agentmods.dev

the-adversary: Skill for Claude Code

.claude/skills/adversary/SKILL.md

adversary is a skill for Claude Code from tasumermaf/the-adversary. It costs 118 tokens per session (1,831 once invoked), scanned A, original, MPL-2.0.

A three-stage adversarial audit process for checking a paper or code repository at a specific Git commit. It finds possible problems, independently verifies them, and produces a release decision queue without applying fixes automatically.

In plain words
What is it for?
Running release-gate or paper audits, reproducing findings, and preparing confirmed issues for human decisions.
Why use it?
It reduces the risk of shipping an unverified finding or changing the work before a person has decided what to do.

Skill for Claude Code

Written for Claude Code: installed under .claude/.

This is tasumermaf/the-adversary's own configuration. It tells Claude Code how to work on the-adversary itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything the-adversary configures →

not rated 2repo 1mo ago A scan 118 tokens original MPL-2.0
View source ↗ tasumermaf/the-adversary
Reuse

Borrowing it

Nothing to install: this file belongs to tasumermaf/the-adversary. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/tasumermaf/the-adversary/master/.claude/skills/adversary/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/tasumermaf/the-adversary

Made for: Claude Code.

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 adversary

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

Your own site · 80×15
<a href="https://agentmods.dev/skills/tasumermaf/the-adversary/adversary"><img src="https://agentmods.dev/badge/skills/tasumermaf/the-adversary/adversary.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 118 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,831 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.
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 $0.00118 $0.01831
Opus 5 $0.00059 $0.00915
Sonnet 5 $0.00024 $0.00366
Haiku 4.5 $0.00012 $0.00183

Measured 9d ago against content hash b351d60d2737, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-09, from the pricing page.

Security

Grade A, and why

adversary 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 9d 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.

.claude/skills/adversary/SKILL.md · 142 lines

How it starts

The opening of the file, as written. The whole thing — 142 lines — stays where its author put it; the contents beside it link to each section on GitHub.

The Adversary — orchestration hub

You are the hub. Your job is to run one audit end to end and hand the human a clean, verified, dispositionable verdict queue. You do not write the artifact and you do not fix it. Finding is cheap; verification is the product.

Hard rule, above everything else: the skill NEVER auto-fixes a CONFIRMED finding. Every fix waits for an explicit human verdict (FIXED / NOTED / DEFERRED / REJECTED-with-reason). You may draft a fix on request after a verdict, never before.

0. Pre-flight (stop if any fails)

  1. Repo clean. git status --porcelain is empty. A dirty tree means the commit SHA you pin will not describe what the lenses actually read. If dirty, stop and ask the human to commit or stash.

  2. Commit pinned. Capture git rev-parse HEAD. Every finding and every verification is stamped with this SHA (DESIGN R8). Record it.

  3. Profile + bars declared. Confirm the profile and its ex-ante FAIL bar before any lens runs (DESIGN R2). Do not discover the bar after seeing the findings.

    Profile Workflow FAIL bar (ex ante)
    release workflows/release-gate.js any blocker, or >= 2 confirmed majors
    paper workflows/loop-until-dry.js (default ladder) submission = 0 confirmed >= major in final round
    code workflows/loop-until-dry.js with a code ladder built from METHODOLOGY §7's lens mapping any blocker, or >= 2 confirmed majors
    protocol workflows/audit-round.js with lenses: ["provenance-prereg", "statistical-validity", "domain-correctness"] any blocker

    The code/protocol specialist rosters are partially shipped — METHODOLOGY §7 maps each documented slot to the nearest shipped lens; pass those lists explicitly. Both scripts refuse an empty lens set.

  4. Boundary spec (if the artifact is public-facing): confirm the IP boundary spec exists so ip-boundary can certify against something real, and pass its path as boundarySpecPath in the workflow args. A lens with no spec certifies against a boundary it reconstructed from the artifact — pass the path, or expect (and accept) the lens's missing-spec finding at major.

Read the full file on GitHub · 142 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. 9d ago First seen · 142 lines · 118 tokens per session scan A b351d60d2737

Subscribe to this mod's changes

adversary is a skill published in the GitHub repository tasumermaf/the-adversary (2 stars, last pushed 1mo ago), licensed MPL-2.0. It adds 118 tokens to every session and 1,831 once invoked, about $0.0006 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.

Code reviewadversarial-testingai-agentspaper-auditresearch-toolsscientific-writing
Related

Other skills, from other repositories

rhetorilex

Free local academic paraphrase and evidence-safe writing skill. Use for meaning-preserving paraphrase, Indonesian-friendly academic English, research framing, literature writing, theses, methods/results/discussion, reviewer responses, evidential verbs, claim control, and original pattern retrieval. Do not use to…

rezaprama/RhetoriLex · 85 tokens

food-chain-code

Adversarial architecture stress-tester. Selects attacker agents from a code-specific behavioral DNA library matched to the technical decision. Each attacks under strict role-lock. Weakest eliminated, survivor absorbs and evolves. Tests architecture decisions before a line of code is written. Works in Claude.ai, Claude…

CodedRichy/food-chain-ideation · 75 tokens

elegance-trap-probe

Strategy: Attack a beautiful unified result on the suspicion that its beauty is the bug. Distinguishes EARNED simplicity (forbids/predicts/subsumes) from DECORATIVE simplicity (re-describes/relabels/accommodates). Directly serves the Occam aesthetic by making it a falsifiable bar, not a vibe. Methods: Sober…

yogsoth-ai/stress-test · 103 tokens

debate-defender

Responds to attacks with counter-evidence and counter-arguments. Defends artifact using evidence, clarification, and rebuttal while acknowledging valid criticisms.

yogsoth-ai/stress-test · 34 tokens

perspective-critic

Evaluates artifact from a specific assigned perspective. Produces assessment grounded in that viewpoint's values, priorities, and expertise.

yogsoth-ai/stress-test · 29 tokens

adversarial-persona

Strategy: Role-play attacks from hostile personas — competing lab researcher, hostile reviewer, funding skeptic, domain outsider — each with distinct attack motivations and blind spots.

yogsoth-ai/stress-test · 37 tokens