zsh-craft

Claude Code Plugins

A collection of plugins for Claude Code that teach Claude things it doesn't know out of the box — idiomatic shell scripting, GitHub API patterns, full fzf syntax, multi-agent code review, and more.

The problem: Claude Code is good at general programming but has blind spots. It writes bash when you're in zsh. It uses bare | fzf without previews. It doesn't know gh milestone list isn't a real command. These plugins fix that by giving Claude domain-specific knowledge that activates automatically when relevant.

How plugins work: Skills auto-activate based on context (mention a VM and exe-dev kicks in, write a zsh script and zsh-craft takes over). Commands are explicit (/code-review, /exe-ls). You install what you need — they're independent.

Install

Add the marketplace, then install what you want:

# From the terminal
claude plugin marketplace add metcalfc/claude-plugin
claude plugin install chad-tools
claude plugin install gh-recipes
claude plugin install fzf-power
claude plugin install zsh-craft
claude plugin install exe-dev
claude plugin install claude-craft
claude plugin install lefthook

# Inside Claude Code
/plugin marketplace add metcalfc/claude-plugin
/plugin install chad-tools

---

chad-tools

Multi-agent code review and dev workflow automation.

The headline feature is /code-review — a single command that auto-detects what you've changed (unstaged, staged, last commit), selects the right review agents, runs them in parallel, and either posts a GitHub review (if a PR exists) or reports findings in your terminal. It also reviews PRs by number (/code-review #123).

Five specialized agents, each focused on a different aspect of code quality:

Agent	Focus	Activates when
code-reviewer	Security, correctness, architecture, style	Always
silent-failure-hunter	Swallowed errors, lost context, misleading fallbacks	Error handling in diff
pr-test-analyzer	Test correctness, coverage gaps, flaky patterns	Test files in diff
comment-analyzer	Doc accuracy, stale comments, misleading docs	Comments/docstrings in diff
type-design-analyzer	Type design, breaking changes, leaky abstractions	Type definitions in diff

Agent selection uses comprehensive trigger patterns across Go, Rust, JS/TS, Python, Ruby, Bash/Zsh, and more. Findings below 80% confidence are filtered. Duplicates are deduplicated. Every comment posted matters.

Commands

Command	What it does
/chad-tools:code-review	Multi-agent code review — local diff or PR
/chad-tools:deslop	Strip AI code slop from branch diff
/chad-tools:humanize	Rewrite prose to remove AI writing patterns
/chad-tools:done	Mark worktree done for cwprune
/chad-tools:pick-next	Pick next issue and launch worktree
/chad-tools:audit-plugins	Audit all marketplace plugins for accuracy

Skills (auto-activate)

Skill	What it does
resume-branch	Check rebase status, PR state, orient to where you left off
gen-script	Generate standalone bash/python/JS scripts
crystallize	Turn a repeated pattern into a new Claude Code skill
protect-branch	Add branch protection hooks to a repo
resolve-reviews	Reply to PR review comments and resolve conversations

---

gh-recipes

Recipes for gh CLI operations that don't have built-in subcommands.

Ever tried gh milestone list and got "unknown command"? This plugin teaches Claude how to use gh api for the operations GitHub CLI doesn't cover natively. It auto-activates when Claude hits an unknown gh subcommand or when you ask about GitHub operations that need the API directly. Includes a PostToolUse hook that detects gh errors and nudges toward the right recipe.

What's covered