菜单
This skill should be used when the user asks to "combine figures", "make a figure panel", "arrange subfigures", "create a composite figure", or needs to lay out multiple SVGs/PDFs into a grid for a paper figure.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
This skill should be used when the user asks to "create an architecture diagram", "draw a figure for the paper", "make a publication-quality diagram", "visualize the model architecture", "diagram the pipeline", "typst diagram", "fletcher diagram", or needs scientific figures in Typst following petarv/janosh visual principles. Generates vector diagrams using fletcher/cetz with semantic color coding and a reusable ML component library.
This skill should be used when the user asks to "create a plot", "make a chart for the paper", "visualize metrics", "plot training curves", "render an ablation table as a figure", or needs matplotlib/plotly figures exported as SVG for embedding in Typst reports.
This skill should be used when the user wants to "render a TikZ diagram", "compile LaTeX to SVG", "create an architecture diagram", "draw a neural network figure", "convert .tex to SVG", "use PetarV diagrams", "use nntikz", or asks to draw any specific architecture (BiGRU, BiMamba, JEPA, KAN, CTC trellis, ST-GCN, transformer, LSTM, ResNet). Handles both inline TikZ snippets and full .tex files. Outputs clean SVG files suitable for papers and dashboards.
This skill should be used when the user asks to "create a report", "write a Typst report", "generate a research report", "make a PDF report", or needs a full document with mixed diagram types (fletcher + TikZ SVG). Provides report templates and the @local/diagrams integration pattern.
| name | figure-compose |
| description | This skill should be used when the user asks to "combine figures", "make a figure panel", "arrange subfigures", "create a composite figure", or needs to lay out multiple SVGs/PDFs into a grid for a paper figure. |
Lay out multiple SVGs or Typst sub-diagrams into a grid using Typst's #grid(). Produces a single numbered figure with subfigure labels (a), (b), (c)...
#figure(
grid(
columns: 2,
gutter: 8mm,
[
#image("figures/early-fusion.svg", width: 100%)
#align(center)[*(a)* Early fusion]
],
[
#image("figures/late-fusion.svg", width: 100%)
#align(center)[*(b)* Late fusion]
],
),
caption: [Fusion strategies. *(a)* Early fusion concatenates modalities at the input. *(b)* Late fusion combines predictions at the output.],
) <fig:fusion>
#figure(
grid(
columns: 3,
gutter: 6mm,
[#image("figures/arch-bigru.svg") #align(center)[*(a)*]],
[#image("figures/arch-bimamba.svg") #align(center)[*(b)*]],
[#image("figures/arch-transformer.svg") #align(center)[*(c)*]],
),
caption: [Encoder architectures: *(a)* BiGRU, *(b)* BiMamba (causal), *(c)* Transformer.],
) <fig:encoders>
#figure(
grid(
columns: 2,
gutter: 10mm,
align: horizon,
figure(
ml-diagram(
data-in((0,0), [Input]),
edge("-|>"),
encoder((1,0), [Encoder]),
edge("-|>"),
ctc-head((2,0), [CTC]),
),
caption: none,
),
image("figures/ctc-trellis.svg", width: 100%),
),
caption: [*(a)* CTC pipeline (left) and *(b)* CTC trellis lattice (right).],
) <fig:ctc>
For panels with different natural widths, use fractional units:
grid(
columns: (2fr, 1fr), // left panel is twice as wide
gutter: 8mm,
image("figures/wide-arch.svg"),
image("figures/detail.svg"),
)
Preferred: bold *(a)* inline below each panel.
Alternative for formal publications (no separate subfigure support in Typst):
#let subfig(body, label) = stack(
body,
v(2mm),
align(center, text(8pt)[#label]),
)
#subfig(image("figures/a.svg"), "(a) Early fusion")
| Layout | Gutter |
|---|---|
| 2-column tight | 6mm |
| 2-column standard | 8–10mm |
| 3-column | 5–6mm |
| Mixed types (diagram + SVG) | 10–12mm |