If you have used Claude Code, Cursor, or Codex for real work, you know the pattern. You ask for a feature, and the agent immediately starts writing code. No spec. No plan. No tests. Twenty minutes later you have something that looks right and is subtly wrong.
The fix is not a smarter model. It is a better process. That is exactly what Superpowers is.
What it is
Superpowers is an open-source software development methodology for coding agents, built as a plugin around a library of composable "skills." It works with Claude Code, Codex CLI, Gemini CLI, Cursor, GitHub Copilot CLI, and a few others.
The key idea: the skills trigger automatically. You do not have to remember to invoke anything. The moment the agent sees you are building something, it stops and runs a real workflow instead of guessing.
It is MIT licensed and, at the time of writing, has over 190k GitHub stars.
How the workflow actually runs
When you ask Superpowers-equipped agent to build something, here is what happens instead of an immediate code dump:
- Brainstorming. Before any code, it asks questions and refines your rough idea into a written spec. It shows you the design in chunks short enough to actually read.
- Git worktrees. Once you approve the design, it creates an isolated workspace on a new branch and verifies a clean test baseline.
- Writing plans. It breaks the work into bite-sized tasks of roughly two to five minutes each. Every task has exact file paths, the code to write, and verification steps.
- Subagent-driven development. It dispatches a fresh subagent per task, with a two-stage review: first does the work match the spec, then is the code actually good.
- Test-driven development. Strict RED-GREEN-REFACTOR. Write a failing test, watch it fail, write the minimum code to pass, watch it pass, commit. Code written before its test gets deleted.
- Code review and branch cleanup happen between tasks, with critical issues blocking progress.
The plans are written to be clear enough for, in the project's own words, an enthusiastic junior engineer with no context and an aversion to testing. That is the bar. If a plan is that explicit, an agent can follow it for a long stretch without drifting.
Getting started
For Claude Code, install from the official plugin marketplace:
/plugin install superpowers@claude-plugins-official
For other harnesses (Codex, Gemini CLI, Cursor, Copilot CLI), the repo has a one-line install for each. If you use more than one agent, install it separately for each.
That is the whole setup. After that, the skills just fire on their own.
The philosophy in one line
The project rests on four principles: test-driven development always, systematic process over ad-hoc guessing, simplicity as the primary goal, and evidence over claims. Verify before declaring success.
None of this is new advice. What is new is having an agent that enforces it instead of an agent that ignores it.
Is it useful?
Yes, with one honest caveat.
It is genuinely useful if you do non-trivial work and have been burned by an agent confidently shipping broken code. Because the agent follows a plan you already approved, it can run autonomously much longer without going off the rails. The structure is the value.
The caveat: it is deliberately slower for trivial work. If you just want to rename a variable or fix a typo, the brainstorm-then-spec-then-plan ceremony is overkill. Use Superpowers for features and real changes, not one-line fixes.
The takeaway
The model is rarely the bottleneck anymore. The bottleneck is that agents skip the boring, disciplined parts of engineering. Superpowers makes those parts mandatory. If your agent keeps producing plausible-looking garbage, this is the cheapest fix available: it is free, it installs in one command, and it changes nothing about how you work except the quality of the output.
Top comments (0)