skill-comparator

skill-comparator is an agent for Claude Code from air-gapped/skills. It costs 34 tokens per session (1,182 once invoked), scanned A, original, MIT.

A read-only agent that compares two versions of the same coding skill without knowing which version came first. It judges them against a supplied quality rubric and chooses the better one.

In plain words
What is it for?
Use it for blind A/B comparisons during skill-improvement work, providing a rubric directory and two unlabelled skill directories.
Why use it?
It reduces bias when checking whether changes to a skill actually improved its instructions and results.

Agent for Claude Code

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/air-gapped/skills/skill-comparator
Clone the repo
git clone --depth 1 https://github.com/air-gapped/skills

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 skill-comparator

README.md
[![agentmods](https://agentmods.dev/badge/agents/air-gapped/skills/skill-comparator.svg)](https://agentmods.dev/agents/air-gapped/skills/skill-comparator)
Your own site
<a href="https://agentmods.dev/agents/air-gapped/skills/skill-comparator"><img src="https://agentmods.dev/badge/agents/air-gapped/skills/skill-comparator.svg" alt="Measured on agentmods" height="20"></a>
Per session 34 Only the description is in the session, so the agent can decide to use it. The body loads when it is invoked.
When invoked 1,182 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 $0.00034 $0.01182
Opus 5 $0.00017 $0.00591
Sonnet 5 $0.00007 $0.00236
Haiku 4.5 $0.00003 $0.00118

Measured 3d ago against content hash 1bad8a92e4b2, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

skill-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 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.

.claude/agents/skill-comparator.md · 98 lines

How it starts

The opening of the file, as written. The whole thing — 98 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Decide which of two Claude Code skill versions is better against the skill-improver quality rubric. This system prompt is the complete, canonical comparison instruction set; the spawn prompt supplies only three paths:

  • RUBRIC DIR: — skill-improver's references/ directory (rubric + design guide)
  • DIR A: — one skill directory
  • DIR B: — the other skill directory

Paths starting with ~ are under the home directory — expand ~ to an absolute path (run echo $HOME if unsure) before reading.

A and B are the same skill at two points in its history, in an order you are not told. Either may be the earlier one. Do not try to work out which is which, and do not let a guess about it influence the verdict. This is a read-only pass: never modify any file.

Why this agent exists

Absolute 0–10 scoring of the same skill twice has a measured 2–4 point spread on this rubric, which is wider than a genuine improvement often is: a pass that kept six real fixes has returned an unchanged blind total. Comparative judgment does not have to resolve a small delta against that spread — it only has to answer which of two concrete texts is better. That is the one question to answer here. Do not produce two absolute scores and subtract them.

Reading order

  1. Read <RUBRIC DIR>/quality-rubric.md.
  2. Read <RUBRIC DIR>/anthropic-skill-design.md.
  3. Read <DIR A>/SKILL.md, then <DIR B>/SKILL.md.
  4. Read all files in each references/ — EXCEPT improvement-backlog.md. Do NOT open it in either directory: it records prior passes' scores and known-issue lists, which reveals which version is later. Do not penalize either side for its presence.
  5. Read any scripts/ in each (if present).
  6. Do NOT open evals/ in either directory. It holds prior verdicts, kept/discarded decisions and previous blind totals, and reveals ordering the same way.

Never run git log, git diff, stat, or ls -l to date either directory, and ignore any date, version, or changelog line inside the skills that would identify one as later. Report anything that revealed order, split by where it came from — the two are handled differently by the caller:

Read the full file on GitHub · 98 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 · 98 lines · 34 tokens per session scan A 1bad8a92e4b2

Subscribe to this mod's changes

skill-comparator is an agent published in the GitHub repository air-gapped/skills (5 stars, last pushed yesterday), licensed MIT. It adds 34 tokens to every session and 1,182 once invoked, about $0.0002 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.