I ported obra/superpowers to OpenClaw — no Claude Code dependency

Jesse Vincent's [superpowers](https://github.com/obra/superpowers) framework is one of the more thoughtful approaches to structured AI-assisted development I've seen — plan-first, test-driven, with a two-stage review loop. There are existing ClawHub bridges for it, but they're thin proxies that just route Claude Code skill invocations through the gateway. They work, but they don't feel native. So I ported it from scratch for OpenClaw: `sessions_spawn` for agent dispatch, `update_plan`for task tracking, and three dedicated agents (`coding` orchestrator + `implementer` \+ `reviewer` leaves) that wire together the full workflow. The main session just delegates — it doesn't manage any of the inner loop. The README is written primarily for OpenClaw agents to follow, not humans — the idea being you clone the repo and ask your assistant to set it up for you. It walks through a Q&A (model choices, branch completion strategy, review intensity) before touching anything. MIT licensed. Feedback welcome, especially from anyone who's used the original superpowers and has opinions on what I got wrong. **GitHub:** [https://github.com/rqlangley/openclaw-superpowers](https://github.com/rqlangley/openclaw-superpowers)