CS Experiment Figure Studio

Codex skill for turning computer-science experiment outputs into dense, publication-style multi-panel figures.

The skill helps an agent:

• inventory result folders containing CSV, TSV, JSON, TXT, or log files;
• infer experiment semantics such as baseline comparisons, efficiency, ablations, robustness, failures, and case studies;
• plan a compact research-question figure before plotting;
• render editable Matplotlib figures as PDF, SVG, and PNG;
• apply restrained CS/Nature-style visual conventions and run figure QA.

Layout

.
├── SKILL.md
├── agents/
│   └── openai.yaml
├── references/
│   ├── cs-experiment-semantics.md
│   ├── nature-style-transfer.md
│   ├── orn-resource-suite.md
│   ├── panel-archetypes.md
│   ├── qa-checklist.md
│   └── style-system.md
└── scripts/
    ├── inventory_experiments.py
    ├── plan_panels.py
    ├── plot_orn_resource_suite.py
    └── render_cs_figure.py

Quick Start

Create an inventory from a result directory:

python scripts/inventory_experiments.py /path/to/experiment/root --output inventory.json

Plan candidate panels:

python scripts/plan_panels.py inventory.json --output panel_plan.json

Render a first draft:

python scripts/render_cs_figure.py panel_plan.json --output figure_cs_experiment --formats svg,pdf,png

For ORN-Miner-style resource and scalability results:

python scripts/plot_orn_resource_suite.py /path/to/orn-miner/results --output orn_resource_suite --formats svg,pdf,png

Requirements

The scripts use Python 3 with:

• matplotlib
• numpy

Notes

This repository contains the skill instructions, reusable plotting scripts, and style references. It does not include private experiment data or generated figure outputs.