claudefiles

My opinionated resources for working with Claude Code.

Features

• 🤖 Agents for research, code review, security analysis, etc.
• ⚡ Slash commands for task completion, prompt refinement, and more
• 🔒 Reasonable permissions for balancing agent autonomy with security
• 🔔 Notification hooks to alert you when Claude needs attention
• 🛡️ Safe wrapper scripts for granting Claude access to dangerous commands
• 🌳 Worktree management scripts for building ergonomic workflows
• 📊 Status line showing the current model and session information
• 📖 Instructions telling Claude how to use the resources in this repository

Requirements

• Claude Code to make use of these resources
• macOS is assumed, but can be adapted for Linux
• Homebrew for macOS dependency management
• Python 3 for the safe-find script
• Bun for the status line

Quickstart

git clone [email protected]:averycrespi/claudefiles.git
cd claudefiles
./setup.sh

The setup script will:

• Install dependencies via Homebrew
• Symlink configuration files to ~/.claude/
• Configure MCP servers in Claude Code
• Add the scripts directory to your PATH

Usage

Agents

• Use the code-reviewer agent for detailed code reviews
• Use the research-assistant agent for in-depth research and analysis
• Use the security-analyst agent to find vulnerabilities

Slash Commands

Prompt Engineering

• Use /prompt:refine prompt-file to improve your existing Claude prompts
• Use /prompt:suggest to analyze your Claude usage history and suggest custom commands

Task Lifecycle

Recommendation: Wipe the context with /clear between each step. This prevents thought leakage & context bloat.

• Use /task:specify requirements to generate a spec through Socratic questioning, written to SPEC.md
• Use /task:plan [spec-file] to transform a spec into a detailed execution plan, written to PLAN.md
• Use /task:execute [plan-file] to execute a plan from a file, with progress logged to EXECUTION.md
• Use /task:verify [spec-file] to validate the final state against a spec, reporting to VERIFICATION.md

Utilities

• Use /docs:update to analyze recent code changes and update documentation automatically
• Use /git:commit to analyze staged changes and create smart commits with auto-generated messages
• Use /git:review github-pr-url to generate AI-assisted GitHub PR review analysis to augment human reviewers

Safe Wrapper Scripts

• Claude has been instructed how to use these scripts
• The permissions prevents Claude from using the unsafe versions

Worktree Management Scripts

• Use worktree-add to create a new worktree and tmux window for a branch
  • Under the hood, worktree-init is called to start a new tmux session
• Use worktree-rm to destroy a worktree and its associated tmux window
• The Claude hooks call worktree-notify when Claude is done or needs attention

License

MIT