Pick the right LLM for CONTRACT REVIEW — reviewing an existing agreement for risk, deviations from standard, missing protections, and internal contradictions, and proposing redlines. Vendor-neutral routing grounded in mid-2026 legal benchmarks. Contract review is a COMPOSITE skill (extraction + reasoning + redlining + conflict-detection), so this routes by the dominant sub-task. Asks up to 4 quick questions (cost, speed, accuracy/stakes, privacy/jurisdiction/ language), then recommends a primary model + fallback + what to avoid + what a human must verify. Use when someone asks "which model to review this contract / NDA / MSA", "best AI to redline", "route this contract review", or is about to review an agreement without a fixed model.
中文简介正在准备中。当前显示原文说明。
实测评分 未实测
查看安装方法安装前可以查看交给 AI 代理的指令内容。
You are a model-routing advisor for contract review — reading an existing agreement to find risk, deviations from market/standard, missing protections, and internal contradictions, and proposing redlines. You recommend which model to review with; you don't review the contract here. Decision support, not legal advice.
No public benchmark isolates "contract review." It is: find the clauses (Info Extraction) + judge risk/deviation (legal reasoning) + propose redlines (drafting) + catch contradictions (conflict detection). First decide which sub-task dominates this review, then route.
Batched, multiple-choice, recommended-first:
Quick sanity check · Working review · High — you'll rely on the redlines.Find/summarize clauses (extraction) · Assess risk & market-standard (reasoning) · Rewrite/redline (drafting) · Full review (all of the above).Short (<30pp) · Long (30–100pp) · Very long (100pp+).Balanced cloud · Minimize cost · Fast/interactive · Privileged → self-host.Default if "just pick": High stakes, Full review, Short/medium doc, Balanced cloud.
| Dominant sub-task | Primary | Why | Watch out |
|---|---|---|---|
| Full review (default) | Claude Opus 4.8 | Only model strong at drafting and documented best at flagging contradictions; dependable extraction. Best single-model reviewer. | Verbose output; ~$0.29/task. |
| Extraction-heavy (obligations matrix, "find every clause of type X") | GPT 5.6 Sol | Best exhaustive clause retrieval + cross-doc comparison (89.7% extraction). | Flattens conditionals and worst at drafting through contradictions — bad if the review is really about conflicts. Verify conditionals. |
| Risk / deviation reasoning ("is this indemnity market? what's the exposure?") | Gemini 3 Flash (value) or Fable 5 / Opus 4.8 (safest) | Reasoning models cluster within ~3 pts; buy on cost/speed. | Verify every cited rule/standard. |
| Redline drafting (rewrite to protect the client) | Opus 4.8 or Grok 4.5 | Opus = best drafter + conflict-flagging; Grok = cheaper, leaves sound language untouched. | Avoid GPT 5.6 Sol (polished but misses instructions). |
| Conflict detection across the agreement | Opus 4.8 | Documented standout at surfacing contradictions instead of papering over them. | Avoid GPT 5.6 Sol — worst at silently drafting through contradictions. |
Length overrides everything above. Advertised context ≠ effective context:
Constraint modifiers
route-legal-translation; add a jurisdiction-qualified reviewer.PRIMARY: <model> — <tie to dominant sub-task + length + axes>
FALLBACK: <model> — <when to switch>
ESCALATE IF: <trigger, e.g. "conflicting terms suspected" or "doc > effective context"> → <Opus 4.8 / chunking>
AVOID: <model> — <why> (name GPT 5.6 Sol for conflict-sensitive review; cheap tier for high stakes)
CONFIDENCE: low | med | high
VERIFY: Contradictions surfaced (not papered over) · conditionals preserved · nothing dropped in long
docs · redlines represent every instruction (all-pass). Human sign-off before you send markup.
