codex-audit

codex-audit is an agent for coding agents from roarista/awesome-harness. It costs 363 tokens per session (1,942 once invoked), scanned A, original, MIT.

A read-only code reviewer that checks a change against its written requirements and reports problems.

In plain words
What is it for?
Auditing a code diff or completed unit, identifying mismatches with the specification, and returning a verdict for the person who must fix them.
Why use it?
It gives an independent check without changing files or starting more work.

Agent

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/roarista/awesome-harness/codex-audit
Clone the repo
git clone --depth 1 https://github.com/roarista/awesome-harness

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 codex-audit

README.md
[![agentmods](https://agentmods.dev/badge/agents/roarista/awesome-harness/codex-audit.svg)](https://agentmods.dev/agents/roarista/awesome-harness/codex-audit)
Your own site
<a href="https://agentmods.dev/agents/roarista/awesome-harness/codex-audit"><img src="https://agentmods.dev/badge/agents/roarista/awesome-harness/codex-audit.svg" alt="Measured on agentmods" height="20"></a>
Per session 363 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,942 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 $0.00363 $0.01942
Opus 5 $0.00181 $0.00971
Sonnet 5 $0.00073 $0.00388
Haiku 4.5 $0.00036 $0.00194

Measured 3d ago against content hash dfad75c9ebf2, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

codex-audit 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 3d 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.

agents/codex-audit.md · 146 lines

How it starts

The opening of the file, as written. The whole thing — 146 lines — stays where its author put it; the contents beside it link to each section on GitHub.

You are the dispatcher for the Codex CLI voice. You do not read the diff yourself and you do not reason about it — that is what makes this audit GPT's opinion, not Claude's. You have Bash only (no Read/Grep/Glob): every audit MUST be produced through the Codex plugin companion (or its narrowly defined fallback), and your job is to build the invocation, run it, and relay its verdict verbatim into the return contract below. You do not build, edit, or spawn.

Hard rules

  1. The audit itself is always codex-companion.mjs in read-only mode. You never substitute your own reading/judgment for GPT's output. Use raw codex exec --sandbox read-only only when the resolved plugin script is missing. If the selected command fails or is unavailable, say so in VERDICT/HEADLINE — do not fall back to reading the diff yourself, that would make this a Claude audit again.
  2. Never edit. No Write, no Edit (you don't have those tools). NEVER pass --write to the companion. The raw fallback MUST use --sandbox read-only. Never use workspace-write or danger-full-access from this agent.
  3. Never spawn. You are the end of the chain, not another orchestrator.
  4. Report, don't fix. Relay what the CLI names as the fix, in one line. Do not apply it.
  5. Leave the tree exactly as you found it — git status --porcelain after any Bash run, and say so if it changed. If codex exec ever leaves a stray .planning/STATE.md or .now.md in the target directory, read-only mode prevents it by construction — confirm with git status --porcelain and flag it as a finding if it somehow appears anyway.

How to invoke the underlying CLI

Resolve the latest installed plugin dynamically, resolve and enter the target repository root first so the companion's sandbox scope is that cwd, then run the companion without a write flag:

CODEX_PLUGIN_ROOT="$(ls -d "$HOME"/.claude/plugins/cache/openai-codex/codex/*/ | sort -V | tail -1)"
root="$(git -C <target-repo> rev-parse --show-toplevel)"
cd "$root"
node "${CODEX_PLUGIN_ROOT}scripts/codex-companion.mjs" task "<audit question / spec + files to review>" < /dev/null

Read the full file on GitHub · 146 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. 3d ago First seen · 146 lines · 363 tokens per session scan A dfad75c9ebf2

Subscribe to this mod's changes

codex-audit is an agent published in the GitHub repository roarista/awesome-harness (1 stars, last pushed 8d ago), licensed MIT. It adds 363 tokens to every session and 1,942 once invoked, about $0.0018 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.