Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Manual-only skill for minimizing total codebase size. Only activate when explicitly requested by user. Measures success by final code amount, not effort. Bias toward deletion.
.claude/skills/davila7-reducing-entropy/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 1% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -34% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 0% | 0% |
| case-17 | ✗→✓ | ▲ Improved | 38% | 0% |
More code begets more code. Entropy accumulates. This skill biases toward the smallest possible codebase.
Core question: "What does the codebase look like after?"
Load at least one mindset from references/
Do not proceed until you've done this.
The goal is less total code in the final codebase - not less code to write right now.
Measure the end state, not the effort.
Not "what's the smallest change" - what's the smallest result.
Count lines before and after. If after > before, reject it.
Every change is an opportunity to delete. Ask:
See references/ for philosophical grounding.
To add new mindsets, see adding-reference-mindsets.md.
Bias toward deletion. Measure the end state.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 10,916 | 11,686 | +7% | 1 | 1 | 0% | 2,354 | 2,007 | -15% | 0 | 0 | — |
case-02 | fail→fail | 7,761 | 3,357 | -57% | 1 | 1 | 0% | 1,409 | 996 | -29% | 0 | 0 | — |
case-03 | fail→pass | 11,456 | 10,601 | -7% | 1 | 1 | 0% | 2,261 | 2,285 | +1% | 0 | 0 | — |
case-04 | pass→pass | 11,803 | 10,015 | -15% | 1 | 1 | 0% | 1,909 | 2,167 | +14% | 0 | 0 | — |
case-05 | pass→pass | 11,384 | 11,087 | -3% | 1 | 1 | 0% | 1,649 | 2,192 | +33% | 0 | 0 | — |
case-06 | fail→pass | 9,564 | 10,922 | +14% | 1 | 1 | 0% | 1,455 | 2,451 | +68% | 0 | 0 | — |
case-13 | fail→pass | 11,931 | 3,337 | -72% | 1 | 1 | 0% | 1,746 | 1,150 | -34% | 0 | 0 | — |
case-07 | pass→pass | 3,459 | 6,829 | +97% | 1 | 1 | 0% | 656 | 1,695 | +158% | 0 | 0 | — |
case-08 | pass→pass | 10,811 | 6,677 | -38% | 1 | 1 | 0% | 1,593 | 1,766 | +11% | 0 | 0 | — |
case-09 | pass→pass | 11,406 | 7,841 | -31% | 1 | 1 | 0% | 1,839 | 2,033 | +11% | 0 | 0 | — |
case-10 | pass→pass | 9,692 | 7,991 | -18% | 1 | 1 | 0% | 1,525 | 1,921 | +26% | 0 | 0 | — |
case-11 | pass→pass | 12,989 | 8,740 | -33% | 1 | 1 | 0% | 2,025 | 1,896 | -6% | 0 | 0 | — |
case-12 | pass→pass | 12,136 | 9,098 | -25% | 1 | 1 | 0% | 1,896 | 2,027 | +7% | 0 | 0 | — |
case-14 | pass→pass | 5,316 | 4,071 | -23% | 1 | 1 | 0% | 780 | 1,080 | +38% | 0 | 0 | — |
case-15 | fail→pass | 10,125 | 6,162 | -39% | 1 | 1 | 0% | 1,617 | 1,614 | -0% | 0 | 0 | — |
case-16 | pass→pass | 6,042 | 5,262 | -13% | 1 | 1 | 0% | 895 | 1,422 | +59% | 0 | 0 | — |
case-17 | fail→pass | 12,055 | 11,458 | -5% | 1 | 1 | 0% | 1,805 | 2,484 | +38% | 0 | 0 | — |
case-18 | fail→pass | 9,794 | 3,534 | -64% | 1 | 1 | 0% | 1,331 | 1,123 | -16% | 0 | 0 | — |
case-19 | fail→pass | 8,707 | 3,240 | -63% | 1 | 1 | 0% | 1,296 | 1,161 | -10% | 0 | 0 | — |
case-20 | pass→pass | 8,228 | 5,139 | -38% | 1 | 1 | 0% | 1,054 | 1,377 | +31% | 0 | 0 | — |
case-21 | fail→pass | 6,597 | 5,235 | -21% | 1 | 1 | 0% | 915 | 1,533 | +68% | 0 | 0 | — |
case-22 | fail→pass | 8,495 | 5,319 | -37% | 1 | 1 | 0% | 1,438 | 1,394 | -3% | 0 | 0 | — |
case-23 | fail→pass | 6,535 | 3,020 | -54% | 1 | 1 | 0% | 937 | 1,010 | +8% | 0 | 0 | — |
case-24 | pass→pass | 13,175 | 5,683 | -57% | 1 | 1 | 0% | 2,048 | 1,539 | -25% | 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. The headline lift of +42 percentage points is the difference between those two pass rates over the 24 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.