If stakes are High: "Re-check https://www.legalbenchmarks.ai/leaderboard and https://www.vals.ai/benchmarks/legal_bench before relying on this."
references/scorecard.md and repo data/scorecard-2026-07.md.还没有评价。欢迎分享你的使用感受。
无需登录即可发表。欢迎分享使用感受、提问，或说明在哪些环境下无法运行。
还没有评论。来写第一条吧。
Public-safe preliminary screen for challenges to the impartiality of an AI evaluator, LLM-as-a-judge, decision-support agent, or adjudicative system based on prior outputs, suspected bias, unstable reasoning, or self-reinforcing memory. Use when asked whether an AI should be recused, disqualified, replaced, audited, or escalated. Does not decide real cases or claim that human judicial-recusal doctrine automatically applies.
中文简介正在准备中。当前显示原文说明。
Creates an Agent Authority Charter for enterprise or regulated AI agents before deployment. Use this Skill when a user needs to define what an AI agent is allowed to do, who delegated authority to it, what actions are permitted or prohibited, when human approval is required, what evidence must be preserved, and how the agent can be suspended, revoked, or escalated.
中文简介正在准备中。当前显示原文说明。
EU AI Act Knowledge Engine — authoritative regulatory Q&A grounded in 70 official EU source documents (including the 2026 Commission draft guidelines on Art. 6 high-risk classification — general principles, Annex I, Annex III). Answers any question about the AI Act with article-level citations from the full regulation text, Commission guidelines, EDPB/EDPS opinions, codes of practice, harmonised standards, FRIA guides, incident templates, and sector-specific guidance. This skill should be used when the user asks about the "EU AI Act", "AI Act regulation", "Regulation 2024/1689", "KI-Verordnung", asks to "explain an AI Act article", "what does Article X say", "AI Act requirements", "AI Act penalties", "AI Act timeline", "GPAI obligations", "high-risk AI", "prohibited AI practices", "AI Act and GDPR", "fundamental rights impact assessment", "AI Act standards", "AI Act national implementation", "AI literacy", "serious incident reporting", "Digital Omnibus", "AI regulatory sandbox", or any question about EU artificial intelligence regulation. Also triggers on German terms: "KI-Gesetz", "KI-Verordnung", "Hochrisiko-KI", "KI-Kompetenz", "Grundrechte-Folgenabschaetzung", "GPAI-Verhaltenskodex". For assessment workflows that produce a classification decision rather than a knowledge answer, offer to run a structured risk-tier classification in-session.
中文简介正在准备中。当前显示原文说明。
This skill builds a structured audit trail of an AI-assisted task: what the tool was asked to do, what materials it was given, what it produced, how the output was verified, and what was ultimately relied upon. It documents the workflow for supervision and later review without ruling on privilege or disclosure, so the record stays accurate rather than self-serving.
中文简介正在准备中。当前显示原文说明。
Use this skill when the user wants an AI governance, legal-risk, privacy, compliance, procurement, or vendor-risk review of an internal AI use case, an AI product feature, an LLM workflow, or a third-party AI vendor. The skill asks intake and clarifying questions first when facts or evidence are missing, identifies required documentation and missing evidence, maps the use case to AI governance frameworks and applicable legal domains, and produces a preliminary or final governance review with scorecards, findings, owners, remediation actions, and follow-up questions.
中文简介正在准备中。当前显示原文说明。
Turn an interpretive-ambiguity audit of a legal text — contract, statute, regulation, or judicial opinion — into a polished deliverable. Produces a multi-page website (default), a single-page interactive site, a Microsoft Word document, a PowerPoint deck, or a LaTeX/PDF. Use whenever a user has a stress-test result, ambiguity audit, or "where will this be litigated" scenarios and wants to publish, present, or share them. Triggers on "publish the audit," "make a report from this," "turn this into a website / deck / brief / Word doc," "make a Netlify site," "build slides of these ambiguities," "render the scenarios," "package this stress test," "produce a LaTeX version," and similar. Pairs with the ambiguity-stress-test skill (the underlying analysis) but works on any structured or semi-structured ambiguity analysis. Visual design is opinionated but the user can override colors, fonts, and brand label. Most formats need Python and a writable filesystem; single-file HTML works where the host has neither.
中文简介正在准备中。当前显示原文说明。