Overview
obra/superpowers treats coding-agent autonomy as a chain of governed transitions: bootstrap process awareness, clarify and approve intent, isolate and plan work, test and debug from evidence, delegate with review, then verify claims before human-controlled integration. The method can make accountability more explicit, but it cannot guarantee correctness; its ceremony should match task risk, harness capability, and available evidence.
Learning outcomes
- Superpowers governs autonomy by making attention, intent, execution, evidence, and integration transitions explicit.
- Approval clarifies what may be built, while worktrees and executable plans protect where and how it is built.
- TDD and root-cause debugging make behavior and explanations falsifiable instead of relying on agent confidence.
- Delegation, review, and fresh verification create additional evidence surfaces, but cannot guarantee correctness or replace human integration authority.
Keywords
Preview every chapter · Unlock full text, audio, and video from chapter 4
PlansFAQ
What is obra/superpowers for coding agents?
obra/superpowers is a software-development methodology that guides coding agents from an initial request through tested, reviewed, and deliberately integrated changes. It organizes autonomy around decisions, boundaries, and evidence.
How does obra/superpowers keep agent autonomy accountable?
obra/superpowers assigns attention, intent approval, execution conditions, behavioral evidence, review, and integration authority to explicit stages and owners. It makes important transitions observable without claiming the process guarantees correct software.
Does obra/superpowers require every task to use the same process?
No. obra/superpowers scales ceremony by task shape: a spike may use a disposable probe, a bounded change a short design, and architectural work a reviewed specification and plan. Approval and evidence boundaries still matter.
Can obra/superpowers guarantee that an AI-generated change is correct?
No. obra/superpowers can make assumptions, tests, reviews, and verification more visible, but incomplete requirements, test suites, harnesses, or evidence still limit what the workflow proves.