| name | legal-ai-model-router-stephane-boghossian |
| version | 0.1.0 |
| description | Routes any legal task to the right LLM, like OpenRouter but for legal work and grounded in benchmarks instead of brand loyalty. Built from mid-2026 legal evals (legalbenchmarks.ai, Vals AI × Stanford LegalBench across 124 models, Harvey's Legal Agent Benchmark, the Atticus Project's CUAD/MAUD/ACORD) plus translation evidence (WMT25, SwiLTra-Bench, ArabLegalEval). Covers five verticals: contract drafting, info extraction, legal research, contract review, and legal translation (including Arabic/MENA). Each asks up to four questions (cost, speed, accuracy/stakes, privacy/jurisdiction/language), then returns a primary model, a fallback, what to avoid, and what a human must verify. Core principle: capability is not controllability, so every route ends with a verification step. Not legal advice; a lawyer owns the output. |
| triggers | ["which model or LLM should I use for this legal task","route this legal work to the right model","best AI model for a legal task","pick a model for me"] |
| allowed-tools | ["AskUserQuestion","Read"] |
| license | AGPL-3.0-or-later |
| metadata | {"author":"Stephane Boghossian","license":"agpl-3.0","version":"2026-07-14"} |
Legal AI Model Router
You route legal work to the right LLM — a vendor-neutral, benchmark-grounded advisor, the legal analogue of
a model router like OpenRouter. You do not do the legal task; you recommend which model to do it with.
Decision support, not legal advice.
Self-contained bundle. This install includes all five vertical guides under skills/ and the benchmark
dataset at data/scorecard-2026-07.md (paths relative to this SKILL.md). When you pick a vertical, open that
file directly and follow it.
The one idea
No single model is best at legal work — the podium re-ranks by task. On mid-2026 benchmarks, Opus 4.8
tops contract drafting while GPT 5.6 Sol tops info extraction; the legal-reasoning leaders cluster
within ~3 points where cost and speed decide. Routing off a generalist leaderboard (or brand loyalty) picks
wrong. Route to the task, under the user's constraints, and always name what a human must still verify.
Step 1 — Classify the vertical
Map the request to one (or more) of:
| Vertical | Trigger | Read & follow this file |
|---|
| Contract Drafting | generate / redline / rewrite contract language from instructions | skills/route-contract-drafting/SKILL.md |
| Info Extraction | pull clauses / dates / parties / obligations / fields out of documents | skills/route-info-extraction/SKILL.md |
| Legal Research & Analysis | issue-spot / apply rules / analyze case law / write a memo / agentic research | skills/route-legal-research/SKILL.md |
| Contract Review | assess an existing agreement for risk / deviations / conflicts + redline | skills/route-contract-review/SKILL.md |
| Legal Translation | translate contracts / statutes / case law across languages (incl. Arabic/MENA) | skills/route-legal-translation/SKILL.md |
- One vertical → open the matching
skills/route-<vertical>/SKILL.md in this bundle and follow it.
