ts-paper-latex
Stage 7 (final assemble + compile) of the ts-paper suite. Assemble drafted LaTeX section bodies + refs.bib + blueprint into a compilable paper in the active TEMPLATE and compile it to PDF.
Install / Use
npx skills add Spark-To-Paper-Skills/spark-to-paper-skills --skill ts-paper-latexInstalls into whichever agent you are using.
SKILL.md
Installable skill definition
Quality Score
Category
Development & EngineeringSupported Platforms
Our assessment of ts-paper-latex
ts-paper-latex scores 88/100 on our quality scale, 1957th of 4,700 Development & Engineering skills we index (top 42%).
Its SKILL.md is 7.2 KB long, well organised into 8 sections with 1 code example: a thorough specification that gives an agent plenty to work with.
With 1,126 GitHub stars, it is one of the more widely adopted skills in the catalogue.
Maintenance, license and trust
- The repository was last updated about 3 months ago, so ts-paper-latex is actively maintained.
- Our last check on 2026-11-01 found the source still online.
- It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
- Its trust signals score 100/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.
Safety scan
No issues foundOur scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.
AI review by kimi-k2.7-code on 2026-10-04. Automated pattern scan on 2026-10-04. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.
ts-paper-latex compared with similar skills
All 4 of these similar skills score higher than ts-paper-latex; compare them before choosing.
| Skill | Score | Stars | Updated | Format |
|---|---|---|---|---|
| ts-paper-latex (this skill)by Spark-To-Paper-Skills | 88 | 1.1k | 3mo ago | SKILL.md |
| taste-skillby Leonxlnx | 100 | 116.8k | 8d ago | CLAUDE.md |
| ai-job-searchby ai-job-search-dev | 100 | 72.1k | today | CLAUDE.md |
| claude-howtoby luongnv89 | 100 | 47.3k | 2d ago | CLAUDE.md |
| lobehubby lobehub | 100 | 85.6k | today | MCP Server |
Frequently asked questions
- How do I install ts-paper-latex?
- Run
npx skills add Spark-To-Paper-Skills/spark-to-paper-skills --skill ts-paper-latex. The install tabs above show the steps for each supported agent. - Which AI agents does ts-paper-latex work with?
- It is written for Universal, as a SKILL.md file. Other agents that read the same format can often use it too.
- Is ts-paper-latex safe to use?
- Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is MIT-licensed and scores 100/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
- Is ts-paper-latex still maintained?
- The repository was last updated about 3 months ago, so ts-paper-latex is actively maintained.
Skill content
View source on GitHubname: ts-paper-latex description: > Stage 7 (final assemble + compile) of the ts-paper suite. Assemble drafted LaTeX section bodies + refs.bib + blueprint into a compilable paper in the active TEMPLATE and compile it to PDF. Copies the template's .sty/.cls + assets, runs the deterministic template-driven post-processes (caption position, merge adjacent \cite for numeric styles, canonical headings from the spec, keyword formatting), and compiles once per call; the agent drives a bounded error-fix loop around it. Use to build/compile the final PDF.
ts-paper-latex — assemble & compile the paper PDF (template-driven)
Turn the finished sections/<id>.tex + refs.bib + blueprint.json into a compiled
main.pdf. This stage never authors or alters content — no renaming methods/datasets, no
filling result numbers (proposal = blank cells stay blank). It is format + assembly + compile only.
Template-driven: assemble_paper.py reads template.json from the workdir and threads it
through everything — section order + headings, table/figure caption position, keyword cap,
whether to merge \cite (numeric templates only), and the preamble (the template's own
main.tex.tmpl, filled by safe @@token@@ substitution) + which .sty/.cls + assets to copy.
The default template is ts_iieta (two-column IIETA); neurips (single-column author-year) is also
bundled and compiles through the same script — nothing here is hardcoded to TS. If no template.json
is in the workdir it falls back to bundled ts_iieta (backward-compatible).
🔴 HARD RULE — NEVER fabricate a venue template. A template's style files (
.sty/.cls) must come from exactly ONE of: (a) a user-provided template (the user drops atemplates/<name>/dir or points at the official files), or (b) the venue's OFFICIAL style files, fetched verbatim from the official source (the conference/journal style-file URL, e.g. NeurIPSmedia.neurips.cc/.../neurips_<year>.sty) and copied UNCHANGED. Do NOT hand-author, approximate, or "make it look like" a venue — a self-invented.stygets margins/fonts/notice/line-numbers wrong and is unusable for submission. If a requested venue has neither a user-provided nor an obtainable official template, STOP and ask the user for the official files — never fabricate one. ⚠️ The currently-bundledts_iietaandneuripsstyles are unofficial approximations ("official": falsein theirtemplate.json); treat them as demo-only and replace with the venue's official.sty/.cls(or a user-provided template) before any real submission.
