Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Compare two document versions and produce analysis or a Word redline. Use when the user has two versions of a document and wants to understand what changed.
.claude/skills/anylegal-ai-compare/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 173% | 0% |
| case-21 | ✓→✗ | ▼ Worse | -25% | 0% |
| case-03 | ✓→✓ | ＝ Same ✓ | -1% | 0% |
| case-07 | ✓→✓ | ＝ Same ✓ | 6% | 0% |
Use this skill when:
read_documentcompare with both document paths or texts — this returns a structured diff, similarity percentage, and visual outputrun_code (default Python) to generate a DOCX with tracked changes. Open the original with python-docx, compute a diff against the revised text, and insert w:ins/w:del tracked-change markup via lxml. See /docx-editing skill for OOXML tracked change patterns. Verify the output with structural checks.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 15,904 | 8,645 | -46% | 1 | 1 | 0% | 2,954 | 1,361 | -54% | 0 | 0 | — |
case-02 | fail→pass | 21,241 | 22,902 | +8% | 1 | 1 | 0% | 3,650 | 4,840 | +33% | 0 | 0 | — |
case-03 | pass→pass | 10,415 | 7,819 | -25% | 1 | 1 | 0% | 1,932 | 1,914 | -1% | 0 | 0 | — |
case-04 | fail→fail | 8,293 | 5,400 | -35% | 1 | 1 | 0% | 1,317 | 868 | -34% | 0 | 0 | — |
case-05 | fail→fail | 3,128 | 23,606 | +655% | 1 | 1 | 0% | 474 | 3,691 | +679% | 0 | 0 | — |
case-06 | fail→fail | 4,469 | 28,647 | +541% | 1 | 1 | 0% | 747 | 4,057 | +443% | 0 | 0 | — |
case-07 | pass→pass | 8,730 | 6,415 | -27% | 1 | 1 | 0% | 1,531 | 1,627 | +6% | 0 | 0 | — |
case-08 | fail→fail | 9,598 | 13,189 | +37% | 1 | 1 | 0% | 1,560 | 1,571 | +1% | 0 | 0 | — |
case-09 | fail→fail | 9,046 | 11,570 | +28% | 1 | 1 | 0% | 1,672 | 1,180 | -29% | 0 | 0 | — |
case-10 | fail→fail | 3,030 | 17,633 | +482% | 1 | 1 | 0% | 504 | 2,527 | +401% | 0 | 0 | — |
case-11 | fail→fail | 7,212 | 8,705 | +21% | 1 | 1 | 0% | 1,124 | 938 | -17% | 0 | 0 | — |
case-12 | pass→pass | 8,486 | 4,077 | -52% | 1 | 1 | 0% | 1,567 | 1,230 | -22% | 0 | 0 | — |
case-13 | fail→fail | 11,455 | 4,660 | -59% | 1 | 1 | 0% | 1,831 | 1,329 | -27% | 0 | 0 | — |
case-14 | pass→pass | 8,024 | 2,712 | -66% | 1 | 1 | 0% | 1,426 | 932 | -35% | 0 | 0 | — |
case-15 | fail→fail | 4,839 | 2,675 | -45% | 1 | 1 | 0% | 638 | 894 | +40% | 0 | 0 | — |
case-16 | fail→fail | 8,247 | 3,135 | -62% | 1 | 1 | 0% | 1,441 | 844 | -41% | 0 | 0 | — |
case-17 | pass→pass | 8,229 | 4,216 | -49% | 1 | 1 | 0% | 1,341 | 1,191 | -11% | 0 | 0 | — |
case-18 | pass→pass | 8,904 | 5,940 | -33% | 1 | 1 | 0% | 1,755 | 1,582 | -10% | 0 | 0 | — |
case-19 | fail→pass | 6,518 | 15,263 | +134% | 1 | 1 | 0% | 1,125 | 3,073 | +173% | 0 | 0 | — |
case-20 | fail→fail | 2,039 | 37,737 | +1751% | 1 | 1 | 0% | 261 | 6,516 | +2397% | 0 | 0 | — |
case-21 | pass→fail | 8,408 | 10,119 | +20% | 1 | 1 | 0% | 1,573 | 1,182 | -25% | 0 | 0 | — |
case-22 | fail→fail | 10,560 | 8,286 | -22% | 1 | 1 | 0% | 1,163 | 980 | -16% | 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 15 counted toward the lift figure. The other 7 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 +5 percentage points is the difference between those two pass rates over the 15 comparable cases. 3 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.