Skip to content

Releases: Albus-White/spark-to-paper-skills

v1.1.0 — Claude Code plugin support

26 Jun 05:16
@u7079256 u7079256
v1.1.0
052524c

Choose a tag to compare

View all tags
v1.1.0 — Claude Code plugin support Latest
Latest

What's new

Plugin install support — the repo is now a proper Claude Code plugin. One command to install:

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

Skills auto-load as /spark-to-paper:ts-paper, /spark-to-paper:ts-idea2story, etc.

Changes

  • Restructured as Claude Code plugin: all skill dirs moved into skills/ with .claude-plugin/plugin.json manifest
  • Four install options: plugin clone (recommended), --plugin-dir (try first), standalone copy, git submodule
  • Soft update notification (from v1.0.1): checks GitHub Releases API on each run, 24h cache, never blocks
  • MIT License added

Breaking change

Skill directories moved from ts-*/ to skills/ts-*/. If you previously copied skills with cp -r ts-*, update to cp -r skills/ts-*.

Full changelog

v1.0...v1.1.0

Assets 2
Loading

v1.0.1 — Soft update check + install guide

26 Jun 05:03
@u7079256 u7079256
v1.0.1
9d6a8cb

Choose a tag to compare

View all tags
v1.0.1 — Soft update check + install guide

Patch release with post-v1.0 improvements.

What's new

  • Soft update notification — check_update.py queries GitHub Releases on each skill run and shows a one-line notice when a newer version is available (24h cache, silent when up-to-date, never blocks the pipeline)
  • README install guide rewritten — three clear options: clone & copy, release download, or git submodule
  • VERSION tracking — VERSION file at repo root + inside ts-paper/ so the update check works for both repo and installed-skill scenarios

Fixes

  • Fixed false-positive update notification when skills were installed without the root VERSION file (added FALLBACK_VERSION constant + ts-paper/VERSION)

Full changelog

v1.0...v1.0.1

Loading

v1.0 — Drop a spark. Get a paper.

26 Jun 04:29
@u7079256 u7079256
v1.0
3a60a60

Choose a tag to compare

View all tags
v1.0 — Drop a spark. Get a paper.

spark-to-paper-skills v1.0

The only Claude Code skill suite that goes fully end-to-end — from a one-line idea to a compiled, publication-format PDF with editable vector figures and machine-checked integrity.

Highlights

  • End-to-end pipeline: idea → literature → writing → experiments → figures → compiled PDF — all inside Claude Code
  • Editable vector figures: AI-generated rasters are reconstructed as editable SVG/PDF/PPTX via the DrawAI hybrid engine (~0.91 SSIM, pixel-exact render with editable text overlay)
  • Machine-checked integrity: no fabricated numbers (ever), every citation verified via WebSearch + Crossref, deterministic gates fail the build on violations
  • Two integrity modes: proposal mode (forward-looking, no numbers) and data-aware mode (every number traced to real data)
  • Template-agnostic: NeurIPS and IIETA templates bundled; add any venue by dropping a template directory
  • Adversarial peer review: built-in multi-reviewer hardening stage with verbatim-quote anti-skim
  • Auto-experiments: Stage 8 runs feasible experiments on real data, fills result tables, and recompiles

Skills Included (13)

Skill Role
ts-paper Orchestrator — routes input and drives the 7-stage chain
ts-idea2story Raw idea → structured research story + citation seed
ts-kg-build Corpus → research-pattern knowledge graph (optional)
ts-paper-plan Proposal → blueprint.json (title, keywords, contributions, plan)
ts-paper-cite Real bibliography via WebSearch + Crossref (floor 40 refs)
ts-paper-write Draft all sections as LaTeX in one holistic pass
ts-paper-refine Right-size + de-AI scrub + logic self-check
ts-paper-review Adversarial peer-review hardening
ts-paper-figure Figure routing: matplotlib for data plots, image model for schematics
ts-paper-data Data-aware mode: real results → filled tables + plots
ts-figure-optimize Raster → editable vector via DrawAI hybrid engine
ts-paper-latex Assemble + compile final PDF
ts-paper-experiment Run feasible experiments, fill tables, recompile

Quick Start

# Global install
git clone https://github.com/Albus-White/spark-to-paper-skills.git
cp -r spark-to-paper-skills/ts-* ~/.claude/skills/

# Or per-project
cp -r spark-to-paper-skills/ts-* <your-project>/.claude/skills/

Then just ask Claude:

Run ts-paper on this proposal.

Paste your idea, proposal, or proposal + data — the orchestrator auto-routes and runs the full chain.

Requirements

  • Claude Code (skills run inside Claude Code)
  • Python 3.10+ with pip install -r ts-figure-optimize/requirements.txt
  • LaTeX (latexmk + TeX distribution) for compilation
  • Optional: DrawAI runtime (~4 GB), image-model endpoint, LibreOffice

Links

Loading