ModelViz Scientific Plotting

Problem

Writing paper- or competition-ready figures is less about “making a chart” and more about connecting a natural-language request, real tabular data, a fixed template, and reproducible code. ModelViz Skill (canonical name @user_147de1c0/modelviz-skill) works with csv / xlsx / xls files, selects an existing local template, preserves its layout, color scheme, and legend conventions, replaces demo data with real column values, and outputs chart files, the adapted script, and quality reports.

Workflow

• Requirement parsing: extracts goal, chart type, style words, negative requirements (for example, “no radar chart”), and ambiguity; if the request is vague or column meanings are unclear, it asks for clarification first.
• Template recall and selection: the local template library covers 12 categories and 89 templates; it performs deterministic candidate recall, then selects a final template from 5–10 candidates using data facts instead of blindly choosing the top score.
• Data adaptation and dependency handling: program code reads files, builds sampling facts, checks and installs validated dependencies; the model maps columns and adapts template code without inventing nonexistent columns.
• Execution and quality checks: runs adapted_plot.py through a subprocess, validates output location, PNG readability, pixel variance, runtime errors, font/legend/layout warnings, then performs visual comparison and limited local repairs when needed.

Boundaries

Best for “data is ready and I need publication-quality figures.” It is not intended for pure math explanations, modifying user data, running arbitrary non-plotting code, or forcing unsupported chart types when no suitable template exists. Intermediate JSON and final image artifacts are kept separate, which helps engineers audit the workflow or reproduce the generated code.