when writing or improving CLAUDE.md files — project context that Claude Code reads every session, global vs project rules, what to include for maximum AI effectiveness, and memory-aware documentation patterns.
npx skillsauth add kienbui1995/magic-powers claude-md-authoringInstall this skill globally with one command. Works with Claude Code, Cursor, and Windsurf.
3 of 9 scanners reported clean
Some scanners were skipped, did not run, or reported a non-clean status. Review each row below.
CLAUDE.md is the most powerful lever for making Claude Code effective on your project. It's read at the start of every session — think of it as the briefing document Claude reads before doing any work. A well-written CLAUDE.md means Claude follows your conventions without being asked, and a missing or stale one means Claude reinvents conventions every session.
Claude Code reads files in priority order:
1. ~/.claude/CLAUDE.md (global — applies to all projects)
2. /project/CLAUDE.md (project root — main project context)
3. /project/subdir/CLAUDE.md (subdirectory — specific area rules)
Higher priority overrides lower. Use global for personal preferences, project for team conventions.
High-value content:
# Project: [Name]
## What This Is
[2-3 sentences: what the project does, who uses it, tech stack]
## Architecture
[Key architectural decisions Claude should know: monorepo, microservices, etc.]
- API: FastAPI on port 8000, PostgreSQL via SQLAlchemy
- Frontend: Next.js 14 App Router, Tailwind CSS, shadcn/ui
- Auth: Clerk (not custom auth — never implement auth yourself)
## Conventions
- Python: use `ruff` for formatting, type hints required on all functions
- Commits: conventional commits (feat:, fix:, chore:)
- Never use print() — always use the logger: `from app.core.logger import logger`
## Do Not
- Never modify migrations directly — always use `alembic revision --autogenerate`
- Never commit .env files
- Never use synchronous DB calls in async functions
## Current Focus
[Optional: what's being built right now, context for current sprint]
Low-value content (don't include):
Personal preferences that apply across all projects:
## My Preferences
- I'm [role] working primarily in [languages]
- Communication style: concise, direct, minimal explanation unless asked
- When in doubt, ask before making large changes
- Always run tests before declaring done
## Tools I Use
- Package manager: uv (Python), pnpm (Node)
- Formatter: ruff (Python), prettier (JS/TS)
- Testing: pytest (Python), vitest (JS/TS)
## Workflow
- Read existing code patterns before suggesting new approaches
- Prefer editing existing files over creating new ones
- Commit frequently with conventional commit messages
Update CLAUDE.md as the project evolves:
~/.claude/CLAUDE.md — personal preferences across all projects/project/CLAUDE.md — shared team conventions/subdir/CLAUDE.md — area-specific rules (e.g., /frontend/CLAUDE.md)claude-project-settings for complete Claude Code configurationclaude-hooks (document hook behaviors in CLAUDE.md so team understands them)@architect helps write the architecture section of CLAUDE.mdcontent-media
Use when designing for XR (AR/VR/MR), choosing interaction modes, or adapting 2D UI patterns for spatial computing
testing
Use when creating new skills, editing existing skills, or verifying skills work before deployment
development
Use when you have a spec or requirements for a multi-step task, before touching code
development
Use when executing a structured workflow — select and run a feature, bugfix, refactor, research, or incident template with correct agent and model assignments per phase.