Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Semiont 4.5-beat full heartbeat cycle (twmd- namespaced; 既有 /heartbeat 雙軌共存). v3.0 super-thin routing. TRIGGER when: user says "twmd 心跳", "twmd-heartbeat", "完整心跳 (twmd)", or prefers twmd- namespace consistency.
.claude/skills/frank890417-twmd-heartbeat/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | -62% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -57% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -46% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 161% | 0% |
跟 /heartbeat 完全等價，只是 namespace 統一為 twmd-。
/twmd-become full（per BECOME §Step 0 high-stake）。docs/semiont/HEARTBEAT.md v3.0（super-thin shell ~170 行）取得 4.5 拍 conceptual framework + pipeline pointer。v3.0 reframe：HEARTBEAT.md 從 745 行 SOP 全載降級為 super-thin pipeline router（per 哲宇 dialogue「heartbeat 我也很少用 routine 取代了」）。Routine 飛輪自轉 cover 日常；手動 heartbeat = 跨 routine / 全器官 audit / strategy 場景。
設計背景：reports/become-boot-mode-design-2026-05-13.md + HEARTBEAT v3.0 super-thin 2026-05-13。
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 5,111 | 5,600 | +10% | 1 | 1 | 0% | 843 | 744 | -12% | 0 | 0 | — |
case-02 | fail→fail | 4,324 | 5,095 | +18% | 1 | 1 | 0% | 297 | 691 | +133% | 0 | 0 | — |
case-08 | fail→pass | 9,225 | 1,998 | -78% | 1 | 1 | 0% | 1,721 | 661 | -62% | 0 | 0 | — |
case-03 | fail→fail | 9,719 | 5,390 | -45% | 1 | 1 | 0% | 1,761 | 735 | -58% | 0 | 0 | — |
case-04 | pass→pass | 12,683 | 8,168 | -36% | 1 | 1 | 0% | 2,206 | 1,731 | -22% | 0 | 0 | — |
case-05 | pass→pass | 6,374 | 3,200 | -50% | 1 | 1 | 0% | 1,245 | 858 | -31% | 0 | 0 | — |
case-06 | pass→fail | 4,132 | 3,864 | -6% | 1 | 1 | 0% | 700 | 495 | -29% | 0 | 0 | — |
case-07 | fail→pass | 8,103 | 1,680 | -79% | 1 | 1 | 0% | 1,564 | 665 | -57% | 0 | 0 | — |
case-09 | fail→fail | 7,750 | 5,465 | -29% | 1 | 1 | 0% | 1,382 | 726 | -47% | 0 | 0 | — |
case-10 | fail→fail | 8,868 | 4,420 | -50% | 1 | 1 | 0% | 1,553 | 694 | -55% | 0 | 0 | — |
case-11 | fail→pass | 10,175 | 3,543 | -65% | 1 | 1 | 0% | 1,615 | 865 | -46% | 0 | 0 | — |
case-12 | fail→pass | 18,857 | 10,794 | -43% | 1 | 1 | 0% | 1,690 | 2,057 | +22% | 0 | 0 | — |
case-13 | fail→pass | 24,653 | 12,963 | -47% | 1 | 1 | 0% | 863 | 2,249 | +161% | 0 | 0 | — |
case-14 | fail→pass | 16,534 | 9,101 | -45% | 1 | 1 | 0% | 837 | 1,797 | +115% | 0 | 0 | — |
case-15 | fail→pass | 10,840 | 10,181 | -6% | 1 | 1 | 0% | 1,878 | 1,667 | -11% | 0 | 0 | — |
case-16 | fail→pass | 8,639 | 14,452 | +67% | 1 | 1 | 0% | 1,526 | 2,426 | +59% | 0 | 0 | — |
case-17 | fail→pass | 3,995 | 2,777 | -30% | 1 | 1 | 0% | 717 | 888 | +24% | 0 | 0 | — |
case-18 | pass→pass | 8,046 | 4,212 | -48% | 1 | 1 | 0% | 1,465 | 1,161 | -21% | 0 | 0 | — |
case-19 | fail→pass | 7,570 | 2,396 | -68% | 1 | 1 | 0% | 1,353 | 815 | -40% | 0 | 0 | — |
case-20 | fail→pass | 8,612 | 2,397 | -72% | 1 | 1 | 0% | 1,482 | 783 | -47% | 0 | 0 | — |
case-21 | fail→pass | 10,639 | 4,479 | -58% | 1 | 1 | 0% | 1,963 | 1,287 | -34% | 0 | 0 | — |
case-22 | fail→pass | 12,402 | 10,812 | -13% | 1 | 1 | 0% | 2,338 | 2,781 | +19% | 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 14 counted toward the lift figure. The other 8 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 +55 percentage points is the difference between those two pass rates over the 14 comparable cases. 2 cases got worse with the skill loaded, and they are 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.