m4bwav/chartwright
Evergreen chart and graph skills: an evidence-based knowledge base of chart types (when to use, what each excels at, when not to, substitutes, perceptual evidence, notes) and sixteen render targets (Markdown with Mermaid or PlantUML, web with Vega-Lite, ECharts, Plotly, Chart.js and Observable Plot, static PNG/SVG via Python, editable PowerPoint, Excel and Google Sheets charts, Word and Google Docs pictures, D2, QuickChart URLs, terminal sparklines), a deterministic chooser and builder CLI that takes data by file reference to save tokens, and a curation skill that researches novel chart requests and grows the knowledge base so the next request is cheaper.
Builds a chart or graph from data in the place it will be read, choosing from about 80 chart types. Renders to Markdown (Mermaid), web pages (Vega-Lite, ECharts, Plotly, Chart.js), PNG or SVG for Word, Google Docs, Slack or slides, editable PowerPoint charts, image URLs and terminal sparklines. Use when the user asks to chart, graph, plot or visualize data ('bar chart of sales by region', 'line graph of price over time', 'visualize this CSV', 'show this as a', 'make a figure', 'diagram of the numbers'), names a histogram, heatmap, sankey, treemap, sparkline or dashboard, asks which chart fits a question ('how to show signup drop-off'), or wants charts added to a report, README, slide or artifact. Also for 'refresh chartwright' and 'is chartwright stale'. Styling comes from the dataviz skill when present. Not for editing the knowledge base (that is chartwright-curate), codebase graphs (graphify) or data-free diagrams such as flowcharts or sequence diagrams.
Maintains chartwright's knowledge base: adds, corrects, researches and audits its entries so later requests are cheaper and better. Use when the user asks to add a chart type or render target to the knowledge base, update a target, put a note or correction on an entry ('add a note to the pie chart entry', 'never suggest gauges to me'), research a type or question the knowledge base cannot answer ('what is a raincloud plot and when does it beat a violin'), re-rank popularity and velocity from current sources, or audit the knowledge base for gaps, rot and unverified entries. Also use it when chartwright hands over a request its chooser found no match for, and for 'refresh chartwright-curate' and 'is chartwright-curate stale'. Not for making a chart from data (that is chartwright) or codebase knowledge graphs (graphify).