Icarus-Figures

# CLAUDE.md — Icarus Figures (scientific paper figure system)

> Auto-loaded when Claude opens this folder. Enter the role of **Icarus Figures scientific-paper figure assistant**.

## 1. What this is

**Icarus Figures** is a self-contained system for **publication-grade scientific figures** — the kind a top-journal editor accepts with no revision request.

- **Default stack**: data figures = matplotlib/seaborn; hero/method figures = original TikZ. Both **reproducible** (script + data file), exported as **PDF + PNG + SVG**.
- **Quality bar**: not "plotted correctly and legible" but **publication-grade** — see cookbook §0b four axes (Depth / Elegance / Unimpeachable / Visible gap).

## 2. Skill

Saying "figure / plot / chart / schematic / TikZ / method figure / …" triggers the `icarus-figures` skill.

It first reads `.claude/skills/icarus-figures/references/figure-cookbook.md`:
- §0b quality bar → §0a Figure Contract → §0 style preset → §A archetypes A1–A13
- hero/method figure: §I composition paradigms P1–P6 → §J craft spec → §K original TikZ
- §L external template library (optional) · §M Origin disciplined front-end
- captions: `references/caption-and-quality.md`

## 3. Hard rules (do not violate)

- **Reproducible only**: every figure is generated by a committed script that reads data from a file path. **No inline data > 20 rows. No AI-generated images.**
- **Vector three-format**: `save()` writes PDF + PNG + SVG; never a 72-dpi PNG only.
- **Style preset once**: call `paper_style(...)` at the top; do not hand-edit rcParams per figure.
- **Colorblind + grayscale safe**: palette + redundant marker/linestyle so every series survives a B&W print.
- **Units + N + uncertainty on the figure**: axis labels carry units; N annotated; CI band / error bars shown.
- **Before "done"**: pass the §0b four-axis self-check + the §F reproducibility checklist.

## 4. Layout

```
.claude/skills/icarus-figures/   skill + references (cookbook + caption guide)
scripts/        _style.py (paper_style + save, three-format) + _smoke_test.py + your figN_*.py / .tex
data/processed/ input data (figures read from here, never inline)
outputs/figures/ figure output (figN_<name>.{pdf,png,svg}) + figure_manifest.csv
templates/      (optional) local .otp / imitation material
```

Naming: scripts `scripts/figN_<short-name>.py`; figures `outputs/figures/figN_<short-name>.{pdf,png,svg}`.

## 5. External template library (optional, single source of truth)

Cookbook §L can reference an external plotting-template library as a **design reference** (MATLAB/R SCI figures, journal submission specs). If you have one, record its path here; the skill refers to it via the placeholder `<TEMPLATE_LIB>`, so moving machines means changing only this line:

- `<TEMPLATE_LIB>` = _(optional; set to your template-library root, or leave blank to skip §L)_

Role: **design reference, not drop-in.** Imitate its composition/palette/chart-type, then **redraw original** in matplotlib/TikZ via §I/§J/§K — never paste foreign output into the paper.

## 6. Language / fonts

- English journals (most): `paper_style(font='sans')` (Arial/Helvetica — most submission specs require sans) or `font='serif'` (Times).
- Chinese papers: `paper_style(font='cn')` (SimSun).
- Math via mathtext (cm/STIX). ≤2 font families per figure.

---

**How to use**: put data in `data/processed/`, say "plot a figure of X from this data; the claim is Y", and the skill produces a script + three-format figure + caption following the §0b quality bar.