Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use this skill when the user asks "/think" or when you face a complex architectural challenge.
.claude/skills/majiayu000-skills/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | -18% | 0% |
| case-01 | ✗→✓ | ▲ Improved | 2% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -14% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -59% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -20% | 0% |
name: starlight-core description: Enhance Claude with Deep Thinking (First Principles, Systems Thinking)
Use this skill when the user asks "/think" or when you face a complex architectural challenge.
You MUST pause and construct a markdown response based on the active Strategy.
Active Strategy: 02_PROTOCOL/STRATEGIES/FIRST_PRINCIPLES.md (Default) Active Strategy: 02_PROTOCOL/STRATEGIES/SYSTEMS_THINKING.md (Use for Architecture)
> Thinking Process: > Principle: (e.g., Separation of Concerns) > Application: (e.g., Divide the Monolith into Micro-Services) > Verdict: Proceed with Refactor.
Before finalizing, ask: "Does this solution serve the User's long-term vision?"
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-11 | fail→pass | 27,659 | 22,073 | -20% | 1 | 1 | 0% | 3,519 | 2,889 | -18% | 0 | 0 | — |
case-05 | pass→fail | 11,382 | 9,998 | -12% | 1 | 1 | 0% | 1,084 | 1,079 | -0% | 0 | 0 | — |
case-01 | fail→pass | 25,695 | 24,477 | -5% | 1 | 1 | 0% | 3,091 | 3,146 | +2% | 0 | 0 | — |
case-02 | fail→pass | 28,321 | 23,760 | -16% | 1 | 1 | 0% | 3,460 | 2,961 | -14% | 0 | 0 | — |
case-03 | fail→pass | 26,427 | 15,248 | -42% | 1 | 1 | 0% | 4,126 | 1,678 | -59% | 0 | 0 | — |
case-04 | pass→pass | 1,999 | 8,952 | +348% | 1 | 1 | 0% | 319 | 773 | +142% | 0 | 0 | — |
case-06 | pass→fail | 5,614 | 10,919 | +94% | 1 | 1 | 0% | 1,207 | 1,403 | +16% | 0 | 0 | — |
case-07 | fail→pass | 26,448 | 13,844 | -48% | 1 | 1 | 0% | 3,408 | 2,732 | -20% | 0 | 0 | — |
case-08 | fail→pass | 25,300 | 19,754 | -22% | 1 | 1 | 0% | 3,807 | 2,589 | -32% | 0 | 0 | — |
case-09 | fail→pass | 28,225 | 23,743 | -16% | 1 | 1 | 0% | 3,105 | 3,468 | +12% | 0 | 0 | — |
case-10 | fail→pass | 26,288 | 17,939 | -32% | 1 | 1 | 0% | 3,738 | 2,983 | -20% | 0 | 0 | — |
case-12 | fail→pass | 19,581 | 19,668 | +0% | 1 | 1 | 0% | 3,115 | 2,037 | -35% | 0 | 0 | — |
case-13 | fail→fail | 26,622 | 21,980 | -17% | 1 | 1 | 0% | 3,835 | 2,202 | -43% | 0 | 0 | — |
case-14 | fail→fail | 24,205 | 17,347 | -28% | 1 | 1 | 0% | 3,566 | 2,934 | -18% | 0 | 0 | — |
case-15 | fail→pass | 17,622 | 14,332 | -19% | 1 | 1 | 0% | 1,823 | 2,528 | +39% | 0 | 0 | — |
case-16 | fail→pass | 34,935 | 16,477 | -53% | 1 | 1 | 0% | 3,959 | 2,828 | -29% | 0 | 0 | — |
case-17 | fail→pass | 19,619 | 42,128 | +115% | 1 | 1 | 0% | 2,890 | 3,363 | +16% | 0 | 0 | — |
case-18 | fail→pass | 19,991 | 16,730 | -16% | 1 | 1 | 0% | 2,966 | 1,959 | -34% | 0 | 0 | — |
case-19 | fail→pass | 26,294 | 24,667 | -6% | 1 | 1 | 0% | 3,164 | 3,295 | +4% | 0 | 0 | — |
case-20 | fail→pass | 23,005 | 14,226 | -38% | 1 | 1 | 0% | 2,721 | 2,481 | -9% | 0 | 0 | — |
case-21 | fail→pass | 26,682 | 22,204 | -17% | 1 | 1 | 0% | 3,149 | 2,975 | -6% | 0 | 0 | — |
case-22 | fail→pass | 20,366 | 18,268 | -10% | 1 | 1 | 0% | 3,126 | 2,978 | -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. The headline lift of +67 percentage points is the difference between those two pass rates over the 22 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.