A Claude Code skill that scaffolds and maintains an ACM sigconf conference paper (BuildSys, SenSys, IMC, MobiCom, …) directly from a project's experimental results.
Point it at a project with a paper/paper_config.yaml and it will:
- Scaffold a full ACM
acmartsigconf template populated with your title and authors. - Read
.npz/.jsonfiles inresults/, evaluate expressions you declare, and fill\textit{…}placeholders with the computed numbers. - Swap figure placeholders (
\fbox{FIG_name}) with real\includegraphicsentries pointing at your screenshots. - Compile the PDF cross-platform (tectonic → latexmk → pdflatex auto-detection).
- Report exactly which placeholders filled, which remain, and how many prose
% TODO:s are left.
Invoke it manually in Claude Code at milestone points. Instead of a two-hour paper-writing session, each invocation is ~30 seconds.
WSL2 / Linux / macOS / Git Bash:
git clone https://github.com/MsMelissa/acm-sigconf-paper-skill.git
cd acm-sigconf-paper-skill
./install.shWindows PowerShell (native):
git clone https://github.com/MsMelissa/acm-sigconf-paper-skill.git
cd acm-sigconf-paper-skill
.\install.ps1The installer copies the skill into your user skills directory (~/.claude/skills/acm-sigconf-paper/ on Unix, %USERPROFILE%\.claude\skills\acm-sigconf-paper\ on Windows). Claude Code discovers it automatically from that point on.
Copy the .claude/skills/acm-sigconf-paper/ directory into any project repo. When you run claude from that repo's root, the skill is auto-discovered. Great for shared class projects where every teammate should have the same skill.
# Build the zip:
cd acm-sigconf-paper-skill/.claude/skills
zip -r ../../acm-sigconf-paper.zip acm-sigconf-paper/Then in Claude Code: Settings → Skills → Upload the zip.
In any Claude Code session (inside a project with paper/paper_config.yaml), just type:
- "create my sigconf paper" — scaffolds
paper/paper.texfrom the template - "update my paper" — re-reads results, refills placeholders, recompiles PDF
- "compile the paper" — rebuilds PDF only
Claude invokes the skill automatically when it matches these phrases.
The skill is three Python scripts wrapped with instructions Claude reads:
.claude/skills/acm-sigconf-paper/
├── SKILL.md ← instructions Claude reads
├── assets/
│ ├── sigconf_template.tex ACM BuildSys skeleton with placeholder markers
│ └── paper_config.example.yaml Annotated mapping config
├── references/
│ ├── latex-setup.md Cross-platform LaTeX install
│ ├── acm-sigconf-conventions.md acmart quirks, binhex.tex workaround
│ └── placeholder-conventions.md Marker grammar reference
└── scripts/
├── create_paper.py Scaffold .tex from template
├── update_paper.py Fill placeholders from results
└── compile.py Cross-platform LaTeX compile
A declarative file you put in paper/paper_config.yaml maps your project structure to paper sections:
title: "ThermalSight: Autonomous Thermal Monitoring…"
author: "Mel Torres"
affiliation: "Carnegie Mellon University"
sections:
results.control:
data: "results/hover_test.npz"
fields:
hover_mean_z:
expr: "mean(z[100:])"
format: "%.2f"
overshoot_pct: "100 * (max(z) - 1.5) / 1.5"
figures:
scene: "docs/screenshots/day4_wide.png".tex file contains placeholders:
The drone hovered at \textit{hover_mean_z} m with
\textit{overshoot_pct}\% overshoot (\autoref{fig:scene}).
\begin{figure}\fbox{FIG_scene}\caption{…}\end{figure}Running update_paper.py turns that into:
The drone hovered at 1.48 m with 2.1\% overshoot (\autoref{fig:scene}).
\begin{figure}\includegraphics[width=\linewidth]{docs/screenshots/day4_wide.png}\caption{…}\end{figure}Full marker grammar: references/placeholder-conventions.md.
-
Claude Code — Pro, Max, Team, or Enterprise plan.
-
Python 3.9+ with
numpyandpyyaml:pip install numpy pyyaml
-
A LaTeX engine — one of:
tectonic(recommended — single binary)latexmk(TeX Live / MacTeX / MiKTeX)pdflatex(fallback)
The skill auto-detects. See references/latex-setup.md for per-OS install.
For a class project where everyone should have the same skill:
- Commit the
.claude/skills/acm-sigconf-paper/directory into your project repo. - When a teammate clones and runs
claudefrom the repo, the skill is auto-discovered. - No manual install step required.
Alternatively, each teammate clones this repo and runs ./install.sh once.
PRs welcome — especially:
- Additional data-file formats (
.csv,.parquet). - Additional ACM class templates (
\documentclass[manuscript],\documentclass[acmsmall]). - Example
paper_config.yamlfiles for other CMU capstone projects.
MIT — see LICENSE.
Built by Mel Torres (HelioViews.ai) for CMU 12-831 capstone & beyond.