Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, negative parallelisms, and excessive conjunctive phrases.
.claude/skills/majiayu000-humanizer/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -70% | 0% |
| case-13 | ✗→✓ | ▲ Improved | -10% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-19 | ✗→✓ | ▲ Improved | -7% | 0% |
| case-21 | ✗→✓ | ▲ Improved | -30% | 0% |
Use this skill to remove AI-writing artifacts while preserving the author's meaning, factual claims, and intended voice.
references/full-guide.md before executing detailed commands, applying templates, or relying on examples from this skill.references/full-guide.md for task-specific details.references/full-guide.md - complete command patterns, examples, checklists, and edge cases.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 9,105 | 19,161 | +110% | 1 | 1 | 0% | 1,549 | 2,211 | +43% | 0 | 0 | — |
case-02 | pass→pass | 6,259 | 4,739 | -24% | 1 | 1 | 0% | 960 | 1,044 | +9% | 0 | 0 | — |
case-03 | pass→pass | 15,144 | 14,222 | -6% | 1 | 1 | 0% | 2,632 | 2,381 | -10% | 0 | 0 | — |
case-04 | fail→fail | 3,973 | 12,802 | +222% | 1 | 1 | 0% | 498 | 2,542 | +410% | 0 | 0 | — |
case-05 | fail→pass | 9,629 | 1,505 | -84% | 1 | 1 | 0% | 1,599 | 476 | -70% | 0 | 0 | — |
case-06 | pass→pass | 8,999 | 4,704 | -48% | 1 | 1 | 0% | 1,298 | 871 | -33% | 0 | 0 | — |
case-07 | fail→fail | 5,543 | 3,964 | -28% | 1 | 1 | 0% | 820 | 876 | +7% | 0 | 0 | — |
case-08 | fail→fail | 8,370 | 3,769 | -55% | 1 | 1 | 0% | 1,304 | 812 | -38% | 0 | 0 | — |
case-09 | pass→fail | 11,137 | 3,218 | -71% | 1 | 1 | 0% | 1,705 | 678 | -60% | 0 | 0 | — |
case-10 | fail→fail | 5,557 | 5,227 | -6% | 1 | 1 | 0% | 833 | 979 | +18% | 0 | 0 | — |
case-11 | pass→pass | 6,651 | 5,777 | -13% | 1 | 1 | 0% | 1,021 | 1,158 | +13% | 0 | 0 | — |
case-12 | fail→fail | 5,084 | 4,906 | -4% | 1 | 1 | 0% | 898 | 1,011 | +13% | 0 | 0 | — |
case-13 | fail→pass | 6,266 | 3,050 | -51% | 1 | 1 | 0% | 788 | 707 | -10% | 0 | 0 | — |
case-14 | pass→pass | 9,541 | 5,390 | -44% | 1 | 1 | 0% | 1,412 | 972 | -31% | 0 | 0 | — |
case-15 | fail→pass | 4,958 | 3,959 | -20% | 1 | 1 | 0% | 690 | 873 | +27% | 0 | 0 | — |
case-16 | pass→pass | 4,094 | 3,418 | -17% | 1 | 1 | 0% | 587 | 760 | +29% | 0 | 0 | — |
case-17 | pass→pass | 9,990 | 3,358 | -66% | 1 | 1 | 0% | 1,479 | 685 | -54% | 0 | 0 | — |
case-18 | pass→pass | 5,493 | 3,243 | -41% | 1 | 1 | 0% | 845 | 673 | -20% | 0 | 0 | — |
case-19 | fail→pass | 5,361 | 3,235 | -40% | 1 | 1 | 0% | 765 | 710 | -7% | 0 | 0 | — |
case-20 | pass→fail | 6,496 | 4,758 | -27% | 1 | 1 | 0% | 1,050 | 466 | -56% | 0 | 0 | — |
case-21 | fail→pass | 6,708 | 3,465 | -48% | 1 | 1 | 0% | 980 | 684 | -30% | 0 | 0 | — |
case-22 | pass→fail | 6,527 | 2,224 | -66% | 1 | 1 | 0% | 957 | 531 | -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. 22 cases were attempted, and 21 counted toward the lift figure. The other 1 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 +9 percentage points is the difference between those two pass rates over the 21 comparable cases. 4 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.