Superpowers: The 280k-Star Framework That Gives Claude Code a Development Methodology

At 280k GitHub stars, Superpowers is the most-adopted Claude Code skills framework in the ecosystem. That number inflates quickly because the idea resonates with a lot of people — but the underlying tool is genuinely worth understanding on its own terms.

What it is

Superpowers, by Jesse Vincent and the Prime Radiant team, is a set of 14 skill files shipped as a single plugin. Each file is plain Markdown with instructions for one engineering practice: test-driven development, systematic debugging, writing plans, executing plans, parallel subagent dispatch, code review (both giving and receiving), git worktree management, and a few meta-skills for running the whole system.

The install for Claude Code is one command:

/plugin install superpowers@claude-plugins-official

After that, Claude picks up the relevant skill based on what it's doing — no extra prompting required. It also works with Cursor, Codex, Gemini CLI, OpenCode, and Devin; install commands differ per platform and are all documented in the repo.

When it actually helps

The framework's thesis is that AI agents aren't failing because they lack capability — they're failing because they skip the process a disciplined developer would follow. They jump straight to coding before writing a plan. They don't write tests first. They declare victory without checking if the thing actually works. Superpowers installs that process as composable skills the agent reaches for on its own.

In a typical session: the brainstorming and writing-plans skills fire before any code gets written. test-driven-development nudges Claude to write the failing test first. verification-before-completion kicks in before closing out a task. requesting-code-review structures how Claude hands off to a human or a subagent. These are the kinds of habits you'd otherwise have to encode in every project's CLAUDE.md — here they're reusable across everything.

The real limitation

Skills are bootstrapped at session start, which means session compaction kills them. If a long session hits Claude's context limit and compacts, the skills bootstrap may quietly evict from context. The agent keeps going — minus the process discipline — and you won't necessarily notice. The recommended fix is to start a fresh session rather than continuing a compacted one. This is a real footgun in long autonomous runs.

The methodology overhead is also not free. Superpowers is designed for hour-long autonomous coding tasks, not quick one-off fixes. If you're tweaking a config value or renaming a variable, the skill scaffolding adds friction rather than removing it. And telemetry is enabled by default — check the docs and opt out via environment variable before pointing it at anything you'd rather keep private.

Credit and links

Built by Jesse Vincent and the Prime Radiant team. The full framework, all 14 skill source files, and platform-specific install commands are at obra/superpowers on GitHub.