Spark-To-Paper-Skills/spark-to-paper-skills

One sentence in, one draft paper out: spark-to-paper-skills automatically reviews papers, plans and runs experiments, and writes a research draft—typically using about $10 in API costs per paper.

What it solves

Spark-to-Paper automates the end-to-end process of turning a research idea or proposal into a fully compiled academic paper. It addresses the common pain points of academic writing, such as managing real citations, creating high-quality editable vector figures, and ensuring the integrity of data and claims without fabricating results.

How it works

The system operates as a plugin for Claude Code, consisting of one orchestrator (ts-paper) and 13 composable skills. It follows a 7-stage pipeline (plus an optional experiment stage) that routes input based on whether it is a bare idea, a proposal, or a proposal with real data.

Key technical components include:

  • The Figure Engine: Uses a hybrid approach where candidates are rendered via PaperBanana, then redrawn natively as audited SVGs to ensure they are editable and high-resolution.
  • Integrity Gates: Deterministic Python scripts (like run_gates.py and audit_svg.py) that check for citation completeness, fabrication of numbers, and vector structure.
  • Adversarial Review: A process where isolated reviewers critique the paper and skeptics attempt to refute issues until the draft is hardened.
  • Template-Agnostic Assembly: Uses LaTeX to compile the final PDF, supporting various venue styles (e.g., NeurIPS) via simple configuration files.

Who it’s for

Researchers and academics who want to accelerate the drafting process of their papers while maintaining strict machine-checked integrity and professional vector graphics.

Highlights

  • Editable Vector Figures: Redraws AI-generated images into native SVGs with live text and verified geometry.
  • Machine-Checked Integrity: Prevents the fabrication of results; in proposal mode, result cells remain blank rather than invented.
  • End-to-End Workflow: Handles everything from literature search (via WebSearch and Crossref) to experiment execution and PDF compilation.
  • Template-Agnostic: Supports any conference format by adding a template directory with LaTeX assets.
  • Claude Code Integration: Operates as a plugin with zero infrastructure requirements (no separate server or database).

Related

  • Project
    luwill/research-skills

    A collection of Claude Code skills designed to automate academic research workflows, including exhaustive literature retrieval, verified proposal writing, and high-fidelity slide generation.

  • Project
    alaliqing/claude-paper

    A research-paper learning plugin for AI agents that transforms PDFs into interactive study environments with automated summaries, code demonstrations, and a web viewer.

  • Project
    K-Dense-AI/claude-scientific-writer

    An AI-powered research and writing tool that generates publication-ready scientific papers, grant proposals, and reports backed by real-time literature search and verified citations.

  • Project
    claesbackman/AI-research-feedback

    A collection of Claude Code skills that provide multi-agent academic reviews for research papers, grant proposals, and pre-analysis plans, including paper-code reproducibility checks.

  • Project
    Rzna-5559/Mrite

    A one-stop repository for mathematical modeling competitions that integrates automated solving via Claude Code, LaTeX paper templates, and a library of common algorithms.