spec-expansion

Turn a sparse seed into a high-recall spec draft. The job: when a coding task starts from a few lines of intent, the agent silently picks its own definition of "done" and builds happy-path-only slop. This skill systematically expands the seed across objects / states / paths / edge cases so the right and complete thing gets specified on the first pass — then hands the result to code-toolkit's VERIFY layer.

This is the GENERATE layer of the GENERATE → DECLARE → VERIFY pipeline. It produces a spec draft; it does not run TDD, write code, or review code (see Boundary below).

Executor model — who does what

You (the agent running this skill) are the executor. You supply the LLM reasoning (object extraction, OOUX fan-out, lens pruning, scenario drafting) and the parallel fan-out. There is no external runtime, no API key, no program to install — the method rides on the host agent you are already in.