Claude Code User Configuration

Personal ~/.claude settings for Claude Code.

Structure

Path	Purpose
CLAUDE.md	Global instructions loaded into every conversation
settings.json	Claude Code settings (model, permissions, env vars, plugins)
rules/	Modular instruction sets referenced from CLAUDE.md
skills/	Custom slash-command skills (issue planning, code review, etc.)
commands/	Lightweight slash commands (commit, discover, codex review)
mcp-servers/	MCP server configuration (docker-compose + env)
agents/	Agent definitions

Rules

• comments.md - When comments are acceptable (why, not what)
• cw-cli.md - CoreWeave CLI reference
• git-spice.md - Stacked PR workflow with gs
• grug-brain.md - Anti-complexity development philosophy
• issue-tracking.md - br CLI patterns for issue management
• python.md - Python conventions (uv, style)
• session-protocol.md - Session startup, progress, and completion
• testing.md - Test real behavior, not coverage numbers

Skills

Custom skills invoked via /skill-name:

• Issue planning: issue-plan, issue-plan-codex, issue-plan-hybrid, issue-plan-user
• Code review: team-branch-review, parallel-branch-review, grug-review, codex-*-review
• Review follow-up: team-branch-fix, team-branch-comment, pr-review-reply, pr-review-import
• Workflow: git-commit, issue-create, remember, repo-explore, discover
• Documentation: diataxis-documentation, humanizer
• CoreWeave: cw-repo, cw-scaffold, cw-explore, cw-dev
• Infrastructure: kubernetes

Key Conventions

• All work tracked via br (beads_rust) - beads are picked up and implemented by atari
• Commits created through /commit skill
• No git push or GitHub writes without explicit approval
• Python uses uv for everything