Agent skills
systematic-debugging
Forces a method before any fix is proposed.
Fires on any bug, failing test or surprise, and it fires before the agent is allowed to suggest a change. The order is the whole point: understand the failure, then propose. It is the counterweight to an agent that is eager to be helpful.
npx skills add https://github.com/obra/superpowers --skill systematic-debugging
DebuggingFull page & trust audit →test-driven-development
Test first, on every feature and every bugfix.
The Superpowers take on TDD, written to trigger before implementation code exists rather than after. Pairs with verification-before-completion, which is what stops the agent declaring the tests green without running them.
npx skills add https://github.com/obra/superpowers --skill test-driven-development
TestingFull page & trust audit →writing-plans
Turns a spec into a plan before anything is touched.
For multi-step work, where the failure mode is an agent that starts confidently in the wrong place. It produces a plan you can read and correct while correcting it is still cheap.
npx skills add https://github.com/obra/superpowers --skill writing-plans
PlanningFull page & trust audit →executing-plans
Works a written plan through to the end.
The other half of writing-plans. It keeps the agent on the plan it agreed to, step by step, rather than quietly redesigning halfway down.
npx skills add https://github.com/obra/superpowers --skill executing-plans
requesting-code-review
Asks for review at the points where review is worth having.
Triggers on finishing a task, landing a significant feature or preparing to merge, and frames the request around what the work was meant to do rather than what changed.
npx skills add https://github.com/obra/superpowers --skill requesting-code-review
receiving-code-review
Handles review feedback with rigour instead of agreement.
The interesting one. It requires the agent to verify a suggestion before implementing it, and to push back where the feedback is unclear or wrong, which is the opposite of what an agent does by default.
npx skills add https://github.com/obra/superpowers --skill receiving-code-review
verification-before-completion
No success claim without the command output behind it.
Fires the moment the agent is about to say something is done, fixed or passing. It has to run the verification and read the output first. If you have ever been told the build is green when it is not, this is the skill for that.
npx skills add https://github.com/obra/superpowers --skill verification-before-completion
using-git-worktrees
Gives the work an isolated workspace before it starts.
Sets up a worktree so a feature or a plan runs somewhere other than your current checkout. Cheap insurance when an agent is about to touch a lot of files.
npx skills add https://github.com/obra/superpowers --skill using-git-worktrees
DeliveryFull page & trust audit →brainstorming
Explores intent before any creative work begins.
Runs ahead of building a feature or a component and digs into what is actually wanted. It exists because the most expensive agent mistake is building the wrong thing correctly.
npx skills add https://github.com/obra/superpowers --skill brainstorming
PlanningFull page & trust audit →subagent-driven-development
Runs the independent parts of a plan in parallel.
For a plan whose tasks do not depend on each other. It hands them to separate agents in the same session, which is faster and keeps each one's context focused on its own piece.
npx skills add https://github.com/obra/superpowers --skill subagent-driven-development
DeliveryFull page & trust audit →
Jesse Vincent