Follow-up to PR #278 (interview frontier-rounds rewrite). Blocked until it merges.
The one-question-at-a-time cadence the interview skill abandoned still appears in sibling planning skills. A fresh-context audit classified every site and staged ready-to-apply diffs (priority order):
- Diff A (must-fix):
architect/SKILL.md Step 4.6 + architect/context/tag-decisions.md — both describe "an interview round (one question at a time, recommendation + basis)"; stale the moment interview means frontier-rounds. The parenthetical is duplicated across the two files — collapse to a single source while fixing.
- Diff B:
architect/SKILL.md scope-clarity clause ("prose, one question at a time") — already internally inconsistent with the adjacent Open Decisions block, which batches.
- Diff C:
prd/SKILL.md Step 4 — "never batch unrelated questions" is the direct inverse of the frontier philosophy; PRD's section-organized questions are the best-fit conversion case. PRD evals need no change.
- Diff D+E:
design/SKILL.md Key behaviors + two design/evals.json assertions — design threads' dependency tracking already IS a frontier graph.
- Leave as is:
brainstorm/SKILL.md single intake question (one question is genuinely correct; a frontier of one degenerates anyway).
Also: siblings should read the same use_ask_user_question userConfig rather than re-deciding prose-vs-card inline.
Audit provenance: fresh-context sub-agent, 2026-07-17, verified against files (full memo machine-local; diffs reproducible from this description). Scale: one small PR, planning plugin minor bump.
🤖 Generated with Claude Code
Follow-up to PR #278 (interview frontier-rounds rewrite). Blocked until it merges.
The one-question-at-a-time cadence the interview skill abandoned still appears in sibling planning skills. A fresh-context audit classified every site and staged ready-to-apply diffs (priority order):
architect/SKILL.mdStep 4.6 +architect/context/tag-decisions.md— both describe "an interview round (one question at a time, recommendation + basis)"; stale the moment interview means frontier-rounds. The parenthetical is duplicated across the two files — collapse to a single source while fixing.architect/SKILL.mdscope-clarity clause ("prose, one question at a time") — already internally inconsistent with the adjacent Open Decisions block, which batches.prd/SKILL.mdStep 4 — "never batch unrelated questions" is the direct inverse of the frontier philosophy; PRD's section-organized questions are the best-fit conversion case. PRD evals need no change.design/SKILL.mdKey behaviors + twodesign/evals.jsonassertions — design threads' dependency tracking already IS a frontier graph.brainstorm/SKILL.mdsingle intake question (one question is genuinely correct; a frontier of one degenerates anyway).Also: siblings should read the same
use_ask_user_questionuserConfig rather than re-deciding prose-vs-card inline.Audit provenance: fresh-context sub-agent, 2026-07-17, verified against files (full memo machine-local; diffs reproducible from this description). Scale: one small PR, planning plugin minor bump.
🤖 Generated with Claude Code