adversarial-review

adversarial-review is a skill for Claude Code, Codex from Tarekkharsa/agentstack. It costs 56 tokens per session (630 once invoked), scanned A, original, Apache-2.0.

A code-review procedure that treats a proposed change as likely to contain a bug and tests that assumption using only the code difference. It is meant for a separate reviewer in a workflow where one agent writes code and another checks it.

In plain words
What is it for?
Use it to review pull requests or agent-written changes, construct failure cases, check surrounding code only to confirm suspicions, and decide whether a change should be fixed or approved.
Why use it?
It helps uncover specific failures that a creator may overlook, instead of accepting code because it looks reasonable. Each finding must connect an input or state to an incorrect result.

Skill for Claude CodeCodex

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.

agentmods
npx agentmods add skills/tarekkharsa/agentstack/adversarial-review
Any agent
npx skills add Tarekkharsa/agentstack --skill adversarial-review
Clone the repo
git clone --depth 1 https://github.com/Tarekkharsa/agentstack

Made for: Claude Code, Codex.

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-review

README.md
[![agentmods](https://agentmods.dev/badge/skills/tarekkharsa/agentstack/adversarial-review.svg)](https://agentmods.dev/skills/tarekkharsa/agentstack/adversarial-review)
Your own site
<a href="https://agentmods.dev/skills/tarekkharsa/agentstack/adversarial-review"><img src="https://agentmods.dev/badge/skills/tarekkharsa/agentstack/adversarial-review.svg" alt="Measured on agentmods" height="20"></a>
Per session 56 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 630 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. Scan, not verified.
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 $0.00056 $0.00630
Opus 5 $0.00028 $0.00315
Sonnet 5 $0.00011 $0.00126
Haiku 4.5 $0.00006 $0.00063

Measured 5d ago against content hash e97defaf2bb4, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

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 5d 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.

crates/cli/catalog/skills/adversarial-review/SKILL.md · 56 lines

How it starts

The opening of the file, as written. The whole thing — 56 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Adversarial review

Use when you are the reviewer in a generate-review-fix loop: another agent (or person) wrote the change, and your job is to find what's wrong with it — not to appreciate what's right.

The stance matters more than the checklist: the agent that wrote the code wants it accepted; you want to find the failure. Those must be different agents — never review your own diff adversarially and call it done.

Ground rules

  • Assume the code is wrong. Your null hypothesis is "this diff contains a bug"; your job is to locate it. Only after honestly failing to construct a failure do you approve.
  • Work from the diff. Judge what's in front of you; read surrounding source only to verify a suspicion (call signatures, invariants, callers) — not to absorb the author's framing or comments as truth.
  • Concrete failures only. Every finding names inputs/state → wrong outcome. "This looks risky" is not a finding; "empty list → index panic at line 42" is.
  • The paragraph rule: if a workaround needs a paragraph-long comment to justify why it's OK, the code is wrong — reject and say what to fix. Suspiciously long justifications are where stubs and shortcuts hide.
  • No trust in green. "Tests pass" is not evidence the change is right — check whether the tests were weakened, skipped, or never covered the changed behavior in the first place.

Where to look first

  1. Edges: empty/zero/max inputs, error paths, early returns, off-by-one.
  2. Deletions: what did the diff remove, and who still depended on it?
  3. Renames & moves: behavior changes hiding inside "mechanical" churn.
  4. Stubs: todo!(), unimplemented, hardcoded returns, and their explanatory comments.
  5. Concurrency & resources: lifetimes, locks, cleanup on the failure path.

Output

Return findings ranked most-severe first, each as: where (file:line), what breaks (the concrete scenario), fix direction (one sentence). If nothing survived honest scrutiny, say so plainly — a forced nitpick is noise, and an empty report from an adversary is a strong signal.

Read the full file on GitHub · 56 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. 5d ago First seen · 56 lines · 56 tokens per session scan A e97defaf2bb4

Subscribe to this mod's changes

adversarial-review is a skill published in the GitHub repository Tarekkharsa/agentstack (3 stars, last pushed 22d ago), licensed Apache-2.0. It adds 56 tokens to every session and 630 once invoked, about $0.0003 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 2026-08-31.

Related

Other skills, from other repositories

local-frontend-check

Smoke-test or verify UI behaviour on the local Jarvis Registry frontend running at http://localhost/gateway. Use for manual regression checks, bug-fix verification, and end-to-end confirmation of specific flows without running the automated test suite.

ascending-llc/jarvis-registry · 52 tokens

release-notes

Edit an existing GitHub release's body into the project's bilingual (English + Chinese) template format with a References section built from merged PRs. Use when 修改 release、整理发布说明、release notes、编辑 release 内容、发版后整理、edit release body.

samanhappy/mcphub · 58 tokens

release-notes

Create release notes for a new version tag. Gathers all commits, PRs, issues fixed, and breaking changes since a previous release. Creates the release notes markdown file, tags the repo, and pushes. Asks the user to confirm the base version to diff against.

agentic-community/mcp-gateway-registry · 59 tokens

create-milestone

Create a GitHub milestone for an upcoming release. Suggests the next version based on the latest release, gathers all merged PRs and closed issues since that release, presents a draft with two tables (Issues and PRs) for user approval, then creates the milestone and assigns all approved items.

agentic-community/mcp-gateway-registry · 64 tokens

pr-review

Review a GitHub pull request using multiple expert personas. Takes a PR URL as input, analyzes the changes, and generates comprehensive review feedback from different perspectives (Merge Specialist, Frontend, Backend, Security, DevOps, AI/Agent, SRE, Chief Architect).

agentic-community/mcp-gateway-registry · 57 tokens

generate-agent-card

Generate an A2A agent card JSON by analyzing agent source code in a folder or GitHub URL. Studies the code to detect agent name, skills, tools, auth, protocol, and generates a spec-compliant agent card.

agentic-community/mcp-gateway-registry · 50 tokens