mirror of
https://github.com/wondelai/skills.git
synced 2026-09-14 14:38:52 +08:00
Michal Jaskolski
0dea03f507
* refactor(skills): lift all 50 skills to 9-10 vs writing-great-skills rubric Apply the writing-great-skills rubric to every skill via a shared fix playbook: cut no-op platitudes, enforce single-source-of-truth, add when-to-read pointers, de-synonym descriptions, add concrete scoring bands, and trim sprawl -- without removing framework substance. - Net -751 lines (857 ins / 1608 del) across 50 SKILL.md + 5 references - Removed 48 duplicative "Reference Files" rosters (kept inline pointers) - Cut ~67 net platitude "Ethical boundary" lines + 15 "Why it works" no-ops - Fixed influence-psychology 6-vs-7 principle bug (Unity now triggerable) - Bumped each skill version (MINOR); README/marketplace/CLAUDE untouched - All 50 YAML frontmatters validated; independent re-score 9.0-9.4 Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ygbNRWuZjaizDpcLEQ6jj * refactor(skills): finish Reference Files removal + document new format Reconcile the writing-great-skills style as canonical (per maintainer): - Remove the 2 remaining "## Reference Files" rosters (clean-code, team-topologies); every reference stays reachable via inline when-to-read pointers (verified 0 orphans across all 50). - Update CLAUDE.md's documented SKILL.md format: drop the "## Reference Files" section, document inline when-to-read pointers, mark "Why it works"/"Ethical boundary" as keep-only-if-behavior-changing, and Scoring as concrete bands tied to the diagnostic. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ygbNRWuZjaizDpcLEQ6jj * fix(skills): address Codex review on PR #11 (scoring math + INP semantics) - influence-psychology: Scoring counted 7 diagnostic rows "one point each" but the bands required 9-10 — unreachable (capped at 7). Drop the row-counting; score against qualitative bands (added a 7-8 band for a full reproducible scale) using the diagnostic as the checklist. - high-perf-browser: INP was described as blocking "the first input" (FID semantics). INP covers all interactions across the page lifecycle (matches references/core-web-vitals.md). Fixed the Why-it-works line and the Key-insights bullet. - Both bumped PATCH 1.4.0 -> 1.4.1. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ygbNRWuZjaizDpcLEQ6jj * fix(skills): repair unreachable scoring bands across 6 skills Codex flagged that several Scoring sections counted N diagnostic rows at "1 point each" while the bands required 9-10 — making the top band mathematically unreachable (max < 10). Root cause was a buggy example ("1 point per satisfied row") in CLAUDE.md's documented format. - system-design, microinteractions, refactoring-patterns, hooked-ux: score = round(passed / N × 10) so full marks reach 10. - blue-ocean-strategy, traction-eos: map row-count to the existing bands (which already span 0-10) instead of "1 point each". - CLAUDE.md: fix the Scoring guidance to require a 0-10-spanning rule and warn against "1 point per row" when there are fewer than 10 rows. - PATCH-bumped the 6 skills. (influence-psychology, same bug class, was fixed in ed83619.) Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_019ygbNRWuZjaizDpcLEQ6jj --------- Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>