---

name: scientific-schematics description: Create polished technical diagrams for CS papers, slides, posters, demos, and repo documentation. Use when a Mermaid diagram is not enough and the task needs a cleaner architecture figure, system overview, pipeline graphic, benchmark setup diagram, or agent workflow illustration.

Scientific Schematics

Use this skill for polished technical figures, not for routine text-native diagrams. Prefer Mermaid first when editability matters more than appearance.

Good Fits

• model architecture overviews
• system component diagrams
• agent or tool-call workflows
• training / inference pipelines
• benchmark or evaluation setup figures
• data processing or serving architecture diagrams

Working Rules

• Start from the message the figure must communicate.
• Label components by role, not by decorative wording.
• Make data flow, control flow, and evaluation boundaries explicit.
• Avoid overloading one figure with both architecture and results.
• Prefer a small number of well-grouped components over exhaustive detail.

Local Resources

Read references/best_practices.md when you need diagram design guidance.

Use local scripts only when the configured image-generation workflow is available and the user actually wants rendered figures:

• scripts/generate_schematic.py
• scripts/generate_schematic_ai.py

The quick-reference files in references/ are inherited upstream support material. Use them selectively.

Relationship to Other Skills

• Use markdown-mermaid-writing for editable diagrams in docs and notes.
• Use this skill when the figure is headed into a paper, poster, deck, or polished README.

Default Output

When the user asks for a diagram without a target format, return:

1. the figure concept
2. required components and labels
3. a recommended diagram structure
4. whether Mermaid or a rendered schematic is the better fit