dotclaude

My Claude Code configuration — behavior rules, custom skills, agent definitions, MCP servers, and a status line that shows what Claude is actually doing.

What this changes

Out of the box, Claude Code is capable but generic. This configuration makes it opinionated:

• It doesn't act without understanding. Before touching code, it checks git history, reads the existing pattern, and understands the why. When asked to diagnose something, it stops after analysis and waits — it doesn't immediately start making changes.
• It thinks about blast radius. Destructive operations get a pause and a confirmation. For Kubernetes, every mutating command must name its context explicitly. No ambient context drift.
• It doesn't pile on changes. One file, one change, test it, move on. If something fails three times the same way, it stops instead of trying a fourth approach.
• It argues back on constraints. When told "we can't change X," it distinguishes between impossible and just-needs-approval, and presents options rather than accepting the constraint as final.

The full ruleset lives in CLAUDE.md.

Status line

A custom status line that mirrors a Powerlevel10k prompt: current directory, AWS/EKS session, virtualenv, git branch with dirty/ahead/behind indicators, model name, and context window usage colored green → yellow → red as it fills.

~/Data/Codebase/my-project | λ dev | main* | claude-sonnet-4-6 | 34%

Reads from ~/.aws_sessions for the EKS indicator — customize the case statement in statusline-command.sh to match your own profile names.

Skills

In this repo

Skill	What it does
code-review-copilot	Multi-model code review via GitHub Copilot CLI — sends diffs to GPT, Claude, and Gemini, synthesizes findings
promptheus	Refines prompts using the Promptheus MCP server
review-board	Document review board powered by multiple LLM CLIs
review-board-copilot	Same as above but scoped to Copilot
writing-assistant	Context-aware text rephrasing and improvement

Marketplace (download separately)

General-purpose skills available from the marketplace:

• algorithmic-art — generative algorithmic art
• brand-guidelines — extract and apply brand guidelines
• canvas-design — canvas-based design tools
• frontend-design — production-grade frontend UI generation
• internal-comms — draft internal communications
• mcp-builder — scaffold MCP servers
• rekhta-qafia — Rekhta qafia/rhyming dictionary search
• skill-creator — guide for creating new skills
• slack-gif-creator — Slack GIF creation
• template-skill — starter template for new skills
• theme-factory — UI theme generation
• ac-triage-dependabot — safely triage open Dependabot PRs
• ac-* — private work-specific cluster health, log tracing, and operational skills
• web-artifacts-builder — web artifact generation
• webapp-testing — Playwright-based web app testing

Plugins

Install with claude plugin install <name>@<registry>:

Plugin	Registry	What it does
superpowers	claude-plugins-official	Meta-skill framework: brainstorming, TDD, systematic debugging, plan writing, code review workflows
codex	openai-codex	OpenAI Codex CLI integration
claude-hud	claude-hud	Status line / HUD display
claude-md-management	claude-plugins-official	CLAUDE.md creation and improvement tools
claude-code-setup	claude-plugins-official	Setup and automation recommendations
gopls-lsp	claude-plugins-official	Go language server (LSP) integration
firebase	claude-plugins-official	Firebase project integration

MCP Servers

Configured in config/mcp.json:

Server	What it does
aws-knowledge	AWS documentation search
kubernetes	K8s cluster management
memory	Persistent memory across sessions
git	Git operations
promptheus	Prompt refinement
aws-api	AWS CLI operations
cloudflare	Cloudflare documentation

Agents

Custom agent definitions in agents/ — specialized subagents for Go, Python, Kubernetes, debugging, cloud architecture, and more. Dropped into ~/.claude/agents/ so Claude can dispatch them automatically.

Setup

# Clone wherever you keep dotfiles
git clone <repo> ~/dotfiles/claude

# Symlink or copy to ~/.claude
cp CLAUDE.md ~/.claude/CLAUDE.md
cp settings.json ~/.claude/settings.json
cp config/mcp.json ~/.claude/config/mcp.json
cp -r skills/* ~/.claude/skills/
cp -r agents/* ~/.claude/agents/
cp claude-powerline*.json ~/.claude/
cp statusline-command.sh ~/.claude/

Then fill in your values:

• config/mcp.json — replace <YOUR_API_KEY> and <YOUR_AWS_PROFILE>
• CLAUDE.md — replace any environment-specific references with your own clusters, namespaces, and tooling
• statusline-command.sh — update the case block with your AWS profile names
• settings.json — adjust permissions to match the tools you actually use

Structure

.
├── CLAUDE.md                       # Behavior rules and conventions
├── settings.json                   # Permissions, hooks, plugins
├── config/
│   └── mcp.json                    # MCP server configurations
├── skills/                         # Custom skills (see above)
├── agents/                         # Custom agent definitions
├── claude-powerline.json           # Powerline theme config
├── claude-powerline-custom.json    # Custom powerline theme
└── statusline-command.sh           # Status line script

Notes

• Only custom skills are in this repo. Marketplace skills and plugin-provided skills install separately.
• commands/ is gitignored — command collections like CCPlugins install there and are project-specific.
• Never commit real API keys. The mcp.json here uses placeholders; keep your real values in ~/.claude/config/mcp.json.