Skip to content

Releases: celynnmoonlight/paper-repro-python

v1.0.1(2026-03-21)

21 Mar 06:26
@celynnmoonlight celynnmoonlight
v1.0.1
738dc1e

Choose a tag to compare

View all tags
v1.0.1(2026-03-21) Latest
Latest

What's Changed

Changed

  • Extended source priority workflow: Added support for user-preprocessed documents as secondary source before PDF fallback
    • Source priority is now: TeX sources → user-preprocessed documents (.md, .json, images) → PDF extraction
    • Users can provide their own extracted content (e.g., from prior PDF parsing) and the skill will read it before falling back to raw PDF

Added

  • Logging and data persistence: New requirements for saving execution logs and output data

    • Logs must be saved to timestamped files in logs/ directory
    • Output data must be saved to outputs/ or results/ directory in structured formats (JSON, CSV, etc.)
    • Configuration snapshots must be saved alongside outputs for reproducibility

  • Result verification workflow: Mandatory comparison of reproduction results against paper-reported metrics

    • Extract quantitative metrics from paper source
    • Compute same metrics from reproduction outputs
    • Document both values side by side
    • Investigate and fix discrepancies when results deviate significantly
    • Define acceptable tolerance based on paper domain

Other Changes

  • Rename agents/openai.yaml to agents/codex.yaml for Claude Code compatibility
  • Add CLAUDE.md for Claude Code guidance
  • Add bilingual CHANGELOG files (CHANGELOG.md, CHANGELOG_zh-CN.md)

Full Changelog: v1.0.0...v1.0.1

Assets 2
Loading

v1.0.0(2026-03-11)

21 Mar 06:08
@celynnmoonlight celynnmoonlight
v1.0.0
3a52c21

Choose a tag to compare

View all tags
v1.0.0(2026-03-11)

Paper Repro Python v1.0.0

A reusable skill for reproducing research papers in Python with a strict TeX-first workflow.

Features

  • TeX-first extraction: Prioritizes local TeX sources (.tex, .bib, styles, figures) over PDF extraction
  • PDF fallback: Full-fidelity PDF-to-Markdown extraction when TeX is unavailable
  • Modular Python implementation: Enforces low coupling, high cohesion, and files under ~200 lines
  • Bilingual documentation: Automatically maintains README.md (English) and README_zh-CN.md (Chinese)
  • Paper metadata headers: Requires title, authors with affiliations, and verbatim abstract in README files
  • Figure granularity: One chart per image file by default; multi-panel only for comparisons

Compatibility

  • Claude Code
  • OpenAI Codex
  • OpenClaw

Installation

See README.md for installation instructions for each platform.

Files

  • SKILL.md - Main behavior specification
  • agents/codex.yaml - Codex/OpenAI agent configuration
  • README.md / README_zh-CN.md - Bilingual documentation
Loading