agentsclimarketplace

Sci figure generation

Skill keithhegit/sci-using-kit/skills/sci-figure-generation

A skill family for evidence-bounded SCI manuscript workflows: literature vetting, claim-evidence alignment, paper revision, reviewer preflight, LaTeX editing, and submission packaging.

Install
npx -y skills add keithhegit/sci-using-kit --skill sci-figure-generation

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Use when creating, revising, validating, or packaging SCI/arXiv manuscript figures, graphical abstracts, system diagrams, mechanism diagrams, evidence snapshots, image-generation prompts, figure captions, or Overleaf figure assets under claim-evidence constraints.

SKILL.md

6.0 KB, as published. Nobody here has run it

SCI Figure Generation

Overview

Create publication-safe figures for SCI/arXiv manuscripts by binding every visual claim to evidence and by separating visual ideation from final submission assets.

Core principle: a figure is a claim surface. Treat generated images, captions, labels, axes, and layout choices as manuscript claims that require the same evidence discipline as prose.

Required Pairing

Use this skill with:

  • sci-using-kit for project routing and evidence bundle discovery.
  • imagegen when a raster image or visual concept is generated.
  • sci-latex-editing when inserting figures into live LaTeX or Overleaf source.
  • sci-submission-preflight before packaging arXiv or journal materials.

Workflow

1. Classify The Figure

Choose the smallest useful figure type:

TypeBest usePreferred production path
Graphical abstractFirst-look system storyImage generation draft, then text QA
System / workflow diagramArchitecture, agent flow, interfacesTikZ/vector if exact; imagegen if visual polish matters
Evidence snapshotRuntime, metrics, logs, resource dataDeterministic chart or card-style graphic; avoid invented visuals
Comparative positioningPrior-work layers and claim boundaryTable/vector; do not imply direct speedup unless comparable
Limitation / release boundaryPublic vs private artifactsUsually text/table; avoid spotlighting weaknesses as main figure

2. Build A Figure Evidence Brief

Before generating or editing a figure, write or mentally fill:

  • Target manuscript and section.
  • Figure purpose in one sentence.
  • Verified facts and numbers allowed in the figure.
  • Forbidden claims, private terms, and unsupported metrics.
  • Whether the figure is for manuscript, appendix, internal preflight, or promotion.
  • Whether labels must be all English for arXiv/journal submission.

If a figure contains a number, unit, axis, comparison, or conclusion, trace it to a table, log, script, paper reference, or confirmed fact.

3. Choose Imagegen vs Deterministic Drawing

Use imagegen when the value is visual polish, composition, or graphical abstract style.

Use deterministic drawing such as TikZ, matplotlib, PowerPoint, BioRender, or vector editing when:

  • exact text must be reliable;
  • axes, scales, units, or values matter;
  • the figure may be challenged by reviewers;
  • imagegen would invent categories, hardware, labels, or metrics.

Hybrid path is often best: use imagegen for layout exploration, then recreate the final figure with controlled text and assets.

4. Prompt For Publication Safety

For generated figures:

  • Require English-only labels for arXiv unless the target venue explicitly allows another language.
  • Specify exact text labels and short labels only.
  • State forbidden private details: IP addresses, hostnames, internal paths, vendor-sensitive labels, model/runtime names, exact private throughput, and unreleased weights unless approved.
  • State unsupported-claim bans: no flight validation, radiation qualification, calibrated power, detector SOTA, geolocation accuracy, P50/P95/P99, or speedup unless evidenced.
  • For mixed-unit evidence, require separate cards or panels instead of a shared axis.

For reusable prompt patterns, read references/prompt-patterns.md.

5. Validate The Image

Inspect the output before any manuscript insertion:

  • Text: all labels are readable, English-only when required, and not misspelled.
  • Evidence: all numbers match the evidence bundle exactly.
  • Units: no mixed units on one misleading axis; no fake precision.
  • Scope: the image does not make the work look like flight validation, SOTA detection, or calibrated spacecraft power unless that is truly supported.
  • Privacy: no private infrastructure, hardware vendor/model, model/runtime, private path, or host detail leaks.
  • Visual role: the figure supports the paper's main story instead of foregrounding limitations unnecessarily.

If generated text is imperfect but composition is useful, keep the image as a draft and recreate the final figure in a controlled tool.

6. Caption And Placement

Write captions with:

  • what the figure shows;
  • what evidence level it represents;
  • what it does not claim when the figure could be misread.

Place figures where they reduce reader confusion:

  • system overview near Introduction/System Architecture;
  • workflow near Method/Agent Workflow;
  • evidence snapshot near Results;
  • release boundary in Appendix or reproducibility notes unless it improves the main argument.

7. LaTeX / Overleaf Packaging

When integrating into LaTeX:

  • copy final assets into a project-local figures/ directory;
  • reference with relative paths only, e.g. figures/fig1_system_overview.png;
  • avoid absolute paths and non-ASCII public source paths;
  • check every \includegraphics{} path exists;
  • keep source package minimal: include final figures, not discarded drafts or prompt notes unless intentionally supplemental;
  • run sci-submission-preflight before external upload.

Red Flags

Stop and revise if a figure:

  • contains Chinese labels for an English arXiv submission;
  • displays different units as if they share a common axis;
  • converts limitations into a visually dominant "weakness list";
  • invents example classes, satellites, hardware, metrics, baselines, or geographic precision;
  • implies a direct performance improvement over incomparable prior work;
  • includes private host access, IPs, local paths, model/runtime labels, or exact private throughput.

Output Contract

For figure tasks, report:

  • figure type and intended manuscript section;
  • evidence consulted;
  • generated or edited asset paths;
  • prompt path or prompt text when relevant;
  • validation performed;
  • residual risks before final submission.

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.