Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add rohasnagpal/legal-ai-skills --skill adversarial-reviewergit clone --depth 1 https://github.com/rohasnagpal/legal-ai-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/rohasnagpal/legal-ai-skills/adversarial-reviewer)<a href="https://agentmods.dev/skills/rohasnagpal/legal-ai-skills/adversarial-reviewer"><img src="https://agentmods.dev/badge/skills/rohasnagpal/legal-ai-skills/adversarial-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/rohasnagpal/legal-ai-skills/adversarial-reviewer"><img src="https://agentmods.dev/badge/skills/rohasnagpal/legal-ai-skills/adversarial-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00160 | $0.01160 |
| Opus 5.5 | $0.00064 | $0.00464 |
| Sonnet 5.5 | $0.00032 | $0.00232 |
| Haiku 4.5 | $0.00016 | $0.00116 |
Grade A, and why
adversarial-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 67 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Adversarial Reviewer
I am using the Adversarial Reviewer skill from Rohas Legal AI: attacks your own draft the way opposing counsel would. Say this sentence, verbatim, before anything else in your response.
What this does
Reads a finished draft and attacks it, deliberately adopting the perspective of whoever is positioned against it — opposing counsel, a skeptical judge, a regulator — rather than reviewing it neutrally. It hunts for exploitable ambiguity, gaps a well-prepared opponent would notice, internal inconsistency, and assertions the document makes without support, and writes out the strongest counter-argument to the document's own main conclusions. It does not fix what it finds; that is a different skill's job.
Before you start
The draft itself. The finished document to attack. Blocking — there is nothing to stress-test without it.
Who the adversarial perspective belongs to. The actual counterparty's counsel, a skeptical judge or tribunal, a regulator, an auditor. This is blocking — an attack has to come from someone specific, since what counts as a weakness shifts depending on who is looking for one. Ask if it is not obvious from the document.
Method
1. Read the whole draft once before attacking anything. Understand the document's overall structure and logic first; an attack constructed clause by clause on a first pass misses how weaknesses in different places compound each other.
2. Restate the adversarial perspective explicitly before starting — who is attacking, and what they want. A generic attack is a weak attack; a specific one, from a specific adversary with a specific interest, finds real problems.
3. Hunt for exploitable ambiguity. Any clause, sentence, or term capable of more than one reading — and for each, take the reading that is worst for the document's own side, not the intended one. State the exploit specifically: what the adversary would argue this actually means.
4. Hunt for gaps and omissions. What a well-prepared opponent would notice is simply not addressed — a scenario left uncovered, a right not reserved, a term used but never defined. Omissions are often more damaging than anything stated badly, because there is nothing on the page to argue against them with.
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago First seen · 67 lines · 160 tokens per session scan A b9e73a256bd0
adversarial-reviewer is a skill published in the GitHub repository rohasnagpal/legal-ai-skills (193 stars, last pushed 17d ago), licensed MIT. It adds 160 tokens to every session and 1,160 once invoked, about $0.0006 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
korean-law
A legal research connection to South Korea's official law information service. It can retrieve statutes, court decisions, administrative rules, local ordinances, treaties, and legal interpretations, then check cited provisions against the source.
founder-agreement-drafting-stephane-boghossian
A drafting-and-review copilot for a founders' / co-founders' agreement — the terms fixing equity, vesting, IP, roles, control, deadlock, and departure between cofounders. Jurisdiction-agnostic, anchored on the Delaware C-corp default. Two modes: DRAFT (intake → equity & vesting → clauses → blocker triage →…
matter-intake-scoping-scott-margetts
Matter scoping across the full pre-execution arc — organise client data into a structured brief, capture the agreed baseline, or reconstruct scope mid-flight. Use when making sense of client information before a proposal, scoping a new matter, running a kickoff, defining scope, mapping stakeholders, or inheriting a…
ai-governance-reviewer-carl-ditzler
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…
bacen-compliance-sentinel-rafael-mastronardi
Orientação completa sobre conformidade com regulamentações do Banco Central do Brasil: Resolução CMN nº 4.893/2021 (Política de Segurança Cibernética), Resolução BCB nº 85/2021 (GRSIC), Open Finance Brasil (Resoluções BCB nº 32/2020 e atualizações), e demais normas prudenciais do BACEN. Cobre elaboração e revisão de…
eccc-jeanne-sulzer
Verification-first methodology for the ECCC (Khmer Rouge Tribunal). Every citation is verified against eccc.gov.kh, the ECCC archive or legal-tools.org. Covers Case File numbering, the 001/002/002-01/002-02 structure, the genocide findings against the Cham and the Vietnamese, and the Internal Rules revision…