Verdict. Wins on discipline. If code is going to production and a bug costs real money, it stops sloppy work. I skipped it in April because it loaded 22K tokens at startup, about 11% of context, and forced a full brainstorm-to-TDD cycle on every task regardless of size. It’s installed now. I haven’t re-measured the startup cost on the current version.
what it is
An agentic skills framework and development methodology for Claude Code. The process runs spec → plan → red-green-refactor → review, with subagent-driven development and a two-stage review. It enforces TDD, worktree isolation, pre-review checklists and evidence over claims. It installs in one command from the plugin marketplace.
On 2026-09-27 the repo had 292,166 stars and 26,147 forks. In mid-April it had about 30K.
how I use it
Installed as a Claude Code plugin from the official marketplace, so its skills load in every session.
what I found
- In April a custom
/implement-speccommand covered the core value for me. It reads a plan and executes it step by step without the overhead. - TDD rigor is overkill for vault automation, agent pipelines and content work. For production SaaS it fits.
- Anthropic’s security-guidance plugin is a separate install, not part of superpowers. It scans for command injection,
eval()andnew Function(), XSS, unsafe deserialization, path traversal, SQL injection and hardcoded secrets.
compared to
- gstack wins on breadth, across design, security, QA, release and retro. The side-by-side table is on the gstack page.