sidiangongyuan/codex-skills-library / skills/paper-visual-craft
paper-visual-craft
VerifiedUse when designing, redrawing, or validating research-paper figures and tables, including Matplotlib plots, LaTeX tables, benchmarks, captions, legends, annotations, color, and typography. Preserves exact evidence and checks rendered PDFs for clipping, overlap, and readability.
| File | Kind | Size |
|---|---|---|
| LICENSE | other | 1.0 KB |
| SKILL.md | skill md | 11.0 KB |
| agents/openai.yaml | config | 290 B |
| references/drivebench-patterns.md | markdown | 2.7 KB |
| references/plot-design.md | markdown | 2.9 KB |
| references/table-design.md | markdown | 9.5 KB |
| references/validation-checklist.md | markdown | 2.3 KB |
references/plot-design.md
View on GitHub# Plot Design
## Data Integrity
- Read the source data directly from JSON, CSV, logs, or tables before plotting.
- Print or assert the key values used in the plot when the task involves paper numbers.
- Keep numeric x-values numeric. Do not convert continuous or ordinal numeric settings into equally spaced categories unless that is the intended comparison.
- Label axes with units. If one panel uses meters and another uses percent, make that visible in the y-axis labels and avoid mixed delta callouts.
## Figure Structure
- Use one panel per question. Split panels into separate output files when the paper benefits from LaTeX subfigure captions.
- Comparison axis first. Before choosing a layout, state what the reader should
compare: models, controls, tasks, classes, metrics, or failure modes. Remove
panels that do not support that comparison.
- Remove plot-internal titles when LaTeX subfigure captions carry the panel names.
- Keep plot-internal text limited to direct data labels, short endpoint labels, or compact annotations in reserved whitespace.
- For appendix diagnostics, make the figure clean and interpretable rather than visually dramatic.
- Diagnostic figures can explain metric behavior, distribution shift, or
decoupling, but their captions must not imply a causal mechanism unless the
experiment directly tests causality.
- Prune redundant views. Duplicate controls, equivalent inputs, uninformative
summary panels, and plain scatter plots with no clear message should be merged
or removed rather than dressed up.
## Matplotlib Style
- Prefer `pdf.fonttype = 42` and `ps.fonttype = 42` for editable vector text.
- Save both `.pdf` and `.png` previews when the figure is paper-facing.
- Use a white background, light gridlines, restrained spines, and clear contrast.
- Use color to distinguish semantic roles, not to decorate. Avoid palettes dominated by one hue unless that is required by the paper style.
- Use consistent legend placement across comparable panels. A compact inside legend is acceptable only when it does not overlap data; otherwise place it just above the axes.
## Labels And Annotations
- Place endpoint labels outside or beside the data trend when there is reserved whitespace.
- Label raw values in the same unit as the axis unless there is a strong reason to show deltas.
- If showing deltas, keep the delta unit consistent across panels: all raw, all relative percent, or all percentage points.
- Use compact tick labels, but do not sacrifice clarity. Examples: `0`, `.1`, `.2`, `.5`, `1.0` can work for dense noise-level plots.
## Visual Checks
- Inspect PNG previews for label overlap, clipped legends, crowded ticks, and excessive whitespace.
- Inspect the compiled PDF page, not only the standalone figure.
- If a figure looks acceptable alone but cramped in the paper, adjust the LaTeX layout or split panels rather than shrinking text below readability.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| brainstormingobra/superpowersYou MUST use this before any creative work - creating features, building components, adding functionality, modifying behavior, or planning anything new, in software or out of it (a talk, a business, a renovation). | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| brand-guidelinesanthropics/skillsApplies Anthropic's official brand colors and typography to any sort of artifact that may benefit from having Anthropic's look-and-feel. Use it when brand colors or style guidelines, visual formatting, or company design standards apply. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| frontend-designanthropics/skillsGuidance for distinctive, intentional visual design when building new UI or reshaping an existing one. Helps with aesthetic direction, typography, and making choices that don't read as templated defaults. | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |
| mcp-builderanthropics/skillsGuide for creating high-quality MCP (Model Context Protocol) servers that enable LLMs to interact with external services through well-designed tools. Use when building MCP servers to integrate external APIs or services, whether in Python (FastMCP) or Node/TypeScript (MCP SDK). | Stars: 198k | Updated | Claude CodeCodex | Verified | Install |