Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Draft a structured legal brief, case summary, or legal argument outline. Use when asked to write a legal brief, case note, legal memo, argument outline, or position paper. Produces a structured document using IRAC format (Issue, Rule, Application, Conclusion).
.claude/skills/mohitagw15856-legal-brief/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 5% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-04 | ✗→✓ | ▲ Improved | 32% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -15% | 0% |
This skill drafts structured legal briefs and memos using IRAC format — the standard structure for legal writing.
One sentence per legal question:
One sentence per issue — conclusion upfront before analysis.
Concise relevant facts only. Flag disputed facts.
What this memo does not cover. What additional research would change the analysis.
WARNING: This draft requires review by a qualified legal professional. It does not constitute legal advice.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-03 | fail→pass | 20,602 | 18,536 | -10% | 1 | 1 | 0% | 3,992 | 4,378 | +10% | 0 | 0 | — |
case-02 | fail→pass | 21,003 | 18,647 | -11% | 1 | 1 | 0% | 3,805 | 3,982 | +5% | 0 | 0 | — |
case-01 | fail→pass | 33,079 | 14,548 | -56% | 1 | 1 | 0% | 6,233 | 3,750 | -40% | 0 | 0 | — |
case-04 | fail→pass | 14,818 | 15,187 | +2% | 1 | 1 | 0% | 2,631 | 3,466 | +32% | 0 | 0 | — |
case-05 | pass→pass | 16,761 | 16,086 | -4% | 1 | 1 | 0% | 2,981 | 3,738 | +25% | 0 | 0 | — |
case-06 | fail→pass | 24,541 | 18,931 | -23% | 1 | 1 | 0% | 4,502 | 3,848 | -15% | 0 | 0 | — |
case-07 | fail→pass | 25,696 | 23,910 | -7% | 1 | 1 | 0% | 3,884 | 4,697 | +21% | 0 | 0 | — |
case-08 | fail→pass | 17,939 | 17,994 | +0% | 1 | 1 | 0% | 2,827 | 3,762 | +33% | 0 | 0 | — |
case-09 | pass→pass | 24,687 | 17,900 | -27% | 1 | 1 | 0% | 4,112 | 3,247 | -21% | 0 | 0 | — |
case-10 | fail→pass | 20,388 | 17,493 | -14% | 1 | 1 | 0% | 3,164 | 3,608 | +14% | 0 | 0 | — |
case-11 | pass→pass | 31,662 | 19,403 | -39% | 1 | 1 | 0% | 5,359 | 4,190 | -22% | 0 | 0 | — |
case-12 | fail→pass | 14,527 | 14,788 | +2% | 1 | 1 | 0% | 2,509 | 3,088 | +23% | 0 | 0 | — |
case-22 | pass→fail | 17,422 | 19,568 | +12% | 1 | 1 | 0% | 2,585 | 3,637 | +41% | 0 | 0 | — |
case-13 | fail→pass | 23,271 | 17,195 | -26% | 1 | 1 | 0% | 3,907 | 3,548 | -9% | 0 | 0 | — |
case-14 | pass→pass | 18,269 | 16,835 | -8% | 1 | 1 | 0% | 3,083 | 3,705 | +20% | 0 | 0 | — |
case-15 | fail→fail | 10,345 | 16,674 | +61% | 1 | 1 | 0% | 1,698 | 3,287 | +94% | 0 | 0 | — |
case-16 | pass→pass | 17,976 | 20,700 | +15% | 1 | 1 | 0% | 3,011 | 4,132 | +37% | 0 | 0 | — |
case-17 | fail→pass | 19,202 | 15,592 | -19% | 1 | 1 | 0% | 3,184 | 3,366 | +6% | 0 | 0 | — |
case-18 | fail→fail | 14,660 | 16,583 | +13% | 1 | 1 | 0% | 2,536 | 3,561 | +40% | 0 | 0 | — |
case-19 | fail→pass | 18,514 | 18,649 | +1% | 1 | 1 | 0% | 3,324 | 4,123 | +24% | 0 | 0 | — |
case-20 | fail→fail | 2,005 | 6,546 | +226% | 1 | 1 | 0% | 353 | 1,641 | +365% | 0 | 0 | — |
case-21 | pass→pass | 20,106 | 19,235 | -4% | 1 | 1 | 0% | 3,239 | 4,008 | +24% | 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 +50 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is 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.