A typed plan scaffold — the structure a fleet of agents executes, written before the plan itself is.
A normal plan is a to-do list — prose an agent has to interpret. A meta-plan is a typed structure: tasks with declared inputs, outputs, and dependencies, so they can be handed to parallel agents without a human re-explaining each one.
You describe the goal in a line. It emits the shape the work takes — what runs in parallel, what must wait, and how the pieces merge back into one shipped result.
Each a typed unit — declared inputs, a single output, an owner.
What must finish first. Everything else runs in parallel by default.
The rule that folds the parallel outputs back into one deliverable.
Returns a typed plan: parallelizable tasks, their dependency edges, and a merge rule — ready to hand to a fleet of agents or work solo, top to bottom.
Serial only where forced.
Machine-executable shape.
One deliverable out.
Reuse the plan shape.
The ideas here aren't invented — the skill generalizes established practice into one reusable, task-agnostic move. Below: the prior art it builds on, and the in-repo contract it actually runs on.