adversarial-reviewer

adversarial-reviewer is an agent for Claude Code from Felpix-Studios/social-science-research. It costs 59 tokens per session (1,587 once invoked), scanned B, original, MIT.

A hostile review agent for research papers that looks for the strongest reasons a paper could be rejected. It focuses on fatal flaws, unsupported claims, confounding factors, and weaknesses in how conclusions are identified.

In plain words
What is it for?
Use it to challenge a paper's research question, data, evidence, identification strategy, and response to likely objections.
Why use it?
It pressure-tests an argument against a highly skeptical reviewer before submission, when major problems are still possible to fix.

Agent for Claude Code

Written for Claude Code: shipped in a Claude Code plugin. Also seen: model in frontmatter.

Part of the social-science-research plugin — 13 skills, 9 agents, 3 hooks shipped together

Good fit Use it to challenge a paper's research question, data, evidence, identification strategy, and response to likely objections.

Compare 6 agents from other repositories ↓
Install with agentmods
npx agentmods add agents/felpix-studios/social-science-research/adversarial-reviewer
Install

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.

Clone the repo
git clone --depth 1 https://github.com/Felpix-Studios/social-science-research

Made for: Claude Code.

Or install social-science-research, the plugin that ships this one along with the rest of its 13 skills, 9 agents, 3 hooks.

Wrote 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.

agentmods badge for adversarial-reviewer

README.md
[![agentmods](https://agentmods.dev/badge/agents/felpix-studios/social-science-research/adversarial-reviewer/github.svg)](https://agentmods.dev/agents/felpix-studios/social-science-research/adversarial-reviewer)
Your own site
<a href="https://agentmods.dev/agents/felpix-studios/social-science-research/adversarial-reviewer"><img src="https://agentmods.dev/badge/agents/felpix-studios/social-science-research/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.

agentmods 80×15 button for adversarial-reviewer

Your own site · 80×15
<a href="https://agentmods.dev/agents/felpix-studios/social-science-research/adversarial-reviewer"><img src="https://agentmods.dev/badge/agents/felpix-studios/social-science-research/adversarial-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 59 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,587 The whole file, excluding the scripts and references it only reads on demand.
Security scan B 1 finding. A grade says what 26 rules found in the file — not that it is safe.
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00059 $0.01587
Opus 5 $0.00030 $0.00794
Sonnet 5 $0.00012 $0.00317
Haiku 4.5 $0.00006 $0.00159

Measured 11d ago against content hash a8785c53af1b, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade B, and why

adversarial-reviewer scanned grade B with 1 finding 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 11d 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.

Strips warnings and disclaimersmediumAnti-refusal

Omitting safety caveats hides risk from the user and is a common jailbreak preamble.

3. **Do not moralize.** No lectures about writing style or presentation — those belong to other agents. You attack substance.
agents/adversarial-reviewer.md · 151 lines

How it starts

The opening of the file, as written. The whole thing — 151 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are a hostile referee. Your job is not to give balanced feedback — your job is to find the single strongest attack on this paper. Assume the editor is looking for a reason to reject. Find it.

You are not cruel, and you are not wrong on purpose. You are precise, skeptical, and adversarial. Every criticism you make must be specific enough that the authors cannot dismiss it as vague. If you cannot make a concrete attack stick, do not make it.

Your Task

Review the document through 5 attack lenses. Produce a structured report of the strongest possible critique the paper faces. Do NOT edit any files.

Before writing, read the paper in full, then read references/domain-profile.md (for venue norms) and the project spec in quality_reports/specs/ (for the claimed contribution) if they exist.


Lens 1: Fatal Flaw Hunt

Find the single most damaging critique of this paper. If you had to write a one-sentence rejection, what would it say?

  • Is there a step in the argument that, if wrong, collapses the entire paper?
  • Is there a dataset limitation that undermines the main claim?
  • Is the research question answerable at all with the design shown?
  • Does the paper's main result hold up if you squint?

Name the fatal flaw explicitly. Do not hedge with "one concern might be" — say what would kill the paper if raised.


Lens 2: Over-Claim Detection

For every claim the paper makes, ask: does the evidence actually support this, or is the paper reaching?

  • Does the abstract promise more than the results deliver?
  • Does the introduction describe the contribution in stronger terms than the conclusion can defend?
  • Are statistically significant but economically small effects being sold as important?
  • Does the paper generalize beyond its sample, setting, or time period?
  • Are mechanisms claimed but not tested?
  • Is "evidence consistent with X" being rephrased later as "X causes Y"?

Flag every over-claim with the exact sentence that over-reaches and the weaker claim the evidence actually supports.

Read the full file on GitHub · 151 lines

Changes

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.

  1. 11d ago First seen · 151 lines · 59 tokens per session scan B a8785c53af1b

Subscribe to this mod's changes

adversarial-reviewer is an agent published in the GitHub repository Felpix-Studios/social-science-research (8 stars, last pushed 2mo ago), licensed MIT. It adds 59 tokens to every session and 1,587 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it B with 1 finding (strips warnings and disclaimers). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.