| name | visualization |
| description | Render clean, presentation-ready charts (bar, line, scatter, pie, heatmap) as PNG files from result tables produced during analysis. |
Visualization
Turn the aggregated tables in data/analysis/*.csv into charts saved to charts/*.png. Every key finding in the report should be backed by a chart.
Script
python3 /.agents/skills/visualization/scripts/make_chart.py \
--workspace ./workspace \
--data data/analysis/revenue_by_category.csv \
--type bar \
--x category --y revenue \
--title "Revenue by Category" \
--output charts/revenue_by_category.png
| Argument | Required | Description |
|---|
--workspace | yes | Root workspace directory |
--data | yes | CSV (relative to workspace) to plot — usually a data/analysis/*.csv |
--type | yes | bar, barh, line, scatter, pie, or heatmap |
--x | for bar/line/scatter | Column for the x-axis (or labels for pie) |
--y | for bar/line/scatter/pie | Column(s) for the y-axis; comma-separated for multi-series line |
--title | recommended | Chart title |
--output | yes | Output PNG path (relative to workspace), under charts/ |
--top | no | Keep only the top N rows by --y (useful for bar charts) |
--xlabel / --ylabel | no | Axis labels |
Chart-type guidance
- bar / barh — compare a measure across categories (revenue by category, count by region). Use
barh when there are many categories or long labels.
- line — trends over time.
--x is the date column; pass multiple --y columns for several series.
- scatter — relationship between two numeric measures (price vs quantity).
- pie — share/composition when there are few categories (≤ 6). Prefer bar otherwise.
- heatmap — a pivot table or correlation matrix (the CSV is read as a matrix; first column is the row index).
Conventions
- Save all charts under
charts/ with descriptive snake_case names that match what you reference in report.json.
- One idea per chart. Keep titles short and business-oriented.
- For heatmaps, pass a CSV whose first column is the row label and remaining columns are the matrix values (e.g. a
pivot_table(...).reset_index() output).
Notes
- The script uses a non-interactive matplotlib backend (
Agg) so it works headless.
- If a chart fails, log the error and continue — never let a chart block the final report.
