I Tested the Superpowers Plugin for Claude Code: Here’s What Happened

Does Superpowers actually make Claude Code better? I audited the Superpowers plugin (300K+ GitHub stars, 1M+ installs) before installing it, then tested it. Same task, same model, with and without Superpowers, graded by 15 hidden tests Claude never saw. Real runs, real numbers. Results (one run per side, Claude Code 2.1.251, Sonnet 5.5, high effort, 20 Oct 2026) • Hidden tests: 15/15 without Superpowers, 15/15 with it. A tie. • Cross-check: B's 50 tests pass on A's code; A's 2,000 random files give 0 mismatches on B's code. • Time: 8 min 01 s without, 9 min 18 s with (+16%). • Tokens processed: 521,737 without, 893,562 with (+71%). • What changed: with Superpowers, Claude wrote the tests first and watched them fail (red), then went green. It planted 17 bugs on purpose and its tests caught all 17. It left 408 lines of tests in the project. Without it, Claude checked its work hard too (2,000 random files), but those checks didn't stay in the project. • One run each is not statistics, and a small task says nothing about big projects. Chapters 0:00 Superpowers in numbers 0:37 What Superpowers is 1:35 What runs in your session (the audit) 3:10 Install it in 60 seconds 4:15 The test: same task, with and without 5:10 Run A: plain Claude Code 5:34 Run B: with Superpowers 6:26 The hidden tests 7:13 Bigger tests by other people 7:52 The verdict 8:42 Next Install Check How to install (Claude Code) /plugin install superpowers@claude-plugins-official Pick "project scope" to try it in one project only. To remove it: /plugin uninstall superpowers Check it works: type "Let's make a react todo list". It should start brainstorming before writing any code. What it puts in your session (our own read of commit bb92a77, v7.0.0, the commit the official marketplace installs) • One SessionStart hook (startup, clear, compact) that injects the using-superpowers skill wrapped in an EXTREMELY_IMPORTANT tag. • No network calls in the hook or startup scripts, no install scripts (pattern search). • The only telemetry is declared in the README: a logo for the optional brainstorming companion, loaded with the version number. Switch it off with SUPERPOWERS_DISABLE_TELEMETRY=1 (we did, for the test). • A pattern search doesn't prove a plugin is safe. Read the skills you install. Sources Superpowers repository: https://github.com/obra/superpowers Release v7.0.0: https://github.com/obra/superpowers/releases Official marketplace page (1,009,371 installs, read 20 Oct 2026): https://claude.com/plugins/superpowers Official marketplace manifest (pins commit bb92a77): https://github.com/anthropics/claude-plugins-official Sonnet 5.5 API prices: https://claude.com/pricing Norbert Laszlo, AgentStackBench, 500 tasks on Codex (11 Jun 2026): https://norbert-laszlo.medium.com/can-a-plugin-improve-codex-benchmarking-the-superpowers-plugin-05d020066565 Mejba Ahmed, 12 Claude Code sessions: https://www.mejba.me/blog/superpowers-plugin-claude-code-review Chase AI, GSD vs Superpowers vs Claude Code: https://www.chaseai.io/blog/gsd-vs-superpowers-vs-claude-code Next Install Check (the fake REA repo): https://github.com/morluto/rea The task, the 15 hidden tests, both transcripts and the code from each run are kept in our project files; the screen recordings shown are real, sped up where labelled. Panels marked "recreated from the real transcript" redraw the real events, times and token counts. Not affiliated with Jesse Vincent, Prime Radiant or Anthropic. Superpowers is MIT-licensed; its icon is shown unaltered. Disclosure: researched and written with help from Claude, a model made by Anthropic. Synthetic narrator voice. Original score. ⚡ Wiredeep: the tech you use every day, explained. 👉 Subscribe for the next Install Check! #claudecode #superpowers #aicoding