<< All versions

Skill v1.0.2

currentLLM-judged scan95/100
miaodx/intuitive-flow/grill-with-docs-batch

7 files

──Details
PublishedSeptember 25, 2026 at 09:47 PM
Content Hashsha256:74e626a6075dc1af...
Git SHA8e0084d94a79
Bump Typepatch
Compare with v1.0.1
──Files
Files (1 file, 1.9 KB)
SKILL.md1.9 KBactive
SKILL.md · 40 lines · 1.9 KB

version: "1.0.2" name: grill-with-docs-batch description: Resolve material plan or domain questions against repository docs in small batches, then stop when decisions are settled. disable-model-invocation: true


Grill With Docs Batch

Challenge unresolved plan or domain decisions against the target docs and code. When asked whether questions remain, answer yes/no first. Do not manufacture questions for implementation defaults, tests, wording, or already-settled choices.

A question earns attention only if its answer changes scope, a public/private boundary, acceptance or rollout, cost, safety, ownership, or domain meaning. Use repository evidence to resolve facts before asking the user.

Supporting Primitives

Use the routed Matt Pocock grilling and domain-modeling skills as the supporting primitives for interview mechanics, glossary sharpening, and durable context/ADR discipline. This skill owns batch pacing, materiality filtering, convergence, and document handoff. It does not depend on the thin Matt grill-with-docs wrapper, so that optional wrapper is not part of this skill's runtime dependency closure.

If material questions remain, read question batches. After answers, apply accepted updates to the named document, then reassess whether another question changes the decision. Read document updates when recording answers or cleaning up plans/ADRs; do not load it for a critique that does not edit documents. Use the user's language.

Stop when an implementer can identify the change, non-goals, boundaries, proof, and next step, or the user signals process fatigue. Report Plan state:, Recommended next action:, and Shortcut:. A settled target can need zero questions.

A short approval applies to the single action just proposed. Preserve accepted scope and execute that action without another approval loop. Ask only if multiple proposed actions make the reply materially ambiguous.

← v1.0.1All versions