What's bundled (clean, copyright-safe assets you own)
assets/ts_iieta.sty— our own two-column IIETA style (masthead with logo + blue band, 10pt Times, ALL-CAPS numbered sections,Figure N./Table N.captions, numeric[n]cites withsort&compress).assets/iieta_logo.png— the masthead logo (compiles fine if absent via\IfFileExists).scripts/assemble_paper.py— the deterministic backbone.
Run it
python scripts/assemble_paper.py <workdir>
The script: reads blueprint.json (title, keywords, abstract, section_order, optional authors/doi/dates/journal); for each sections/<id>.tex applies the post-processes and prepends the canonical ALL-CAPS heading; builds main.tex from the TS template; copies ts_iieta.sty + iieta_logo.png into the workdir; runs latexmk -pdf. It prints a JSON status {ok, compiled, exit, error_count, error_tail, sections, order, assets, template, workdir}.
Deterministic post-processes it applies (no LLM)
- Table captions ABOVE the tabular (IIETA convention) — both inline minipage and floating
table/table*. - Merge adjacent
\citeruns into one (\cite{a} \cite{b}→\cite{a,b}) so they render as[1, 2]/[3-5]with the style'ssort&compress. - Canonical section headings — strips any heading the writer added and forces
Introduction / Related Works / Methodology / Experimental Results / Discussion and Analysis / Conclusion, which the style renders as1. INTRODUCTION… - Strip heading numbers (
\subsection{3.1 Foo}→Foo), table-width safety net (wraps a bare\begin{tabular}in\adjustbox), single-paragraph abstract merge. - Keyword formatting — dedup, cap 6, comma-join into the
\tsSetKeywordsmacro (empty → the line is omitted). - The bundled
ts_iieta.styalso loadsinputenc[utf8]as a Unicode compile-net.
Compile + fix loop (quality gate — do not skip)
- The agent owns the ≤3-try fix loop; the script compiles once per call and only renders the verdict (no script-side loop). The script reports
compiled(PDF exists ANDerror_count == 0— it already trusts the log, not the exit code) and a structurederror_count. Success also requiresmain.bblresolved (no[?]citations). Treaterror_count > 0as a red gate that blocks done. - Run with
--backupon the FIRST attempt only sosections/is snapshotted tosections.bak/(rollback if a fix makes it worse) — do not re-pass--backupon later attempts or you overwrite the good baseline. - If
compiled:falseorerror_count > 0, readerror_tail, then edit the offendingsections/*.texwith minimal, syntax-only fixes (close unbalanced$/environments, escape stray& % # _in text, fix mispaired\begin/\end,1-10→1--10). Never change content, math semantics, citations, labels, or--placeholders. - Re-run. Bounded at ~3 attempts; if
error_countgoes UP, restore fromsections.bak/and report rather than thrash. - One figure-specific error class: a
File \figures/<label>' not foundor a bad-bounding-box error is **not** a content bug — fix it by **re-running the figure stage** (regenerate the missing.pdf/.png`), never by editing section prose. - Finally write
logs/7_latex.io.md(INPUT: section files + refs.bib; DECISIONS: each compile error + the minimal fix; OUTPUT: main.pdf page count, error_count=0), andlogs/index.mdlinking every stage's log.
Figures embed as vector PDF (no extra wiring)
Figures are referenced extension-less — \includegraphics{figures/<label>} — and the figure stage
ships figures/<label>.pdf (editable vector) beside the kept figures/<label>.png. Both bundled .sty
already \RequirePackage{graphicx}, so under pdflatex/latexmk the .pdf is embedded (its extension is
preferred over .png) with no change to assemble_paper.py (it never parses \includegraphics or
figure extensions). The vector is pre-rendered by cairosvg (in ts-figure-optimize's hybrid export), so there is no
\includesvg/Inkscape/--shell-escape dependency. The run_gates.py all vector check confirms every
figure has its .pdf sibling — a raster-only figure is a red gate.
Front matter / metadata
Authors/affiliation/email/DOI/dates default to clearly-marked placeholders ([AUTHORS TBD], 10.18280/ts.XXXXXX) for a proposal — that is correct; fill real values only if the user provides them in blueprint.json (authors, doi, dates, journal).
Done check
Report: page count, sections included, that captions are above tables, cites merged, keywords present, and that the citation linter (ts-paper-cite/scripts/citations_lint.py) passes with zero stubs/orphans.
Related Skills
taste-skill
116.8kAgent skills that give coding agents taste: art direction, reference images and a browser check instead of generic AI pages
ai-job-search
72.1kThe job search that runs on your machine. AI job application framework built on Claude Code: evaluate postings, tailor CVs, write cover letters, prep interviews. Fork it and own it.
claude-howto
47.3kA visual, example-driven guide to Claude Code — from basic concepts to advanced agents, with copy-paste templates that bring immediate value.
lobehub
85.6k🤯 LobeHub - Super-Agents on Your Team. Add super-agents to the chats your team already uses.They take on work by themselves and get better every week.
Languages
Trust signals
From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.
