Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →用語保存計劃月度趨勢觀察 via canonical TERMINOLOGY-TRENDS-PIPELINE — SC 需求 → 多切面搜索 → 缺口對照（雙防線查重）→ 高信心入庫 ≤20 條 → 月度趨勢報告。 TRIGGER when: routine twmd-terminology-trends-monthly fires / user says "跑用語趨勢", "支語趨勢觀察", "terminology trends", "詞庫月度更新".
.claude/skills/frank890417-twmd-terminology-trends/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -5% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -61% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -26% | 0% |
/twmd-become write 完整甦醒（Step 0-9，Write mode self-test 全過）。docs/pipelines/TERMINOLOGY-TRENDS-PIPELINE.md 後照 7 stage 嚴格執行，6 個 hard gate 一個不跳：/twmd-finale（memory 必寫）。故意最小化。SOP 100% 在 pipeline canonical，本 skill 只做 routing（REFLEXES #15：熟了跳步是最常見退化）。首輪範本：reports/terminology-zhiyu-deep-research-2026-08-04.md。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 25,562 | 18,068 | -29% | 1 | 1 | 0% | 2,681 | 892 | -67% | 0 | 0 | — |
case-02 | fail→fail | 26,357 | 17,482 | -34% | 1 | 1 | 0% | 3,183 | 747 | -77% | 0 | 0 | — |
case-03 | fail→fail | 38,015 | 21,786 | -43% | 1 | 1 | 0% | 3,565 | 1,280 | -64% | 0 | 0 | — |
case-04 | fail→pass | 23,407 | 21,680 | -7% | 1 | 1 | 0% | 2,930 | 2,232 | -24% | 0 | 0 | — |
case-05 | fail→pass | 21,003 | 16,642 | -21% | 1 | 1 | 0% | 1,922 | 1,827 | -5% | 0 | 0 | — |
case-06 | fail→pass | 21,919 | 10,736 | -51% | 1 | 1 | 0% | 2,315 | 1,232 | -47% | 0 | 0 | — |
case-11 | fail→pass | 29,895 | 8,256 | -72% | 1 | 1 | 0% | 1,967 | 762 | -61% | 0 | 0 | — |
case-07 | fail→pass | 20,742 | 12,326 | -41% | 1 | 1 | 0% | 1,918 | 1,414 | -26% | 0 | 0 | — |
case-08 | fail→fail | 15,102 | 17,202 | +14% | 1 | 1 | 0% | 2,209 | 2,242 | +1% | 0 | 0 | — |
case-09 | fail→pass | 20,290 | 22,510 | +11% | 1 | 1 | 0% | 2,681 | 1,774 | -34% | 0 | 0 | — |
case-10 | fail→pass | 28,164 | 14,125 | -50% | 1 | 1 | 0% | 2,811 | 1,630 | -42% | 0 | 0 | — |
case-12 | fail→pass | 14,508 | 8,223 | -43% | 1 | 1 | 0% | 1,092 | 760 | -30% | 0 | 0 | — |
case-13 | fail→fail | 15,550 | 2,823 | -82% | 1 | 1 | 0% | 1,938 | 799 | -59% | 0 | 0 | — |
case-14 | fail→pass | 23,751 | 10,943 | -54% | 1 | 1 | 0% | 2,427 | 1,378 | -43% | 0 | 0 | — |
case-15 | fail→pass | 21,102 | 12,970 | -39% | 1 | 1 | 0% | 2,260 | 1,293 | -43% | 0 | 0 | — |
case-16 | pass→pass | 19,614 | 15,584 | -21% | 1 | 1 | 0% | 2,129 | 1,646 | -23% | 0 | 0 | — |
case-17 | pass→pass | 19,943 | 9,084 | -54% | 1 | 1 | 0% | 1,950 | 1,460 | -25% | 0 | 0 | — |
case-18 | fail→fail | 17,854 | 16,462 | -8% | 1 | 1 | 0% | 2,176 | 2,092 | -4% | 0 | 0 | — |
case-19 | fail→pass | 21,911 | 7,985 | -64% | 1 | 1 | 0% | 2,366 | 1,379 | -42% | 0 | 0 | — |
case-20 | pass→pass | 16,986 | 30,941 | +82% | 1 | 1 | 0% | 1,685 | 3,796 | +125% | 0 | 0 | — |
case-21 | pass→fail | 17,490 | 16,126 | -8% | 1 | 1 | 0% | 1,499 | 889 | -41% | 0 | 0 | — |
case-22 | pass→pass | 13,327 | 13,544 | +2% | 1 | 1 | 0% | 1,300 | 1,235 | -5% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted, and 19 counted toward the lift figure. The other 3 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +45 percentage points is the difference between those two pass rates over the 19 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.