Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Review a document against playbook positions. Analyze clauses for risks, flag issues, suggest specific changes, and apply edits. Subsumes redlining — review + fix is one workflow.
.claude/skills/anylegal-ai-review/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-11 | ✗→✓ | ▲ Improved | 282% | 0% |
| case-14 | ✓→✗ | ▼ Worse | -31% | 0% |
| case-21 | ✓→✓ | = Same ✓ | 78% | 0% |
| case-22 | ✓→✓ | = Same ✓ | 177% | 0% |
| case-01 | ✗→✗ | = Same ✗ | -67% | 0% |
Use this skill when:
read_document — always read the full document firstread_document("Playbook/<filename>") to load the one(s) relevant to this contract type, jurisdiction, or clientweb_search with the jurisdiction set to find relevant statutes and regulations for the contract typeweb_fetch to retrieve full statutory text from authoritative sources (government domains, official gazettes)web_search for case law, regulatory guidance, and market standards (recent enforcement trends, standard market positions for the industry)web_fetchweb_search to check market standards if the clause deviates from typical positionsSkill(skill="docx-editing") for the full edit-pattern library, then call edit_document for each change. Edit the document in place — do not clone to a _v2 copy.old_text.edit_document emits <w:ins> / <w:del> tracked-change markup automatically. Author defaults to "Anylegal.ai" — pass an explicit author only if the user supplies a different name.edit_document for each one.run_code for text edits.Provide analysis as:
Skill(skill="docx-editing") and call edit_document for each change. Edit in place. Copy EXACT text from the document — never from memory.web_search and web_fetch against authoritative jurisdiction-specific sources to verify statutory requirements. Cite source URLs when flagging compliance issues.| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 29,363 | 9,135 | -69% | 1 | 1 | 0% | 5,324 | 1,740 | -67% | 0 | 0 | — |
case-02 | fail→fail | 10,250 | 3,303 | -68% | 1 | 1 | 0% | 1,494 | 1,349 | -10% | 0 | 0 | — |
case-03 | fail→fail | 25,174 | 8,097 | -68% | 1 | 1 | 0% | 3,714 | 1,523 | -59% | 0 | 0 | — |
case-04 | fail→fail | 4,645 | 8,441 | +82% | 1 | 1 | 0% | 700 | 1,687 | +141% | 0 | 0 | — |
case-05 | fail→fail | 4,161 | 8,426 | +102% | 1 | 1 | 0% | 209 | 1,783 | +753% | 0 | 0 | — |
case-06 | fail→fail | 2,338 | 10,069 | +331% | 1 | 1 | 0% | 381 | 2,007 | +427% | 0 | 0 | — |
case-07 | fail→fail | 19,958 | 11,832 | -41% | 1 | 1 | 0% | 3,529 | 1,862 | -47% | 0 | 0 | — |
case-08 | fail→fail | 8,397 | 4,641 | -45% | 1 | 1 | 0% | 1,237 | 1,291 | +4% | 0 | 0 | — |
case-09 | fail→fail | 11,251 | 7,156 | -36% | 1 | 1 | 0% | 1,893 | 1,356 | -28% | 0 | 0 | — |
case-10 | fail→fail | 13,459 | 5,754 | -57% | 1 | 1 | 0% | 2,599 | 1,357 | -48% | 0 | 0 | — |
case-11 | fail→pass | 4,236 | 7,248 | +71% | 1 | 1 | 0% | 676 | 2,581 | +282% | 0 | 0 | — |
case-12 | fail→fail | 2,042 | 6,085 | +198% | 1 | 1 | 0% | 362 | 1,496 | +313% | 0 | 0 | — |
case-13 | fail→fail | 2,619 | 6,778 | +159% | 1 | 1 | 0% | 391 | 1,489 | +281% | 0 | 0 | — |
case-14 | pass→fail | 11,102 | 5,809 | -48% | 1 | 1 | 0% | 1,940 | 1,338 | -31% | 0 | 0 | — |
case-15 | fail→fail | 8,732 | 17,549 | +101% | 1 | 1 | 0% | 1,311 | 1,338 | +2% | 0 | 0 | — |
case-16 | fail→fail | 2,121 | 6,963 | +228% | 1 | 1 | 0% | 278 | 1,438 | +417% | 0 | 0 | — |
case-17 | fail→fail | 13,415 | 7,817 | -42% | 1 | 1 | 0% | 1,971 | 1,547 | -22% | 0 | 0 | — |
case-18 | fail→fail | 5,778 | 5,549 | -4% | 1 | 1 | 0% | 1,014 | 1,624 | +60% | 0 | 0 | — |
case-19 | fail→fail | 9,229 | 11,405 | +24% | 1 | 1 | 0% | 1,538 | 1,576 | +2% | 0 | 0 | — |
case-20 | fail→fail | 31,121 | 18,078 | -42% | 1 | 1 | 0% | 6,168 | 2,734 | -56% | 0 | 0 | — |
case-21 | pass→pass | 9,272 | 11,039 | +19% | 1 | 1 | 0% | 1,566 | 2,783 | +78% | 0 | 0 | — |
case-22 | pass→pass | 7,839 | 16,054 | +105% | 1 | 1 | 0% | 1,472 | 4,074 | +177% | 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 3 counted toward the lift figure. The other 19 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 0 percentage points is the difference between those two pass rates over the 3 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.