Skip to content

sidiangongyuan/codex-skills-library / skills/paper-visual-craft

paper-visual-craft

Verified

Use 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.

FileKindSize
LICENSEother1.0 KB
SKILL.mdskill md11.0 KB
agents/openai.yamlconfig290 B
references/drivebench-patterns.mdmarkdown2.7 KB
references/plot-design.mdmarkdown2.9 KB
references/table-design.mdmarkdown9.5 KB
references/validation-checklist.mdmarkdown2.3 KB

references/drivebench-patterns.md

View on GitHub
# DriveBench-Style Patterns

Use DriveBench as a design reference, not an asset source. Borrow layout logic, not files.

## What To Borrow

- Benchmark tables with compact capability columns and visually distinct marks.
- Grouped headers that let readers scan task families, metrics, or controls quickly.
- Icons used as semantic compression, paired with short abbreviations and a concise caption.
- Clean / corrupt / no-image control organization that exposes visual-prior reliance.
- Diagnostic tables that show metric disagreement instead of forcing one synthetic score.

## What Not To Borrow

- Do not copy DriveBench PNG icons, figures, or table images.
- Do not reproduce a full corruption taxonomy unless the paper's claim needs it.
- Do not add large benchmark-style tables to the main paper if the result is only an appendix diagnostic.
- Do not make a figure look DriveBench-like at the cost of hiding exact values, units, or sample counts.

## Applying The Pattern

- Start with the experimental question: capability comparison, input control, corruption diagnostic, QA evaluation, or metric disagreement.
- Put conditions as rows when comparing one model across controls.
- Put models as rows when comparing benchmark or model families.
- Use light highlights for the reference input, proposed dataset/model, and strongest control rows.
- Keep captions short but explicit about mark meanings, control definitions, and whether judge scores are secondary.
- Let the table carry the comparison, including secondary metrics, and let the
  prose carry the main story. DriveBench-style polish works because the reader
  can scan the conditions, controls, and metric families without reading
  implementation notes. State a secondary metric in prose only when it changes
  fairness, applicability, or the claim boundary.
- For unfinished QA or control diagnostics, reserve the DriveBench-style table
  structure with the project's explicit missing-value markers. Preserve pending
  (`TBD`) versus inapplicable (`N/A`) when the draft schema defines that
  distinction, or use `--` when specified. Keep internal completion notes out of
  the paper-facing caption and prose, and do not write result language for unrun
  values or retain pending notes in a submission. Structural `N/A` may remain
  when it is inapplicable by definition.

## Claim Safety

- Say "diagnostic" when the experiment is a stress test, control, or appendix probe.
- Avoid "full robustness" unless the corruption set, protocol, and baselines justify it.
- For QA, distinguish objective accuracy from free-form semantic judge scores.
- For CoT, do not treat fluent explanations as proof of grounded evidence unless the metric actually checks evidence source.
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: 322kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall
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: 198kUpdated Claude CodeCodexVerifiedInstall