publication-figures
Use whenever you generate or review a chart, plot, table, or paper figure in this workspace, including work delegated by paper-writing, literature-survey, and experiment skills. Applies the Open Science publication style, enforces readable final-size layout for figures and tables, and rejects generic diagram-tool output as a publication figure. Interactive Plotly/HTML may be used for exploration, but paper delivery requires a static publication-ready export.
git clone --depth 1 https://github.com/ai4s-research/open-science /tmp/publication-figures && cp -r /tmp/publication-figures/runtime/skills/core/publication-figures ~/.claude/skills/publication-figuresSKILL.md
# Publication Figures and Tables
Make generated figures **publication-grade and on-system by default**. Every
figure you produce with matplotlib must use the bundled Open Science style, so a
figure in a report and a stat tile in the app read as one design system.
When a paper/survey skill or target venue specifies its own publication palette
and physical type scale, that more specific standard overrides this palette;
the final-size, collision, and figure-form rules below still apply.
## Apply the style (always, before plotting)
The style file `openscience.mplstyle` sits next to this SKILL.md. Load it by
absolute path at the top of any figure script:
```python
import matplotlib.pyplot as plt
from pathlib import Path
# This skill's directory — the style ships beside SKILL.md.
STYLE = Path(__file__).resolve().parent / "openscience.mplstyle" if "__file__" in dir() else None
# In a notebook/agent cell, use the skill's deployed path directly:
plt.style.use(str(STYLE)) if STYLE and STYLE.exists() else plt.style.use("default")
```
If you cannot resolve the path, set the palette inline (same hexes as below).
## Choose a paper-appropriate form
- Use matplotlib/seaborn for quantitative evidence; use a purpose-built
TikZ/SVG/PDF schematic for a method or architecture; use scientific image
panels only with source, crop, scale, and processing provenance.
- Do not use Mermaid, PlantUML, generic flowchart/mind-map output, diagram-editor
screenshots, or notebook/UI screenshots as a final paper figure. They may be
scratch aids only. Graphviz is acceptable only when the graph itself is the
analysed data, not as a shortcut for a generic process diagram.
- Every figure must make evidence, a mechanism, or an experimental design easier
to understand. Decorative roadmaps, funnels, icon collages, and stock process
diagrams do not belong in a paper.
## The shared palette (single source of truth)
These are the exact hues the app's native charts use. Assign categorical series
in this fixed order — never a different order, never a cycled 9th hue.
| Slot | Hue | Light hex |
|------|-----|-----------|
| 1 | blue | `#2a78d6` |
| 2 | aqua | `#1baf7a` |
| 3 | yellow | `#eda100` |
| 4 | green | `#008300` |
| 5 | violet | `#4a3aa7` |
| 6 | red | `#e34948` |
| 7 | magenta | `#e87ba4` |
| 8 | orange | `#eb6834` |
Sequential (magnitude, one hue light→dark): `#cde2fb #9ec5f4 #6da7ec #3987e5
#256abf #184f95 #104281`. Diverging: blue ↔ red with a neutral gray midpoint.
## Rules (from the app's dataviz standard)
- **One y-axis.** Never two scales on one plot — use two charts or index to a
common base.
- **Categorical color = identity, assigned in slot order; sequential = one hue
by magnitude; diverging = two hues + gray midpoint.** Never a rainbow.
- **Thin marks, recessive chrome:** 2px lines, ≥6pt markers, hairline y-grid
only, no top/right spines (the style sets these).
- **Label selectively** — the endpoint or the extreme, never a number on every
point. A legend is present for ≥2 series; a single series needs none (the
title names it).
- **Text stays in ink**, never the series color. Identity comes from the mark.
- **Fit by design, not tiny type.** Size for the final column/page first. When
labels compete for space, enlarge the canvas, wrap/shorten labels, show fewer
ticks, move the legend, use a horizontal chart, or split panels. Do not shrink
any text below 7 pt at final size.
- **No collisions.** Labels, ticks, legends, annotations, watermarks, panel
letters, and data marks may not overlap or be clipped. Use constrained layout
where appropriate, but always inspect the rendered result; an automatic
layout call is not proof.
