Legal Red Team

• You've used an LLM to draft a legal memo, contract summary, compliance guide, or regulatory overview and need to verify it before sharing.
• You want to catch plausible-but-wrong article numbers, fabricated deadlines, or confident-but-unsourced claims before distribution.
• You need a structured go/no-go quality gate (4/5 minimum score, zero CRITICAL issues) for AI-generated legal content.
• You're running a citation audit to confirm every statutory reference actually exists and matches the cited provision.

• You need to verify legal conclusions, litigation strategy, or professional legal interpretations — this skill checks factual and citation accuracy, not legal reasoning.
• You require access to paid legal databases (Westlaw, LexisNexis) — source verification relies on freely accessible official sources.
• You are looking for a substitute for qualified legal counsel on high-stakes matters; this skill is a supplement, not a replacement.

Every document is assessed across six structured categories: factual accuracy, legal authority citations, arithmetic validation, source verification, speculation detection, and disclaimer adequacy. Each category has defined red flags and checks, ensuring no common failure mode is missed.

The skill targets five recurring AI hallucination patterns in legal content: plausible-but-wrong article numbers, confidently incorrect dates, guidance mischaracterized as binding law, outdated legal references, and arithmetic errors in timeline calculations. Each pattern has a defined detection technique.

Findings are classified into four severity levels — CRITICAL, HIGH, MODERATE, and LOW — with clear definitions, examples, and required actions at each level. CRITICAL issues must be resolved before any distribution.

The legal_quality_scorer.py tool produces a composite 1–5 score with per-category breakdown. Documents scoring below 4/5 must not be distributed; the pre-distribution gate workflow enforces zero CRITICAL issues and complete disclaimers.

Two Python scripts — legal_fact_checker.py and legal_quality_scorer.py — support text or file input, JSON output, verbose mode, and saved reports. They serve as first-pass scanning tools, designed to feed into manual adversarial review.

Three ready-to-run workflows cover full adversarial review, quick citation checks, and pre-distribution gating. Each workflow includes a validation step to confirm completion criteria have been met.

Run legal_fact_checker.py to flag all citations and dates, verify each against EUR-Lex or eCFR, then run legal_quality_scorer.py to confirm a 4/5 or higher score and zero CRITICAL findings before sending to clients or staff.

Use Workflow 2 (Quick Citation Check) to extract every legal citation from the document and verify each against the relevant official source — catching invented article numbers or misquoted provisions before they reach a legal team.

Integrate legal_quality_scorer.py as an automated gate in a document generation pipeline. Any document scoring below 4/5 is held for human review; only documents meeting the threshold and disclaimer requirements are passed downstream.

Apply the six-step methodology with an adversarial mindset: mark every factual assertion and number, verify dates against statutory text, flag speculation presented as certainty, and produce a severity-classified findings report for the drafting team.