- Skills
- 14
- Publisher
- obra
- License
- MIT
Superpowers is written as one process. With the Claude Code plugin, a session-start hook loads the using-superpowers skill into every conversation, and that skill requires the agent to check for a matching skill before any response, including a clarifying question. From there the skills hand off to each other, from brainstorming to writing-plans to subagent-driven-development.
The skills are written to hold under pressure. Several open with an iron law, such as no fixes without a root cause investigation first, followed by a list of the rationalizations an agent uses to skip a step and why each one is wrong.
Design and planning
Running the work
- Subagent-driven development Execute a plan task by task through fresh subagents, with a review after each one.
- Dispatching parallel agents Send independent tasks to parallel agents with prompts that stand on their own.
- Using git worktrees Make sure feature work runs in an isolated workspace with a clean test baseline.
Proving it works
- Test-driven development Write the failing test first for every feature and bug fix, then the code.
- Systematic debugging Find the root cause in four phases before proposing any fix.
- Verification before completion Run the checks and read their output before claiming anything is fixed or done.
- Requesting code review Ask for a review at the right moments, with a reviewer prompt that checks against requirements.
- Receiving code review Verify review feedback before acting on it, and push back when it is wrong.
- Finishing a development branch Verify tests, pick how to integrate the branch, and clean up the workspace.