Superpowers
A process framework that makes an agent plan before it builds
Superpowers is a large plugin of skills that impose a development process on an agent: brainstorm the design, get it approved, write a spec, plan the implementation, then build. It exists to stop the failure where an agent writes two thousand lines of the wrong thing very quickly.
What it actually changes
The core move is a hard gate before implementation. The agent has to state what it intends to build and get a human yes before touching code, which converts the expensive failure mode into a cheap conversation.
It also classifies work by size, so a one line fix does not get a design document and a new subsystem does not get started on a hunch.
Whether the overhead is worth it
On a substantial change, unquestionably. Being asked clarifying questions before any code exists routinely surfaces a requirement that would have cost a rewrite.
On a genuinely small task the ceremony is friction, which is why the classification step matters. If you find yourself fighting it on trivial work, the classification is what to correct, not the framework.
Reach for it when
- an agent keeps building the wrong thing very efficiently
- the work is large enough that a wrong direction is expensive
- you want design decisions written down rather than implied by code
Watch out for
- the process is real overhead on small tasks
- it is opinionated, and disagreeing with the opinion means editing the skills
- a large skill set consumes context before your task starts
Compare with these
- SkillsFree
Anthropic Skills
The reference implementation of what a skill should look like
- skills
- claude
- documents
No licence file in the repositoryRead - SkillsFree
AGENTS.md
One instruction file that every coding agent agrees to read
- skills
- conventions
- instructions
Open convention, no license requiredRead - SkillsFree
Awesome Claude Code
A community index of commands, hooks and configurations
- skills
- claude-code
- hooks
CC0-1.0Read