This project addresses the complex and time-consuming process of generating research papers from initial ideas. It offers an automated pipeline, powered by Claude Code skills, to draft papers, plan and execute experiments, and create verifiable outputs like cited references and editable vector figures, all at a low API cost. The target audience includes researchers and engineers seeking to accelerate their publication workflow.

How It Works

The system utilizes a suite of 13 composable Claude Code skills orchestrated by a central ts-paper component. It follows a multi-stage pipeline: idea processing, planning, citation gathering (verified via WebSearch/Crossref), writing, refinement, adversarial review, figure generation, and LaTeX compilation. A key differentiator is the DrawAI hybrid engine, which reconstructs AI-generated raster figures into editable vector formats (SVG, PDF, PPTX) while maintaining pixel-exact fidelity. The pipeline emphasizes machine-checked integrity, ensuring all numbers are traceable to source data and citations are complete, with deterministic gates that halt the build on violations.

Quick Start & Requirements

• Primary Install: Recommended installation is as a Claude Code plugin:

  git clone https://github.com/Albus-White/spark-to-paper-skills.git ~/.claude/skills/spark-to-paper-skills
  

  This auto-loads the skills for the next Claude Code session.
• Prerequisites:
  • Claude Code environment.
  • Python 3.10+ with pip install -r skills/ts-figure-optimize/requirements.txt.
  • A LaTeX distribution (e.g., TeX Live) with latexmk.
• Resource Footprint: An optional DrawAI runtime requires approximately 4 GB of disk space and can run on CPU or GPU.
• Links: GitHub Repository

Highlighted Details

• Editable Vector Figures: Generates editable vector figures (SVG, PDF, PPTX) from raster inputs using the DrawAI engine, preserving 100% fidelity of the original render and making text overlays editable.
• Machine-Checked Integrity: Guarantees no fabricated numbers; citations are verified, and numbers are traced to source data. Builds fail on integrity violations.
• End-to-End Automation: Manages the entire research paper lifecycle from a one-line idea to a compiled PDF, including literature review, writing, and experimental validation.
• Template-Agnostic Venue Support: Easily add support for new publication venues by providing a templates/<name>/ directory with necessary LaTeX assets and a template.json file.
• Adversarial Review: Incorporates a multi-stage review process involving isolated reviewers and skeptics to identify and rectify issues.

Maintenance & Community

The project shows signs of active maintenance with recent updates (v1.1.0) and an automatic update notification system querying GitHub Releases. The README does not currently provide links to community channels like Discord or Slack, nor does it detail specific maintainers or sponsorships.

Licensing & Compatibility

The project is released under the MIT License, which is highly permissive and allows for commercial use, modification, and distribution, including within closed-source applications.

Limitations & Caveats

The primary dependency is Claude Code, which may represent an adoption barrier for users not already within that ecosystem. While the DrawAI engine aims for pixel-exact fidelity, AI-driven image reconstruction can introduce subtle artifacts or limitations with highly complex figures. The system operates on API calls, incurring costs (estimated ~$10 per paper), and while rigorous checks are in place, the robustness across all possible research domains and figure types is yet to be fully established.