- **Save clean:** `plt.savefig(path, bbox_inches="tight")` (the style sets dpi).
## Keep paper tables readable
- Design every table for its final column or page width. Body text must remain at
least 8 pt at final size.
- Resolve an over-wide table in this order: shorten or wrap headings and move
units/details into the caption or notes; use flexible text columns and aligned
numeric columns; split the table or move secondary columns to an appendix; use
landscape for a genuinely wide appendix table; reduce cell padding last.
- Do not solve width by scaling the entire table until it fits. If scaling would
make text smaller than 8 pt, redesign or split the table.
- Keep precision purposeful, repeat units once, and avoid dense vertical rules.
- Render the containing PDF or page to an image at the intended output size and
inspect it. A table fails if it crosses margins, clips content, overlaps text,
or is readable only by zooming.
## Final-size inspection
For PDF output, rasterize the actual final artifact before delivery:
```bash
pdftoppm -png -r 180 paper.pdf /tmp/paper-check/page
```
Inspect every page containing a generated figure or table. Source dimensions,
`tight_layout`, `constrained_layout`, and a successful compile are not evidence
that the rendered result is readable.
## Checklist before returning a figure or table
1. Style applied (palette + chrome from `openscience.mplstyle`).
2. Series colors assigned in slot order; ≤8 series (else group into "Other").
3. Single y-axis; legend iff ≥2 series; axis labels + units present.
4. Figure form is appropriate for a paper; no Mermaid/PlantUML/generic diagram.
5. Table fits by structure rather than tiny type; final body text is at least
8 pt.
6. Final PDF/page was rasterized and visually inspected: readable text, no
overlap, clipping, or margin overflow.
7. Saved to the workspace and referenced by path so it surfaces as an artifact.A test skill that says hello. Use when you want to test skill loading or verify that the skill system is working.
Verify apps/desktop frontend changes visually without launching the Tauri app or a live model
Use whenever you write or run scientific analysis code (physics, earth/geo, biology, chemistry, or social science) in this workspace — before executing it and again after generating results. Runs a deterministic domain-correctness gate that catches code which runs but is scientifically wrong (unit/dimension mismatch, Euclidean distance on lat/lon without a CRS, 0-based/1-based coordinate and strand errors, impossible SMILES valence, uncorrected multiple comparisons, averaging a categorical code). Surfaces structured findings; never claims the code is correct.
Use BEFORE reading any data file that could be large (CSV/TSV, Parquet, HDF5, FITS, NetCDF, NDJSON, genomics FASTQ/FASTA/VCF/BAM, GRIB, ROOT, or big text/simulation logs like VASP OUTCAR). Returns a compact memory pointer — header/schema/shape/sample/key numbers — by introspection and sampling in bounded memory, so you never load a file bigger than the context window into the model. Reference data via the pointer; read specific ranges deterministically.
Use when the user asks to run heavy or GPU work on Modal (the cloud compute platform) — writing a Modal function in the workspace, running it with the user's own `modal` CLI + token, and bringing results back. Data-to-compute for jobs too big for the laptop, without a Slurm cluster.
Use when the user asks to run, submit, monitor, or cancel a job on a remote machine over SSH — their own GPU/CPU server, a workstation, or a Slurm cluster ("the cluster", a login node, "my 3090 box", "the compute server"). Picks a saved machine, runs the work directly over SSH (or via Slurm when present), tracks it, and fetches results back into the workspace.
Use whenever you run statistical analysis for the social sciences (regression, hypothesis tests, econometrics) or read Stata (.dta) / SPSS (.sav) data in this workspace. Enforces an execute-don't-interpret boundary (surface estimates, don't volunteer causal claims), checks the analysis against a preregistration plan for HARKing, verifies reproducible seeds, and reproduces .dta/.sav estimates via R. Flags integrity risks; never certifies the analysis is sound.
Use when the user asks to review, verify, or audit a report, manuscript, or analysis in the workspace for traceability — resolving citations, flagging numbers with no source, and checking figures against the code that generated them. Emits a structured review block the app renders as reviewer findings. Verifies traceability, never "correctness".