synthesis-adversarial-review

synthesis-adversarial-review is a skill for Claude Code, Codex from synthesisengineering/synthesis-skills. It costs 83 tokens per session (2,030 once invoked), scanned A, original, Apache-2.0.

A structured process for having separate agents challenge software work from different viewpoints. It records findings, handoffs, decisions, and whether the requested result is sufficient.

In plain words
What is it for?
Use it for bounded adversarial reviews, cross-agent checking, production handoffs, finding ledgers, and final acceptance decisions.
Why use it?
It can reveal mistakes or missing requirements that the main agent did not notice. The process keeps review focused on the requested outcome rather than producing reviews for their own sake.

Skill for Claude CodeCodex

Written for Claude Code and Codex: shipped in a Claude Code plugin, but also agents/openai.yaml present.

Part of the synthesis-skills plugin — 63 skills, 4 hooks shipped together

Good fit Use it for bounded adversarial reviews, cross-agent checking, production handoffs, finding ledgers, and final acceptance decisions.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review
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.

Any agent
npx skills add synthesisengineering/synthesis-skills --skill synthesis-adversarial-review
Clone the repo
git clone --depth 1 https://github.com/synthesisengineering/synthesis-skills

Made for: Claude Code, Codex.

Or install synthesis-skills, the plugin that ships this one along with the rest of its 63 skills, 4 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 synthesis-adversarial-review

README.md
[![agentmods](https://agentmods.dev/badge/skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review/github.svg)](https://agentmods.dev/skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review)
Your own site
<a href="https://agentmods.dev/skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review"><img src="https://agentmods.dev/badge/skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review/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 synthesis-adversarial-review

Your own site · 80×15
<a href="https://agentmods.dev/skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review"><img src="https://agentmods.dev/badge/skills/synthesisengineering/synthesis-skills/synthesis-adversarial-review.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 83 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,030 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
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 $0.00083 $0.02030
Opus 5 $0.00042 $0.01015
Sonnet 5 $0.00017 $0.00406
Haiku 4.5 $0.00008 $0.00203

Measured 11d ago against content hash 738cd39ceba4, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

synthesis-adversarial-review 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 11d ago.

The scan reads SKILL.md. This mod also ships 4 executable files (scripts/finding_ledger.py, scripts/protocol_acceptance.py, scripts/test_finding_ledger.py, …), listed below but not scanned — reading those needs a real analyzer, not pattern matching.

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.

skills/synthesis-adversarial-review/SKILL.md · 187 lines

How it starts

The opening of the file, as written. The whole thing — 187 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Synthesis Adversarial Review

Purpose

Adversarial collaboration is useful when differently shaped agents attack the same work from different blind spots. It is not an invitation to maximize rounds. The review exists to deliver the principal's outcome: the artifacts and enforced boundaries the principal asked to ship, at the accepted quality bar. Reviewer satisfaction, control growth, and a large finding count are not completion criteria.

This skill governs the review protocol. It does not grant publication, deployment, communication, or repair authority. Those approval boundaries survive the review.

Before Round One: Proportionality Contract

Record this section in the engagement plan before dispatching a reviewer:

  1. Principal outcome. State the outcome in the principal's terms, including the artifact or system boundary that must ship.
  2. Closed review universe. Enumerate the artifacts, surfaces, and decision planes. Each assigned plane receives a per-artifact terminal disposition in the same round.
  3. Consequence and depth. Name the harm the review is meant to prevent and the one verifier generation justified by that harm.
  4. Round-trip budget. Set a budget for principal courier crossings. Agent-to-agent transport is not a principal crossing; a required human copy/paste is. Declare, batch, and count every such crossing. Exceeding the budget is a blocked-state alert.
  5. Stop rule. Define green artifact acceptance, allowed open risks, approval gates, and the sufficiency checkpoint. Fewer rounds must come from complete coverage and stronger fixtures, never from fewer checks or lower quality.

If the universe cannot be enumerated, record why and define the bounded derivation that will close it. “Representative samples” do not support a closed-world completion claim.

Roles and Blind-Spot Rotation

Use at least two roles:

  • Executor: owns the principal's artifacts, production implementation, and repairs.
  • Adversarial reviewer: derives attacks independently, attempts to falsify the executor's claims, and does not inherit the executor's preferred abstraction.

Read the full file on GitHub · 187 lines

Files

What ships with it

6 files 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.

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 · 187 lines · 83 tokens per session scan A 738cd39ceba4

Subscribe to this mod's changes

synthesis-adversarial-review is a skill published in the GitHub repository synthesisengineering/synthesis-skills (18 stars, last pushed today), licensed Apache-2.0. It adds 83 tokens to every session and 2,030 once invoked, about $0.0004 per session on Opus 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.