subagent-driven-development or superpowers: how much structure do you want?
superpowers is a broad working method for a whole session. subagent-driven-development is the narrower question of splitting work across agents. One changes how every task is approached, the other changes how one large task is divided.
The subagent-driven-development skill structures work so that several agents can run concurrently without colliding. It pays off when the pieces are truly independent, such as separate modules or parallel investigations, and it produces merge pain and contradictory decisions when they are not.
Best for
Wide, independent work: parallel research, per module changes, large mechanical sweeps.
Skip if
The pieces share state, interfaces or design decisions that have not been made yet.
Which one to install
Choose subagent-driven-development when the problem is size: work that will not fit one pass, that fans out cleanly, or that needs independent verification of its own output. The value is in the split and the handover, not in the working style.
Superpowers is a complete development methodology for coding agents, published by Jesse Vincent under MIT, built from composable skills covering brainstorming, planning, test driven development, systematic debugging, verification and subagent work. Install it when you want the agent to follow a process rather than jump straight to code. Install individual pieces when you already have a process.
Best for
Teams that want agent work to follow a repeatable method instead of depending on who wrote the prompt.
Skip if
You want one narrow behaviour, in which case take the single skill and leave the methodology.
Which one to install
Choose superpowers when you want a consistent method applied throughout, from how a plan is written to how a change is checked. It is the heavier commitment and it shows up on tasks you did not think needed a method.
They are designed to sit together rather than compete, which is why each names the other as related. If you install both, expect superpowers to set the working style and subagent-driven-development to be reached for only when a task is genuinely too big for one pass.