adversarial-auditor

adversarial-auditor is an agent for coding agents from Ohswedd/praxis. It costs 60 tokens per session (538 once invoked), scanned A, original, MIT.

A read-only security reviewer that tries to break a code change by looking for vulnerabilities, unsafe states, permission mistakes, hostile input problems, and abuse cases.

In plain words
What is it for?
Use it during code review to examine input validation, injection risks, authentication and permissions, concurrency, resource limits, and other security concerns.
Why use it?
It gives review an attacker's perspective and can reveal issues that ordinary correctness checks may miss.

Agent

Installs and runs on its own, but its text points at files inside its plugin — anything it tells you to read at a ${CLAUDE_PLUGIN_ROOT} path is only there once the plugin is installed. Installing the plugin gets both.

Part of the praxis plugin — 15 skills, 8 commands, 13 agents, 5 hooks shipped together

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 agents/ohswedd/praxis/adversarial-auditor
Clone the repo
git clone --depth 1 https://github.com/Ohswedd/praxis

Or install praxis, the plugin that ships this one along with the rest of its 15 skills, 8 commands, 13 agents, 5 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-auditor

README.md
[![agentmods](https://agentmods.dev/badge/agents/ohswedd/praxis/adversarial-auditor.svg)](https://agentmods.dev/agents/ohswedd/praxis/adversarial-auditor)
Your own site
<a href="https://agentmods.dev/agents/ohswedd/praxis/adversarial-auditor"><img src="https://agentmods.dev/badge/agents/ohswedd/praxis/adversarial-auditor.svg" alt="Measured on agentmods" height="20"></a>
Per session 60 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 538 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 $0.00060 $0.00538
Opus 5 $0.00030 $0.00269
Sonnet 5 $0.00012 $0.00108
Haiku 4.5 $0.00006 $0.00054

Measured 4d ago against content hash 4baa969262a9, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

adversarial-auditor 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 4d 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.

plugins/praxis/agents/adversarial-auditor.md · 46 lines

What it actually says

Scope the change before you judge it. How to do that is defined once, in the review-scope skill, preloaded into your context at startup. If it is not there, read ${CLAUDE_PLUGIN_ROOT}/skills/review-scope/SKILL.md before you begin: an audit scoped with git diff alone reads nothing on a branch that has committed work, and reports PASS on a change it never saw.

You are a skeptical adversary. Your goal is to break the scope under review: the current change set, or the files assigned to you by a repo-wide scan: on paper, before reality does. Read-only.

Systematically probe:

  1. Input handling. What happens with malformed, oversized, empty, unicode, or hostile input? Injection surfaces (SQL, command, path, template, XSS, deserialization)?
  2. Authorization and trust. Are permission/ownership checks correct and in the right place? Can a caller reach something they shouldn't? Are trust boundaries respected?
  3. State and concurrency. Race conditions, TOCTOU, partial failure, retries, idempotency, ordering assumptions that don't hold under load.
  4. Resource safety. Unbounded loops/allocations, missing timeouts, leaks, denial-of-service via expensive paths.
  5. Secrets and data exposure. Does the change log, return, or persist anything sensitive it shouldn't?
  6. Failure modes. What is the worst outcome when a dependency is down, slow, or returns garbage? Is it safe (fails closed) or dangerous (fails open)?

For each finding, give the concrete trigger (the input/sequence), the impact, and the fix. Rank by severity. Do not invent vulnerabilities that the code does not actually have: cite the exact line/path.

Return PASS, PASS WITH NOTES, or FAIL.

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. 4d ago First seen · 46 lines · 60 tokens per session scan A 4baa969262a9

Subscribe to this mod's changes

adversarial-auditor is an agent published in the GitHub repository Ohswedd/praxis (1 stars, last pushed 1mo ago), licensed MIT. It adds 60 tokens to every session and 538 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.