A review gate that asks separate Claude and OpenAI-based critics to check a build, research result, refactor, or major fix against stated requirements before it is called complete.

Use it to verify medium or large code changes, research deliverables, frontend work, refactors, and other tasks with clear acceptance criteria.

It provides an independent check for missed bugs, unsupported claims, and unfinished work before a result is handed over.

verifier — Skill for Codex by JansenAnalytics

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 JansenAnalytics/claudex --skill verifier

Clone the repo

git clone --depth 1 https://github.com/JansenAnalytics/claudex

Made for: 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.

README.md

[![agentmods](https://agentmods.dev/badge/skills/jansenanalytics/claudex/verifier.svg)](https://agentmods.dev/skills/jansenanalytics/claudex/verifier)

Your own site

<a href="https://agentmods.dev/skills/jansenanalytics/claudex/verifier"><img src="https://agentmods.dev/badge/skills/jansenanalytics/claudex/verifier.svg" alt="Measured on agentmods" height="20"></a>

Per session 91 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.

When invoked 1,788 The whole file, excluding the scripts and references it only reads on demand.

Security scan B 1 finding. 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.

Model	Per session	Once invoked
Fable 5.1	$0.00091	$0.01788
Opus 5	$0.00046	$0.00894
Sonnet 5	$0.00018	$0.00358
Haiku 4.5	$0.00009	$0.00179

Measured 4d ago against content hash b91067840b48, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade B, and why

verifier scanned grade B with 1 finding 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.

Reads agent configuration directoriesmediumAgent snooping

.claude/, .codex/, .gemini/ hold keys, settings and other credentials a mod has no legitimate need for.

`~/.codex/config.toml` `azure` profile. Provider/model/backend are all config- or flag-selectable per task.

skills/verifier/SKILL.md · 110 lines

How it starts

The opening of the file, as written. The whole thing — 110 lines — stays where its author put it; the contents beside it link to each section on GitHub.

verifier — Double-Model Verification (Karpathy Layer 2)

Catches "done when it isn't", shallow/unverified claims, code bugs, frontend bugs, weak trade rigor, and output that falls short of the stated goal — by checking the work with two independent models and only clearing it when they agree. See design: research/2026-06-14-cross-model-verifier-spec.md.

When to run (in scope)

Medium/major builds, refactors, major fixes, research/analysis deliverables, repo work, frontend changes, trade-decision rigor — before you tell the user "done". Also any time on demand. Skip for small Q&A, trivial one-line edits, and anything where verification is overkill.

The flow (do all five — this is the gate)

1. State the goal + evaluation criteria UP FRONT. Before judging, write the concrete acceptance criteria ("what good looks like"). Precise, testable. This is the rubric both models grade against.

2. Layer A — independent Anthropic review. Spawn a FRESH Claude subagent (e.g. reviewer for code, researcher/analyst for research) — not your own builder context — and have it grade the artifact against the criteria, returning the verdict shape below. A fresh subagent reduces "builder grading its own homework."

3. Layer B — independent cross-model critic. Run the external engine (different model family → different blind spots). Always pass a --profile so the critic grades in context (this is what stops it over-flagging theoretical, unreachable issues — see "Context profiles & the gate" below):

export OPENAI_API_KEY="$(grep '^OPENAI_API_KEY=' ~/.claude-agent/.env | cut -d= -f2-)"
node ~/.claude-agent/scripts/verifier/verify.cjs \
  --profile curation-helper|trading|public-web|default \
  --goal "<the stated goal>" \
  --criteria "criterion 1; criterion 2; criterion 3" \
  --scope code|research|frontend|trade|general \
  --target <path | - | "inline text"> \
  --depth medium|major \
  [--invariants "must not drop data; must stay backward-compatible"] \
  [--non-goals "handling untrusted multi-MB input"]

--target - reads the artifact from stdin (e.g. git diff | node verify.cjs --target - ...).
Depth tiers: medium → api backend gpt-5.4; major → escalate (--depth major = gpt-5.5, or --backend codex --model gpt-5.3-codex for deep code review). Override per task: --model, --backend, --provider openai|azure-foundry.
Output is one structured verdict JSON (see schema). Cost is logged to data/verifier-spend.jsonl and capped at the configured monthly budget (config/verifier.json, default $200; refuses when exceeded).

Read the full file on GitHub · 110 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.

4d ago First seen · 110 lines · 91 tokens per session scan B b91067840b48

Subscribe to this mod's changes

verifier is a skill published in the GitHub repository JansenAnalytics/claudex (5 stars, last pushed 2mo ago), licensed MIT. It adds 91 tokens to every session and 1,788 once invoked, about $0.0005 per session on Opus 5. A static security scan graded it B with 1 finding (reads agent configuration directories). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-03.

Testing azure codex cross-model development karpathy-method openai quality-gate research review verification

dev-agent

Analyze repositories, implement code changes, run tests, and optionally commit.

frumu-ai/tandem · 17 tokens

qcsd-cicd-swarm

QCSD Verification phase swarm for CI/CD pipeline quality gates using regression analysis, flaky test detection, quality gate enforcement, and deployment readiness assessment. Consumes Development outputs (SHIP/CONDITIONAL/HOLD decisions, quality metrics) and produces signals for Production monitoring.

summarybotng/summarybot-ng · 61 tokens

verifier