/arc:go

The full arc from idea to shipped code.

A workflow toolkit for Claude Code and Codex with 28 skills and 39 specialized agents. Skills orchestrate workflows: exploring your codebase, asking clarifying questions, spawning agents for parallel review. Agents are specialists: security auditors, performance analysts, architecture reviewers, each with deep domain expertise.

Made by Daniel Howells

$claude plugins install arc@howells

Install from the Claude Code plugin marketplace. Requires Claude Code 2.1.16+ (uses Tasks, not Todos)

$curl -fsSL https://raw.githubusercontent.com/howells/arc/main/.codex/install.sh | bash -s -- --auto-update --interval-hours 6

Codex install (uses ~/.agents/skills) with auto-update every 6 hours. One-time install without auto-update: curl -fsSL https://raw.githubusercontent.com/howells/arc/main/.codex/install.sh | bash.

Updating

$claude plugins update arc@howells

Claude Code auto-updates plugins, but you can trigger it manually. In Cursor, use this command — it doesn't auto-update.

$~/.codex/arc/.codex/update.sh

Codex manual update. If you installed with --auto-update, this runs automatically on a schedule.

Start here

/arc:go

Understands your codebase. Asks what you want to build. Guides you through design → plan → implementation with TDD.

Why this exists

If you've spent time teaching Claude how you like to work (tests first, one question at a time, review as you go) Arc is that knowledge already written down. You get to skip the setup conversations and start building.

It's opinionated because good defaults save time. If something doesn't fit how you work, take what's useful and leave the rest.

Principles

  • Tests first. Not negotiable.
  • Review is continuous, not a gate at the end.
  • Ask one question at a time. Too many questions at once stalls progress.
  • The user decides. Agents advise, they don't dictate.
···

Skills are workflows and capabilities. Most are user-invokable via /arc:name—they orchestrate the development process: gathering context, asking questions, spawning agents, and guiding implementation.