Critique academic figures for format, colorblind safety, legibility, overplotting, and category count. Use when reviewing figures before submission.
Systematically review academic figures and flag common issues that hurt readability, accessibility, and print quality.
presubmit-checks skill as one of its checks (future)Search in this order:
\includegraphics, \includesvg, \includestandalone, and figure-related \input commands in the TeX treefigures/, fig/, then a recursive search for figure files in the paper directory (exclude , , and build artifacts)node_modules.gitOnly ask the user to confirm the figure list if multiple candidate manuscripts are found or the resolved figure set looks suspiciously broad.
Run these checks per figure using Bash:
Classify each figure as:
.pdf, .svg, .eps.png, .jpg, .jpeg, .tiff, .gif, .bmpDo not decide acceptability from extension alone.
Raster is acceptable when the content is inherently image-based, such as microscopy, medical imaging, satellite or photographic imagery, screenshots, or other pixel-native data products.
For .pdf figures only, run:
pdfimages -list <figure.pdf>
Interpretation:
Only flag if the PDF appears to be essentially a raster wrapper rather than a genuinely mixed figure.
If pdfimages is not installed, note this and skip the check (do not fail the whole review).
Read each figure image using the Read tool and evaluate the following. For each check, provide a clear pass/flag verdict with explanation.
Subpanels: Many academic figures contain multiple panels (a, b, c, ...). When a figure has subpanels, evaluate each panel individually — one panel may have issues while others are fine. Reference specific panels in flags (e.g., "panel (c) is overplotted"). Also check that panels are clearly labeled (a, b, c or A, B, C) — unlabeled panels make it hard to reference from the text.
Examine the colors used to distinguish data categories or encode information.
Count the number of distinct categorical colors or legend entries.
Assess whether text elements (axis labels, tick labels, legend text, annotations) are legible at the figure's likely print size (typically a single column ~3.5 in or double column ~7 in).
If the figure is a scatterplot (or similar point-based plot):
A "dynamite plot" is a bar chart with error bars used to show point estimates and uncertainty. It hides individual data points and the underlying distribution, and the bar's visual metaphor of physical accumulation is misleading (especially with non-zero baselines or log scales).
Present a per-figure report in this format:
## Figure: <filename>
| Check | Verdict | Notes |
|-------|---------|-------|
| File format / resolution | PASS / WARN / FLAG | ... |
| Raster-components-in-PDF | PASS / FLAG / SKIP | ... |
| Colorblind risk | PASS / FLAG | ... |
| Category count | PASS / FLAG | ... |
| Font legibility | PASS / FLAG | ... |
| Overplotted scatter | PASS / FLAG / N/A | ... |
| Dynamite plot | PASS / FLAG / N/A | ... |
**Recommendations:**
- ...
After all figures, provide a summary count: N figures reviewed, N flags total.