Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Research a legal topic with web search and inline citations. Use when the user needs legal information, market standards, regulatory guidance, or case law references.
.claude/skills/anylegal-ai-research/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-16 | ✗→✓ | ▲ Improved | 153% | 0% |
| case-07 | ✓→✗ | ▼ Worse | 0% | 0% |
| case-06 | ✓→✗ | ▼ Worse | -6% | 0% |
| case-10 | ✓→✗ | ▼ Worse | -45% | 0% |
| case-11 | ✓→✗ | ▼ Worse | 36% | 0% |
Use this skill when:
web_search with the jurisdiction set, targeting official government and institutional domains for statutory text, regulations, and case law. Frame queries with specific section/article numbers when known.web_fetch — once you have URLs to authoritative sources, fetch them to get complete article/section text. Don't rely on search snippets for citations.web_search for legal commentary, market standards, regulatory guidance, and practitioner analysis. Tag these as "commentary" rather than authority.[[N]](URL) format. Start numbering from 1] in every new response or document — never continue numbering from earlier messages.Structure research responses as:
Every factual claim MUST have an inline citation:
The Companies Act 2006 (UK) requires directors to act in accordance with the company's constitution [[1]](https://url-to-source).End every research response with:
## Sources
1. [Title of source](URL)
2. [Title of source](URL)When the user asks to create a document (memo, note, report) from research:
run_code (default Python) with python-docx to create the document with proper formatting (headings, bold defined terms, tables). Write the complete document in a single script. For simple markdown notes, create_document is acceptable.read_document to verify: correct language, citation numbering starts at 1], all sections present, formatting intact. Fix any issues before reporting success.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | pass→pass | 9,623 | 10,053 | +4% | 1 | 1 | 0% | 1,752 | 2,896 | +65% | 0 | 0 | — |
case-07 | pass→fail | 7,306 | 5,412 | -26% | 1 | 1 | 0% | 1,263 | 1,259 | -0% | 0 | 0 | — |
case-01 | fail→fail | 18,906 | 7,472 | -60% | 1 | 1 | 0% | 3,373 | 1,386 | -59% | 0 | 0 | — |
case-02 | fail→fail | 20,344 | 7,433 | -63% | 1 | 1 | 0% | 3,743 | 1,369 | -63% | 0 | 0 | — |
case-03 | fail→fail | 31,904 | 8,037 | -75% | 1 | 1 | 0% | 5,500 | 1,455 | -74% | 0 | 0 | — |
case-04 | pass→pass | 17,998 | 19,387 | +8% | 1 | 1 | 0% | 3,178 | 3,584 | +13% | 0 | 0 | — |
case-05 | fail→fail | 7,499 | 19,353 | +158% | 1 | 1 | 0% | 1,560 | 1,243 | -20% | 0 | 0 | — |
case-06 | pass→fail | 8,805 | 6,770 | -23% | 1 | 1 | 0% | 1,334 | 1,250 | -6% | 0 | 0 | — |
case-08 | fail→fail | 14,978 | 6,361 | -58% | 1 | 1 | 0% | 2,607 | 1,931 | -26% | 0 | 0 | — |
case-10 | pass→fail | 14,417 | 6,931 | -52% | 1 | 1 | 0% | 2,400 | 1,329 | -45% | 0 | 0 | — |
case-11 | pass→fail | 4,967 | 11,702 | +136% | 1 | 1 | 0% | 891 | 1,210 | +36% | 0 | 0 | — |
case-12 | pass→fail | 23,585 | 13,518 | -43% | 1 | 1 | 0% | 4,173 | 1,863 | -55% | 0 | 0 | — |
case-13 | fail→fail | 6,241 | 2,462 | -61% | 1 | 1 | 0% | 982 | 1,330 | +35% | 0 | 0 | — |
case-14 | pass→fail | 14,852 | 8,427 | -43% | 1 | 1 | 0% | 2,573 | 1,305 | -49% | 0 | 0 | — |
case-15 | pass→fail | 14,056 | 6,915 | -51% | 1 | 1 | 0% | 2,552 | 1,296 | -49% | 0 | 0 | — |
case-16 | fail→pass | 3,650 | 4,245 | +16% | 1 | 1 | 0% | 639 | 1,617 | +153% | 0 | 0 | — |
case-17 | pass→fail | 18,481 | 8,956 | -52% | 1 | 1 | 0% | 2,734 | 1,403 | -49% | 0 | 0 | — |
case-18 | fail→fail | 26,496 | 7,669 | -71% | 1 | 1 | 0% | 6,159 | 1,342 | -78% | 0 | 0 | — |
case-19 | fail→fail | 14,249 | 7,092 | -50% | 1 | 1 | 0% | 2,633 | 1,223 | -54% | 0 | 0 | — |
case-20 | pass→fail | 12,422 | 6,113 | -51% | 1 | 1 | 0% | 2,165 | 1,187 | -45% | 0 | 0 | — |
case-21 | pass→pass | 4,627 | 5,155 | +11% | 1 | 1 | 0% | 815 | 1,771 | +117% | 0 | 0 | — |
case-22 | pass→pass | 12,115 | 11,831 | -2% | 1 | 1 | 0% | 2,264 | 2,567 | +13% | 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 7 counted toward the lift figure. The other 15 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 -36 percentage points is the difference between those two pass rates over the 7 comparable cases. 10 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.