What Happened
A GitHub project called Superpowers has gained significant traction, amassing over 40,900 stars. It is described by its creator as a "complete operating system for AI coding agents." The framework is designed to impose a rigorous, structured workflow on agents like Anthropic's Claude Code, OpenAI's Codex, and OpenCode to combat the common, ad-hoc approach where agents "guess what you want, write code before understanding the problem, skip tests, and produce spaghetti."
What Superpowers Does
Superpowers is not a prompt template but a system that dictates a specific development lifecycle for an AI agent. According to the source, the workflow is as follows:
Specification Brainstorming: Before writing any code, the agent is forced to stop and collaborate with the user. It asks questions to refine the specification of what is being built and presents the plan in digestible chunks for user approval.
Implementation Planning: After spec approval, the agent creates a detailed implementation plan. The plan's goal is to be clear enough for "an enthusiastic junior engineer with poor taste and no judgement" to follow.
Subagent-Driven Development: The system launches fresh subagents for individual tasks. Each task undergoes a two-stage code review: first for specification compliance, then for code quality. This structure is intended to allow the agent to run autonomously for hours without deviating from the original plan.
Enforced Test-Driven Development (TDD): The framework mandates a strict TDD cycle: write a failing test, watch it fail, write minimal code to pass the test, watch it pass, then commit. The system "literally deletes code written before tests."
Verification and Cleanup: Upon task completion, the agent verifies the work, presents the user with options (merge, create a pull request, keep, or discard), and performs cleanup.
The project's stated philosophy is "brutal," emphasizing systematic processes over ad-hoc approaches, evidence over claims, complexity reduction, and verification before declaring success.
Technical Details & Availability
- Compatibility: Works with Claude Code (via plugin installation), Codex, and OpenCode.
- License: 100% open-source under the MIT License.
- Access: The project is available on GitHub, where it has garnered its 40.9K stars.
The core value proposition is shifting the human role from constant babysitter of a chaotic agent to a reviewer and approver within a disciplined, automated software engineering pipeline.
