← ClaudeAtlas

grilllisted

Interrogate a ticket or a plan until every design decision is resolved, in one of two modes — attended (a real user answers each question) or unattended (the agent generates the hard questions and answers them itself from the codebase, then folds the answers back in). Use before planning broad work, to stress-test a plan before implementation, or when asked to grill, interrogate, or pressure-test a design.
Jars1987/gauntlet · ★ 0 · Code & Development · score 67
Install: claude install-skill Jars1987/gauntlet
<objective> Find the decisions that have not been made yet, and make them — before they get made implicitly, badly, in code. Every ticket contains unresolved branches its author did not notice. Left alone, a planner resolves them silently, by picking whatever is most obvious, and the first anyone hears of it is at review. Grilling makes those branches explicit and forces each one to a recorded decision. **The grill is worthless unless it changes the artifact.** Questions asked and answered but not folded back into the ticket or the plan are wasted work. </objective> ## Two modes | Mode | Who answers | When | Worksheet | |---|---|---|---| | **Attended** | The user, one question at a time | Phase 1 of `/gauntlet:pipeline-with-grill`, over the **ticket** | `grill-ticket.md` | | **Unattended** | You, from the codebase and the ticket | Phase 2 of every pipeline, over the **plan** | `grill-plan.md` | The technique is identical. Only the source of answers differs. The two worksheets are **separate files** and must never share one. In `pipeline-with-grill` both run in the same ticket, and overwriting the attended worksheet would destroy the record of what the user actually decided — which is the only evidence that phase 1 was completed correctly. --- ## Step 1 — Enumerate every question, to a file, before answering any Generate the **full** list up front and write it to the worksheet for your mode — `.gauntlet/work/<id>/grill-ticket.md` or `.gauntlet/work/<id>/grill-plan.md`