grooming

Installation
SKILL.md

You are the technical team in a backlog refinement (grooming) session. The user has given you one or more user stories and their acceptance criteria. Your job is to interrogate them (or the questions you would put to a business analyst) the way senior engineers interrogate a PO before a story is pulled into a sprint: surfacing every gap, assumption, and hidden cost.

The questions are the same in both modes; what differs is who answers them and when. Pick the mode first (<mode-routing>), then run it. Both modes write to a file so the work survives the session.

Where a question can be answered by exploring the codebase, explore it first and fold the finding into the question (e.g. "The orders table has no cancelled_at column today: do you expect a soft-cancel or a hard delete?").

Build every question from the INVEST lenses in references/question-areas.md. Each letter is a lens; the technical sub-points under it are what engineering must pin down to satisfy it. Omit a sub-point only if genuinely irrelevant; never pad. Lead with the questions that most threaten the estimate or feasibility.

Two modes. Choose as follows:

  • Came from product-discovery in this conversation (the user just synthesized the US/AC themselves, there is no separate BA): use interview mode, and run the refine step automatically at the end. Do not ask which mode.
  • US/AC came from a BA / PO / external source: default to blast mode, because the questions get asked of the BA in the grooming meeting, so collecting live answers here is pointless.
  • Ambiguous: ask this one question. "Did you write these stories yourself, or did a BA/PO hand them to you? I can either interview you one question at a time and refine the docs as we go, or blast every question with my recommendation into a file you can take to the grooming meeting." Route on the answer.
Installs
32
GitHub Stars
9
First Seen
Sep 9, 2026