Superpowers: Testable Coding Agent Workflows
A practical review of Obra's Superpowers, a cross-agent skill system for brainstorming, worktrees, TDD, code review, and verified completion gates.
Published: 2026-05-12
Summary
Superpowers turns software-development discipline into installable agent behavior. It is strict by design: agents should ask before building, plan before executing, test before coding, review before merging, and verify before claiming completion.
Execution paths from this guide
Move from reading to action: validate by task intent, compare alternatives, then open tool reviews for final checks.
Browse by task • Compare • Tools • Deals
Priority tasks: Code review tasks • Debugging tasks • Code generation tasks
Priority compares: ChatGPT vs Claude • Claude vs Gemini
Priority tool reviews: Claude review • ChatGPT review • Gemini review
What Superpowers really is
Superpowers is a complete software-development methodology packaged as Agent Skills. It supports several coding-agent environments, but its main idea is platform-independent: a coding agent should not jump from user request to production code. It should clarify the problem, write a design, create a worktree, make a plan, execute in small tasks, use test-driven development, request review, respond to review, and verify before completion.
The enforcement layer is the product
Many skill packs describe good behavior. Superpowers tries to make good behavior hard to skip. Its bootstrap instructions tell the agent to check relevant skills before acting, and the workflow skills contain hard gates: no production code before a failing test, no completion claim without fresh verification, no vague debugging without root-cause tracing. That is the difference between documentation and an operating rule.
Why it feels strict
The system is intentionally rigid because its enemy is agent improvisation. A model that writes code too soon can look productive while creating hidden debt. Superpowers pushes the model back into a disciplined loop: brainstorm, plan, isolate, test, review, and finish deliberately. For teams that already believe in TDD and written plans, this rigidity is a strength. For teams that prefer lightweight exploration, it may feel like too much ceremony.
Subagents are treated as review machinery
The subagent-driven-development flow is a useful design example. It does not merely spawn more agents and hope parallelism helps. It gives each task to a fresh worker, then runs separate review passes for spec compliance and code quality. That separation is important. A worker can implement the right feature badly, or implement clean code that misses the plan. Superpowers names both failure modes.
What skill authors should learn
Superpowers treats skill text as behavioral code. That means it adds tests, integration checks, contribution rules, and platform-specific bootstrap logic. This is a mature instinct. If a skill is supposed to change agent behavior, then the project needs a way to test whether the behavior actually changes. A README alone is not enough.
Verdict
Superpowers is one of the clearest examples of Agent Skills as governance rather than decoration. Its best users will be builders who want their agents to slow down at the exact moments where agents usually create risk: before scoping, before coding, before fixing, before merging, and before declaring success.
Primary Sources
These links point to the source repositories or official documentation used for this guide.
Frequently asked questions
Is Superpowers beginner friendly?
It is beginner friendly if the user wants structure. It explains a complete workflow, but it also expects the user to accept planning, TDD, review, and verification as normal parts of development.
Why does Superpowers emphasize TDD so strongly?
TDD gives the agent an objective feedback loop before it writes production code. That reduces the chance of confident but untested changes and makes completion claims easier to verify.
What is the main limitation of Superpowers?
The main limitation is fit. Teams that do not want mandatory TDD, structured brainstorming, worktree isolation, or review gates may find the system heavier than their actual workflow requires.
Related Guides
Explore related tools
Use the directory to compare tools, evaluate offers, and browse by task.