Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Tactic: Compile the feeding plan into an ARA via the external compiler, then run Level-2 rigor review over it
.claude/skills/yogsoth-ai-compile-and-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-01 | ✗→✓ | ▲ Improved | -36% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -20% | 0% |
| case-05 | ✗→✓ | ▲ Improved | -24% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -36% | 0% |
Key question: 把投喂计划编译成一份内部一致的 ARA,并做认识论审查。
Skill load ara-compile —— 把投喂计划整理成 compiler 的 $ARGUMENTS(路径 + 标主干 + 大方向约束 + --output ../ara/),一次 inline 跑 compiler, 得 ../ara/(Seal Level 1 已过)。
Skill load ara-rigor-review —— 对 ../ara/ 跑 rigor-reviewer 的 Level 2六维审查,得 ara/level2_report.json,透传给用户。
整条链是一次性前向流,不是自动迭代循环。D5 低分由用户决定是否回 context-review 补打捞。
ara/(完整 ARA)+ ara/level2_report.json。
<!-- BEGIN available-tables (generated) -->
Optional, no fixed order; the final leaf is always a sop.
| SOP | When to use | | --- | --- | | ara-compile | SOP: Turn the feeding plan into the compiler's $ARGUMENTS and run the external ARA compiler once inline to produce ../ara/ | | ara-rigor-review | SOP: Run the external ARA rigor-reviewer (Seal Level 2, six-dimension semantic review) over ../ara/ and pass its level2_report.json to the user |
<!-- END available-tables (generated) -->
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→pass | 9,160 | 3,830 | -58% | 1 | 1 | 0% | 1,502 | 959 | -36% | 0 | 0 | — |
case-07 | fail→fail | 12,783 | 1,624 | -87% | 1 | 1 | 0% | 1,915 | 552 | -71% | 0 | 0 | — |
case-02 | fail→fail | 11,695 | 1,594 | -86% | 1 | 1 | 0% | 1,936 | 595 | -69% | 0 | 0 | — |
case-03 | fail→pass | 18,126 | 13,438 | -26% | 1 | 1 | 0% | 3,188 | 2,431 | -24% | 0 | 0 | — |
case-04 | fail→pass | 8,033 | 4,162 | -48% | 1 | 1 | 0% | 1,256 | 1,002 | -20% | 0 | 0 | — |
case-05 | fail→pass | 23,665 | 7,265 | -69% | 1 | 1 | 0% | 2,049 | 1,550 | -24% | 0 | 0 | — |
case-06 | fail→pass | 9,156 | 3,318 | -64% | 1 | 1 | 0% | 1,487 | 950 | -36% | 0 | 0 | — |
case-08 | fail→pass | 9,628 | 2,637 | -73% | 1 | 1 | 0% | 1,521 | 801 | -47% | 0 | 0 | — |
case-09 | fail→pass | 9,113 | 2,103 | -77% | 1 | 1 | 0% | 1,394 | 696 | -50% | 0 | 0 | — |
case-10 | fail→pass | 10,472 | 3,071 | -71% | 1 | 1 | 0% | 1,574 | 828 | -47% | 0 | 0 | — |
case-11 | fail→pass | 12,638 | 4,466 | -65% | 1 | 1 | 0% | 2,025 | 1,123 | -45% | 0 | 0 | — |
case-16 | fail→pass | 36,649 | 1,931 | -95% | 1 | 1 | 0% | 748 | 656 | -12% | 0 | 0 | — |
case-12 | pass→pass | 12,255 | 2,860 | -77% | 1 | 1 | 0% | 1,774 | 836 | -53% | 0 | 0 | — |
case-13 | fail→pass | 11,255 | 2,687 | -76% | 1 | 1 | 0% | 552 | 729 | +32% | 0 | 0 | — |
case-14 | fail→pass | 3,715 | 2,328 | -37% | 1 | 1 | 0% | 549 | 672 | +22% | 0 | 0 | — |
case-15 | fail→pass | 5,156 | 2,186 | -58% | 1 | 1 | 0% | 706 | 692 | -2% | 0 | 0 | — |
case-17 | fail→pass | 5,055 | 2,337 | -54% | 1 | 1 | 0% | 664 | 710 | +7% | 0 | 0 | — |
case-18 | fail→pass | 9,956 | 2,471 | -75% | 1 | 1 | 0% | 1,356 | 645 | -52% | 0 | 0 | — |
case-19 | fail→pass | 8,802 | 2,959 | -66% | 1 | 1 | 0% | 1,191 | 749 | -37% | 0 | 0 | — |
case-20 | fail→pass | 9,853 | 1,901 | -81% | 1 | 1 | 0% | 1,662 | 604 | -64% | 0 | 0 | — |
case-21 | pass→pass | 9,103 | 3,474 | -62% | 1 | 1 | 0% | 1,272 | 891 | -30% | 0 | 0 | — |
case-22 | pass→pass | 11,019 | 3,336 | -70% | 1 | 1 | 0% | 1,549 | 849 | -45% | 0 | 0 | — |
case-23 | pass→pass | 25,501 | 5,026 | -80% | 1 | 1 | 0% | 2,020 | 1,158 | -43% | 0 | 0 | — |
case-24 | fail→pass | 28,399 | 5,167 | -82% | 1 | 1 | 0% | 2,220 | 1,220 | -45% | 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. 24 cases were attempted, and 22 counted toward the lift figure. The other 2 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 +75 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.