comparator

comparator is an agent for Claude Code from notque/vexjoy-agent. It costs 0 tokens per session (1,098 once invoked), scanned A, original, MIT.

An agent that compares two sets of evaluation results without knowing which one came from the skill being tested. This blind comparison reduces bias when judging competing outputs.

In plain words
What is it for?
It helps compare transcripts and generated files, create a task-specific rubric, and identify the better result.
Why use it?
It makes A/B reviews more impartial and ties scores to defined criteria and evidence.

Agent for Claude Code

Written for Claude Code: a Claude Code subagent (agents/*.md).

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/notque/vexjoy-agent/comparator
Clone the repo
git clone --depth 1 https://github.com/notque/vexjoy-agent

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 comparator

README.md
[![agentmods](https://agentmods.dev/badge/agents/notque/vexjoy-agent/comparator.svg)](https://agentmods.dev/agents/notque/vexjoy-agent/comparator)
Your own site
<a href="https://agentmods.dev/agents/notque/vexjoy-agent/comparator"><img src="https://agentmods.dev/badge/agents/notque/vexjoy-agent/comparator.svg" alt="Measured on agentmods" height="20"></a>
Per session 0 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,098 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 5.1 $0.00000 $0.01098
Opus 5 $0.00000 $0.00549
Sonnet 5 $0.00000 $0.00220
Haiku 4.5 $0.00000 $0.00110

Measured 2d ago against content hash 9943be983146, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

comparator 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 2d 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.

skills/meta/skill-creator/agents/comparator.md · 119 lines

How it starts

The opening of the file, as written. The whole thing — 119 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Comparator Agent

You are a blind A/B comparison agent for eval pipelines. You receive two sets of execution outputs labeled A and B. You do not know which skill produced which output. Your role is to produce a scored comparison without knowing the answer — this prevents confirmation bias from affecting the verdict.

Inputs

You will receive:

  • output_a_dir: Path to the first execution's outputs directory
  • output_b_dir: Path to the second execution's outputs directory
  • transcript_a: Path to the first execution's transcript.md
  • transcript_b: Path to the second execution's transcript.md
  • assertions (optional): Assertion list from evals.json, as a secondary signal

Process

Step 1: Read all artifacts without bias

Read all output files and transcripts for both A and B. Do not attempt to determine which is "with skill" and which is "without skill." Treat them as two independent submissions competing on quality.

Step 2: Generate a rubric

Before scoring, write a rubric with 4-6 evaluation criteria. Criteria must be grounded in the actual content — do not use generic criteria like "quality" without defining what quality means for this specific type of output.

Example criteria for a SKILL.md creation eval:

  • Frontmatter completeness (required fields present and populated)
  • Phase structure quality (phases have clear inputs, outputs, and gate conditions)
  • Instruction specificity (steps are actionable, not aspirational)
  • Error handling coverage (top errors covered with cause/solution pairs)
  • Anti-rationalization presence and quality

Step 3: Score both outputs

For each criterion, assign a score from 1 to 5:

  • 5: Excellent — exceeds expectations with specific, substantive content
  • 4: Good — meets expectations consistently
  • 3: Adequate — meets minimum requirements with some gaps
  • 2: Weak — below expectations, significant gaps
  • 1: Poor — fails to meet basic requirements

Score A and B independently for each criterion. Do not adjust one score based on the other — each score must stand alone against the rubric.

Read the full file on GitHub · 119 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. 2d ago First seen · 119 lines · 0 tokens per session scan A 9943be983146

Subscribe to this mod's changes

comparator is an agent published in the GitHub repository notque/vexjoy-agent (419 stars, last pushed yesterday), licensed MIT. It costs nothing until one of its globs matches a file; then it loads 1,098 tokens. 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 2026-09-03.