How it works
A plan fails on what nobody asked about. This runs the grilling method: model the plan as a design tree, and each round ask the whole frontier — every question whose prerequisites are already settled — with a recommended answer on each. Your answers push the frontier outward. Done is when it is empty.
The prescan measures, free
Placeholders, vague quantifiers, unresolved choices, undefined acronyms, hedges, untestable commitments — each with a line number and an excerpt you can check by eye. Plus the document-level signals a finished plan should carry. No sign-in, no model call, no charge.
The round asks the frontier
Every question arrives with a recommended answer, the reason it is askable now, and the prescan findings it would settle. Questions that depend on this round's answers are listed as deferred, not asked early.
The checks hold it to the facts
Frontier discipline, unique ids, a recommendation on every question, and coverage of every prescan finding by id — all re-measured in this tab and printed next to the round. A finding the round neither asks about nor dismisses is named.
A derived work of the @mattpocock/grill-me skill (MIT). The grilling method — design tree, rounds, frontier, a recommendation on every question — is theirs; the prescan, the checks, the session state and the brief are this app's.