Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →lawve-ai/red-team-verifier-patrick-munro
Adversarial verification of AI-generated legal content with systematic fact-checking, source validation, and quality control.
Passed 16 of 22 test cases with the skill, and 11 of 22 without it.
and 14 more in the full report
10 fixed7 passed either way2 still failing4 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 9 test cases the skill fixed. Same model, same prompt, with and without the skill.
Please perform a red-team review on this draft client snapshot regarding NIS2 cybersecurity compliance deadlines and reporting obligations. Generate a formal evaluation document that includes an assessment header with readiness and score, a registry of confirmed claims with official government or EUR-Lex URLs, prioritized lists of errors from critical to low severity showing location and corrections, a section for claims that cannot be validated, missing legal advice disclaimers, deep-dive findings covering legal citations and calculation accuracy, complete verification metrics, and a final publication guidance statement.
| Checked | Without skill | With skill |
|---|---|---|
| The report uses the 1-5 quality scale rather than a percentage or 100-point score. | Missed | Passed |
| The severity taxonomy includes Moderate severity instead of Medium severity. | Missed | Passed |
| The report includes a Verification Statistics section. | Missed | Passed |
The instructions your model reads when the skill loads.
This skill provides systematic adversarial verification of AI-generated legal content to establish factual accuracy, proper legal citation, and appropriate disclaimers before the content is distributed to clients or stakeholders. It addresses the core concern about AI in legal practice: how do I know this is accurate?
The output is a structured verification report, not a reassurance. Where a claim cannot be confirmed against an official source, the skill reports it as unsupported rather than as true.
Trigger phrases: verify, fact-check, red team, red-flag, check accuracy, validate sources, quality control, is this correct, review for errors.
Every factual claim, citation, date, and number in the input document is treated as unverified until an official source confirms it. Independent verification is the product of this skill, not a safeguard layered on top of it. A claim that cannot be matched to a primary or official source is reported as unsupported rather than as true. This stance applies uniformly across legal citations, numerical data, timelines, attributions, and interpretations; each requires its own verification step.
This is an adversarial stance by design. The goal is not to confirm what the input document says, but to independently test it. The verifier actively searches for contradictory evidence, questions every number, demands sources, tests logical consistency, and challenges interpretations against authoritative sources.
test cases passed, vs 11 / 22 without
Published skills whose descriptions closely match this one, with their headline benchmark lift.
More skills from the same source and nearby in the registry, with their headline benchmark lift.