grill-with-docslisted

# Grill Interrogate the plan relentlessly until nothing is left silently assumed. Map it as a **design tree**: every decision branches into the decisions that hang off it. You ask the questions and you answer them; there is no user. ## Rounds The **frontier** is every decision whose prerequisites are already settled. Work it in rounds: 1. Write out the whole frontier, numbered. For each question give the candidate answers and the recommended one. 2. Settle each question: - A **fact** question (what does this code do, what does the config say, how does the existing test do it) is answered by looking: read the code, run a command, dispatch a sub-agent. A fact about a library, API, or platform that the codebase cannot answer goes through the `fabrika:research` skill. Facts come from sources, never from assumption. - A **decision** question is answered by taking the recommended answer. Tie-breakers, in order: what the spec says; what the repo already does elsewhere; the smallest change that satisfies the spec; the choice that is easiest to reverse. 3. Settled questions push the frontier outward. Recompute it and start the next round. A question whose answer depends on another still open in this round belongs to the next round. Ask a question only when its answer would change a line of the plan. A question whose every answer leaves the plan unchanged is skipped. ## Record Append each settled question to `## Decisions` in `.fabrika/work/plan.md`: ``` - **Q<n> <title>*