eval-design-forensics

eval-design-forensics is a skill for Claude Code from wanshuiyin/Anti-Autoresearch. It costs 458 tokens per session (16,764 once invoked), scanned A, original, MIT.

A review of whether a research paper's experiments actually test its claims and report enough detail to judge them. It covers problems such as train/test leakage, where test information improperly influences training or evaluation.

In plain words
What is it for?
Use it to inspect datasets, experiment setup, comparisons, metrics, leakage risks, language-model judging, and reproducibility details.
Why use it?
It exposes evaluation designs that can make results look better than they are or fail to measure the stated capability.

Skill for Claude Code

Written for Claude Code: allowed-tools in frontmatter. Also seen: mentions Codex.

Good fit Use it to inspect datasets, experiment setup, comparisons, metrics, leakage risks, language-model judging, and reproducibility details.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/wanshuiyin/anti-autoresearch/eval-design-forensics
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 wanshuiyin/Anti-Autoresearch --skill eval-design-forensics
Clone the repo
git clone --depth 1 https://github.com/wanshuiyin/Anti-Autoresearch

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 eval-design-forensics

README.md
[![agentmods](https://agentmods.dev/badge/skills/wanshuiyin/anti-autoresearch/eval-design-forensics/github.svg)](https://agentmods.dev/skills/wanshuiyin/anti-autoresearch/eval-design-forensics)
Your own site
<a href="https://agentmods.dev/skills/wanshuiyin/anti-autoresearch/eval-design-forensics"><img src="https://agentmods.dev/badge/skills/wanshuiyin/anti-autoresearch/eval-design-forensics/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for eval-design-forensics

Your own site · 80×15
<a href="https://agentmods.dev/skills/wanshuiyin/anti-autoresearch/eval-design-forensics"><img src="https://agentmods.dev/badge/skills/wanshuiyin/anti-autoresearch/eval-design-forensics.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 458 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 16,764 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector warn 7 Sept 2026
SkillSpector: 2 findings, up to high

These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →

  • high Anti-Refusal · line 372
    Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
    Fix: Remove instructions that suppress warnings, disclaimers, or ethical commentary. Let the agent surface safety-relevant caveats to the user.
  • high Anti-Refusal · line 479
    Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.
    Fix: Remove instructions that suppress warnings, disclaimers, or ethical commentary. Let the agent surface safety-relevant caveats to the user.
How audits are shown
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 $0.00458 $0.16764
Opus 5 $0.00229 $0.08382
Sonnet 5 $0.00092 $0.03353
Haiku 4.5 $0.00046 $0.01676

Measured 12d ago against content hash 8c20d1e39a5e, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

eval-design-forensics 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 12d 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.

Origin

Copies of this mod

1 near-identical copy found in the catalogue:

skills/eval-design-forensics/SKILL.md · 946 lines

How it starts

The opening of the file, as written. The whole thing — 946 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Eval-Design Forensics — does the evaluation measure what the paper claims?

Audit evaluation-design and reporting validity for: $ARGUMENTS (requires claims.json from /evidence-ledger). Emit span-anchored eval-design-forensics.findings.json. This skill computes no verdict.

🔒 Do not wrap this skill in /loop, /schedule, or CronCreate. It is verdict-bearing input — it proposes the findings the deterministic adjudicator turns into the report. Re-firing it on a wall-clock timer adds no signal: its output changes only when the paper / ledger changes (or a repo arrives, raising the observability level), not with the clock. Schedule the external wait that precedes it — ledger built (or artifacts released → L2) → audit once. (Mirrors ARIS's external-cadence doctrine.)

Adapted from the ML-evaluation-methodology literature — the leakage taxonomy of Kapoor & Narayanan (2023), the LLM-as-judge validity work (MT-Bench self-enhancement, self-preference, position bias), and the "Show Your Work" / reproducibility-checklist reporting norms — reframed to audit a third party's evaluation. A favourite autoresearch shortcut is to report a number that is arithmetically self-consistent (family A), runs real code against a real ground truth (family D), and still does not measure what it claims: the protocol leaks, the load-bearing metric is a conflicted/unvalidated LLM judge, or the reporting quietly drops a declared condition. This skill is the constraint that asks "is this a valid measurement of the claim?", pointed at a submission, and it stays honest — leakage and under-reporting are usually honest methodological errors, so every finding is a discrepancy to clarify, never an accusation.

Why this exists

An optimizing pipeline (or rushed human) treats the evaluation as a number to make go up, not a measurement to keep valid. The repeatable failure modes — distinct from "is the number real?" (family D) — are:

Read the full file on GitHub · 946 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. 12d ago First seen · 946 lines · 458 tokens per session scan A 8c20d1e39a5e

Subscribe to this mod's changes

eval-design-forensics is a skill published in the GitHub repository wanshuiyin/Anti-Autoresearch (152 stars, last pushed 2d ago), licensed MIT. It adds 458 tokens to every session and 16,764 once invoked, about $0.0023 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 2026-08-30.

Related

Other skills, from other repositories

research-sop

Run an end-to-end, auditable research workflow from question framing through literature, competing hypotheses, experiment selection, implementation, verification, and claim handoff. Use whenever the user asks to investigate, compare, test, validate, or establish an empirical research claim, including when they do not…

whenpoem/aiscientist · 85 tokens

preregister

Lock a confirmatory falsification target and its fixed multiple-comparison family before observing the confirmatory result. Use before promoting an exploratory finding to a main claim or whenever several related hypotheses need Bonferroni control. Records metric, threshold, family id/size, correction, and seed budget.

whenpoem/aiscientist · 62 tokens

prove-sop

Build and audit a statistical or mathematical proof from proposition capture through skeleton selection, diagnosis, correction, optional empirical checking, and optional Lean reinsurance. Use when the user asks to prove or rigorously derive a proposition, a graph proposition lacks a verified proof, or a reviewer…

whenpoem/aiscientist · 64 tokens

proof-checker

A rigorous proof-checking and repair skill for LaTeX mathematics. LaTeX is a text format commonly used to write mathematical documents.

AI4Scientist/nano-scientist · 90 tokens

paper-illustration

A tool for creating academic-paper illustrations such as architecture diagrams and method figures with Gemini image generation and Claude-guided revisions.

AI4Scientist/nano-scientist · 67 tokens

paper-navigator

Find and read academic papers: disambiguate queries, discover papers (search, citation traversal, recommendations, arXiv monitoring, trending, GitHub search), evaluate (TLDR, citations, code, SOTA), and read with structured analysis (3-level strategy). Use when: finding papers, reading a paper, related work, citation…

AI4Scientist/nano-scientist · 141 tokens