Cross-Cutting

/grill-me

Relentless interview that stress-tests any draft artifact before approval. Walks the decision tree one branch at a time.

Hat Stage Upstream gate Output
Cross-cutting Review helper Any draft artifact Inline <!-- grill-me … --> scratch block on target

When to Use

Prerequisites

Stance by Artifact Type

Type Stance
epic-context Business value, success metric, ownership, deadline reality
prd Scope slice, AC coverage, edge cases, non-goals honesty
eng-spec Architecture fit, contract compatibility, rollback, observability
tdd Testability, race conditions, failure modes, public API shape
bdd / test-cases / test-spec Traceability, negative paths, non-functional coverage
Design artifacts Flow gaps, a11y, empty/error/loading states

Protocol

  1. Load target + every referenced upstream. Missing/unapproved upstream → flag as first gap and stop.
  2. Pick stance by artifact type.
  3. Walk decision tree one branch at a time. Format: state question → state own recommended answer with one-line justification → ask user to confirm / amend / override. No batching. No hedging.
  4. Explore before asking — if answer is on disk (repos/, product/, engineering/, .workbench-state/), read first, present finding.
  5. Record findings inline in HTML comment:

     <!-- grill-me session {YYYY-MM-DD}
     - [resolved] non-goal for mobile clients — explicit now at §5
     - [parked]  SLO target — deferred to spec; tracked as GRILL-1
     - [open]    rollback strategy if migration partially applied
     -->
    
  6. Exit when every branch resolved, every [parked] item has owner + date, OR user types “stop grill”. Summarise opens with one-line recommendations.

Do Not

Source

skills/grill-me/SKILL.md