paper-writing-bench

Convert AI research papers into sparse ideas, dense proposals, and experimental logs.

32|1|Updated May 19, 2026
One-click install
npx skills add https://github.com/woodfishhhh/EZ_math_model --skill paper-writing-bench-woodfishhhh
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: paper-writing-bench
Source: https://github.com/woodfishhhh/EZ_math_model/tree/main/skills/ez-math-model/external/paper-orchestra/skills/paper-writing-bench
Command: npx skills add https://github.com/woodfishhhh/EZ_math_model --skill paper-writing-bench-woodfishhhh

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

It helps you evaluate and improve paper-writing pipelines by converting an existing AI research paper into the exact benchmark-style raw ingredients (sparse idea, dense idea, and experimental log) needed for reconstruction.

Core Features & Use Cases

  • Paper-to-benchmark reconstruction: Reverse-engineers the (I, E) tuple by stripping narrative flow from a paper while preserving technical content constraints.
  • Sparse and Dense idea variants: Produces two concept-note forms—one high-level without math and one technical with LaTeX formulas and full variable definitions—both with no experimental content leakage.
  • Ground-truth experimental log: Extracts an exhaustive, past-tense experimental log with raw numeric data organized under dedicated headings to serve as the pipeline’s results source of truth.
  • Benchmark alignment: Mirrors the PaperWritingBench construction procedure (arXiv:2604.05018) and enforces strict anonymization, no citations, no URLs, and no figure/table references.

Quick Start

Use the paper-writing-bench skill to reverse-engineer a benchmark case from your provided paper PDF and output idea_sparse, idea_dense, and experimental_log files.

Frequently Asked Questions about paper-writing-bench

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I reverse-engineer benchmark inputs from an AI research paper?

You can reverse-engineer benchmark inputs by processing the paper PDF to output sparse idea, dense idea, and experimental log files. This strips narrative flow while preserving technical constraints to generate reconstruction tuples for evaluating paper-writing pipelines.

What is the difference between sparse and dense idea variants in benchmark construction?

Sparse idea variants are high-level concept notes without math or experimental content leakage. Dense idea variants provide technical proposals with LaTeX formulas and full variable definitions, ensuring both forms contain no experimental leakage for accurate benchmarking.

How do I extract numeric table data from a PDF for an experimental log?

Extracting numeric table data from a PDF for an experimental log requires accurate numeric table extraction to organize raw data under dedicated headings. The output serves as a ground-truth results source without figure or table references.

Does benchmark reconstruction require anonymization and removal of citations?

Yes, benchmark reconstruction requires strict anonymization and removal of all citations and URLs. This ensures the reverse-engineered benchmark inputs align with the PaperWritingBench construction procedure without leaking source identifiers.

When do I need to generate benchmark cases from existing papers?

You need to generate benchmark cases from existing papers when you want to evaluate paper-writing pipelines, compare text generations, or test paper-orchestration and rater workflows using ground-truth experimental data and technical proposals.

Can I use this benchmark extraction method to test paper-orchestration workflows?

Yes, you can use this benchmark extraction method to test paper-orchestration workflows. It reverse-engineers paper ingredients into sparse ideas, dense proposals, and experimental logs specifically designed to evaluate and compare generation pipelines.