anthropics/claude-code: hookify, PR review, and plugin tools

What is anthropics/claude-code, and what does this listing cover?

anthropics/claude-code is Anthropic's own repository for Claude Code, the terminal based coding tool, and it carries 143,267 GitHub stars. This listing is not the CLI itself. It indexes the plugin-shaped content the repository also publishes: 40 components split into 10 skills, 14 agents, and 16 commands, filed under Flowy's agents category. Because Anthropic wrote both the tool and this bundle, it also shows how the company builds its own plugins, skills, agents, and hooks, not just a set of examples.

Who actually reaches for this

Two kinds of reader use this listing. One is building a plugin, skill, agent, or hook for Claude Code and wants Anthropic's own authoring guidance and helper agents instead of a blank file. The other wants everyday coding helpers: a git commit, a pull request review, a new Agent SDK project, or a rule that enforces a repeated correction. Most people use one side of this, not both.

The skills: eight for building plugins, two for everything else

Eight of the 10 skills form a set for plugin authorship. plugin-structure covers scaffolding a plugin and how plugin.json drives auto-discovery of its parts. skill-development covers writing a skill: structure, progressive disclosure, and a description that fires at the right moment. agent-development covers a subagent's frontmatter, system prompt, and tool access. command-development covers a slash command's frontmatter, arguments, and interactive prompts through AskUserQuestion. hook-development covers PreToolUse, PostToolUse, and Stop events, validating tool use, and blocking dangerous commands. mcp-integration covers wiring an external service into a plugin over stdio, SSE, or HTTP. plugin-settings covers storing configuration so a plugin's behavior can change per project without editing code. writing-rules covers the syntax behind a hookify rule, described below. The other two skills stand apart: claude-opus-4-5-migration moves prompts and code from Sonnet 4.0, Sonnet 4.5, or Opus 4.1 onto Opus 4.5, and frontend-design gives direction on aesthetic choices and typography so new UI does not read like a generic template.

The agents: authoring helpers, code understanding, and review

The 14 agents split into five groups. Three back the authoring skills: agent-creator builds a new subagent from a description of what it should do, plugin-validator checks a plugin's files and plugin.json after a change, and skill-reviewer checks a skill's quality and its description. Two understand existing code: code-explorer traces execution paths and dependencies in a feature, and code-architect turns that into a blueprint of files to change. Six review a change after it is written: code-reviewer for bugs, security issues, and convention drift; code-simplifier for cleaning up code without changing what it does; comment-analyzer for comments that will not stay accurate; pr-test-analyzer for whether new tests cover the change; silent-failure-hunter for error handling that swallows a failure; and type-design-analyzer for whether a new type expresses its own invariants. agent-sdk-verifier-py and agent-sdk-verifier-ts check that a Python or TypeScript Agent SDK application is configured correctly. conversation-analyzer reads back a session transcript to find behavior worth turning into a hook rule, feeding directly into hookify.

The commands: hookify, git and PR work, and the Ralph Wiggum loop

Four of the 16 commands make up the hookify toolset: hookify turns a conversation pattern or an explicit instruction into a hook rule, list shows the rules already configured, configure turns a rule on or off, and help explains the hookify commands. Seven cover everyday git and GitHub work: commit writes a commit, commit-push-pr commits, pushes, and opens a pull request in one step, code-review and review-pr both review a pull request though review-pr runs several specialist agents together, dedupe finds duplicate GitHub issues, triage-issue analyzes and labels them, and clean_gone removes local branches and worktrees whose remote branch is gone. Two are guided workflows: feature-dev for feature development with an architecture focus, and create-plugin for scaffolding, building, and validating a new plugin. new-sdk-app sets up a fresh Claude Agent SDK project, which the two SDK verifier agents above then check.

What do ralph-loop and cancel-ralph do?

The remaining two commands, ralph-loop and cancel-ralph, start and stop what the plugin calls a Ralph Wiggum loop inside a session. The fact pack's own descriptions do not go further than start and cancel, so this guide will not guess at the mechanism. If you plan to rely on it, read the repository's own documentation for it first.

Installing anthropics/claude-code

Install it from the listing page linked at the top of this guide. Flowy indexes what anthropics/claude-code publishes. It does not own, bundle, or modify any of the 40 components, and updates come from Anthropic's own repository, not from Flowy. Given how wide this set is, most people are better off installing for one group at a time, plugin authorship, review, or hookify, rather than all 40 pieces at once.