Science Figure

Science Figure

Use this skill before plotting or rewriting legends. A Science-family figure is an evidence argument, not a decorative arrangement.

First move: figure contract

Before editing or making a figure, define:

• core conclusion
• target audience
• panel list and each panel's question
• evidence hierarchy
• source data for each quantitative claim
• likely reviewer risk
• journal and article type, if known

If the user has not supplied data, create a panel plan and legend scaffold rather than fabricating plots.

Open extra files

| File | Open when | |---|---| | [references/figure-contract.md](references/figure-contract.md) | Need to convert results into panel map, evidence hierarchy, or reviewer-risk checks | | [references/legend-and-panel-style.md](references/legend-and-panel-style.md) | Need legend structure, panel titles, typography, color, or abbreviation rules | | [references/qa-checklist.md](references/qa-checklist.md) | Before final delivery or when auditing an existing figure |

Figure workflow

1. State the figure's one-sentence conclusion.
2. Assign one job per panel:

context, design, validation, main result, mechanism, comparison, generalization, or limitation.

1. Order panels by reader logic, not by experiment chronology.
2. Remove redundant panels unless they test different risks.
3. Map every claim in the legend to a panel, method, statistic, or source-data file.
4. Audit visual encoding:

• color means the same thing across panels
• axes and units are explicit
• error bars and sample sizes are defined
• statistical tests are specified where needed
• image panels have scale bars and acquisition context

1. Return the figure plan, legend, and unresolved data/source issues.

Science-family design defaults

• Favor dense but legible multi-panel evidence.
• Put the strongest result in the visual position the reader sees early.
• Use muted neutral structure plus a small number of meaningful accent colors.
• Avoid rainbow palettes unless the data are continuous and colorbar-labeled.
• Never use color alone to encode categorical distinctions.
• Legends should explain what was done and what is shown, not claim more than the data support.

Output format

Figure conclusion
[one sentence]

Panel map
| Panel | Job | Evidence | Key visual | Reviewer risk |

Legend draft
[legend]

Source-data / methods checks
- [needed files, statistics, scale bars, n, replicates, software, or "None"]

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: shikaijieskj
• Source: shikaijieskj/science-skills
• License: MIT

Install and usage instructions live in the source repository linked above.