superpowers-v

Compound V — a multi-model coding sidekick for Superpowers, running on Claude Code.

You describe a feature. Claude sizes the request, plans it, splits it into non-overlapping pieces, and hands each piece to a worker in its own isolated worktree. Every write is checked against the files that worker was allowed to touch, and a review gates "done". There is no start button — a hook sizes each request as you work.

Requirements

• Claude Code ≥ 2.1.219. Compound V runs on the native Workflow runtime and on native hooks (PreToolUse, UserPromptSubmit, PostCompact, Stop). SessionStart warns below that floor.
• One ambient cost. The lane-guard hook runs on every Write/Edit/Bash call. What it costs depends on the machine — measure it on yours; the recipe is in AGENTS.md.
• One project setting. worktree.baseRef should be head, or a dependent job runs in the main checkout and such jobs are serialized. A native Claude Code setting — see Install.

Install

/plugin marketplace add https://github.com/procoders/superpowers-v
/plugin install superpowers-v@procoders

Other model CLIs are optional. Install and log into them and Compound V picks them up; without them it runs Claude-only.

• Codex (≥ 0.156.1, for the gpt-6 family): npm i -g @openai/codex → codex login
• Cursor: curl https://cursor.com/install -fsS | bash → cursor-agent login
• Antigravity: install the agy CLI → log in

Then the one setting. worktree.baseRef is a native Claude Code project setting in the project's .claude/settings.json, with two values: fresh (the default) and head. It is project-wide: head branches every worktree from the current HEAD, your own --worktree sessions included. A job that depends on another needs it to see that job's work. Without it such a job runs in the main checkout, and a wave with more than one of those is split so each runs alone — it buys parallelism, not correctness: the run completes either way.

{ "worktree": { "baseRef": "head" } }

(Optional) Context7 MCP gives the doc-validator real library docs, but only in a session it is attached to — a pre-flight agent launched inside a Workflow inherits no MCP tools and falls back to WebSearch/WebFetch (12 of the last 16 library audits ran that way). Install it with /plugin install context7@claude-plugins-official.

How to use it

Run /v:init once — it detects which model CLIs you have, saves the config, and offers the baseRef setting above.

Then just work. Describe the feature, or start brainstorming as usual. A UserPromptSubmit hook scores the first change request of each session and sizes it as DIRECT, SCOPED or FULL. There is no command to launch the orchestration; Claude runs it for you.

🎮 New here? Learn it as a game → Compound V Academy

Three gamified episodes walk you through the whole pipeline, and the cheatsheet puts commands, triggers, memory and routing on one page.

How it routes the work