benchmark-archaeology

benchmark-archaeology is a skill for Claude Code, Codex from yogsoth-ai/de-anthropocentric-research-engine. It costs 48 tokens per session (1,228 once invoked), scanned A, original, Apache-2.0.

A research skill for critically examining AI and machine-learning benchmarks, their metrics, and leaderboards. A benchmark is a standard test set used to compare systems.

In plain words
What is it for?
Use it to audit benchmark quality, study score plateaus, test whether a benchmark measures its claimed ability, and investigate protocol differences.
Why use it?
It looks for misleading scores, outdated tests, data contamination, missing coverage, and differences in evaluation procedures.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one. Also seen: mentions subagents.

Good fit Use it to audit benchmark quality, study score plateaus, test whether a benchmark measures its claimed ability, and investigate protocol differences.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology
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 yogsoth-ai/de-anthropocentric-research-engine --skill benchmark-archaeology
Clone the repo
git clone --depth 1 https://github.com/yogsoth-ai/de-anthropocentric-research-engine

Made for: Claude Code, 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.

agentmods badge for benchmark-archaeology

README.md
[![agentmods](https://agentmods.dev/badge/skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology/github.svg)](https://agentmods.dev/skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology)
Your own site
<a href="https://agentmods.dev/skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology"><img src="https://agentmods.dev/badge/skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology/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 benchmark-archaeology

Your own site · 80×15
<a href="https://agentmods.dev/skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology"><img src="https://agentmods.dev/badge/skills/yogsoth-ai/de-anthropocentric-research-engine/benchmark-archaeology.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 48 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 1,228 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: 1 finding, 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 Prompt Injection · line 104
    Hidden instructions were detected in comments or invisible text. These could contain malicious directives. Manual review is recommended.
    Fix: Audit all comments and invisible characters. Remove any instructions that direct the agent to perform unauthorized actions. Use plain, reviewable content.
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 $0.00048 $0.01228
Opus 5 $0.00024 $0.00614
Sonnet 5 $0.00010 $0.00246
Haiku 4.5 $0.00005 $0.00123

Measured 8d ago against content hash 6e9f5d1fed03, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

benchmark-archaeology 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 8d 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/benchmark-archaeology/SKILL.md · 128 lines

How it starts

The opening of the file, as written. The whole thing — 128 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Benchmark Archaeology

Systematic excavation and critical analysis of AI/ML evaluation methodology. Treats benchmarks as historical artifacts requiring forensic examination — uncovering hidden assumptions, methodological drift, validity decay, and coverage gaps that accumulate over time.

Strategy Routing

Signal Route To
"audit this benchmark", "benchmark quality", "BetterBench" benchmark-audit
"saturation", "ceiling", "score plateau", "when will X be solved" saturation-analysis
"does it actually measure", "construct validity", "what does score mean" validity-probing
"what's not tested", "coverage gaps", "missing capabilities" coverage-mapping
"different papers get different scores", "protocol differences" protocol-forensics

Manifest

Strategies (5)

Strategy Purpose
benchmark-audit Systematic quality assessment using BetterBench 46-criterion framework
saturation-analysis Track score trajectories, detect saturation and failure points
validity-probing Challenge construct validity — does benchmark measure claimed capability?
coverage-mapping Map evaluation coverage, identify untested capability dimensions
protocol-forensics Analyze evaluation protocol differences across papers for same benchmark

Tactics (3)

Tactic Purpose
score-trajectory-analysis Collect historical scores, fit saturation curves, detect inflection points
artifact-detection Detect annotation artifacts and shortcuts in benchmarks
evaluation-protocol-comparison Compare implementation differences of same benchmark across papers

Subagent SOPs (9 + 1 shared)

SOP Purpose
benchmark-inventory Identify and catalog all relevant benchmarks in target domain
metric-decomposition Decompose composite metrics into constituent signals
contamination-audit Detect train-test data leakage and memorization artifacts
construct-validity-assessment Evaluate whether benchmark measures its claimed capability
documentation-audit Assess documentation completeness against BetterBench/Datasheets standards
capability-taxonomy-mapping Build capability taxonomy, map existing benchmark coverage
leaderboard-dynamics-analysis Analyze leaderboard score distributions, compression, selective reporting
protocol-element-extraction Extract evaluation protocol parameters from papers
benchmark-synthesis Produce final structured audit report
saturation-detection (shared) Detect saturation signals in score trajectories (from literature-survey)

Read the full file on GitHub · 128 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. 8d ago First seen · 128 lines · 48 tokens per session scan A 6e9f5d1fed03

Subscribe to this mod's changes

benchmark-archaeology is a skill published in the GitHub repository yogsoth-ai/de-anthropocentric-research-engine (462 stars, last pushed 2d ago), licensed Apache-2.0. It adds 48 tokens to every session and 1,228 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 2026-09-03.

Related

Other skills, from other repositories

Deep Research

Produce a deep, structured research report on a topic: decompose into key dimensions, analyze each with evidence and reasoning, synthesize cross-cutting insights, and surface open questions. Use for deep research, analysis, and literature/landscape reviews.

AgentEra/Agently · 53 tokens

literature-review-tools

Recommend AND run open-source AI tools, agents, Claude Code / Codex skills, and MCP servers for any stage of a literature review — searching, reading, extracting, synthesizing, screening, citation-checking, and paper writing. Use when the user asks "what tool should I use to..." OR "install/run/use to ..." for…

brycewang-stanford/lit-review-agent-tools · 288 tokens

papers-reading-skill

Evidence-grounded AI research workflow for turning supplied economics, finance, management, and social-science papers or structured records into versioned PaperReading artifacts. Use when Codex must ingest text, Markdown, or a text-based PDF; separate source-grounded claims from researcher analysis; bind findings to…

AOROM/paperreading · 115 tokens

paper-fulltext-harvest

Batch download academic paper full-text (PDF/XML) from a list of DOIs. Handles 25 DOI prefixes across 19 publisher families via three layered routes: (1) publisher TDM APIs requiring institutional subscription (Elsevier ScienceDirect, Wiley Online, Springer Nature), (2) Open Access sources (Crossref, Unpaywall…

jxtse/scientific-research-skills · 276 tokens

academic-figure-generation

Generates publication-quality academic figures (framework diagrams, pipeline illustrations, system architectures, method overviews) from a paper's method text and a target caption, using a local PaperBanana multi-agent pipeline (Retriever → Planner → Stylist → Visualizer → Critic).

jxtse/scientific-research-skills · 59 tokens

paper-reading

Reads and analyzes academic papers (arXiv preprints, conference / journal PDFs, Zotero items) at three configurable depths: quick skim (2 min), standard read (10 min), or deep analysis (30 min). Produces structured digests covering problem, method, key innovation, results, limitations, reproducibility, hidden…

jxtse/scientific-research-skills · 154 tokens