data-visualization-skill

name	data-visualization-skill
description	Design, implement, and verify decision-useful charts, dashboards, figures, and interactive data stories without distorting the underlying evidence.
allowed-tools	["read_file","list_dir","grep_search","glob_files","write_file","edit_file","apply_patch","notebook_edit","lsp","run_command","artifact_write","browser_capabilities","browser_list_tabs","browser_get_state","browser_snapshot","browser_screenshot","browser_console","browser_network","browser_find_element","browser_assert_visible","browser_open_tab","browser_navigate","browser_reload","browser_wait","browser_select_tab","browser_close_tab","browser_click","browser_hover","browser_type","browser_press","browser_scroll","browser_set_device","[Truncated]"]

Data visualization

Overview

Turn a defined question and reviewed dataset into an honest visual explanation. Choose encodings from the analytical relationship, preserve uncertainty and provenance, and verify both static and interactive outputs for comprehension, accessibility, and rendering defects.

When to Use

Use for charts, dashboards, analytical figures, visual reports, exploratory views, interactive data stories, or audits of a potentially misleading graphic.

Workflow

1. State the audience, decision, comparison, unit of analysis, and the single question each view must answer. Separate exploratory views from publication or decision artifacts.
2. Verify source lineage, filters, joins, missingness, denominators, units, time zones, aggregation level, and uncertainty before choosing a chart.
3. Select the simplest encoding that preserves the relationship: position and length before area or angle; aligned scales for comparison; small multiples before overloaded color, animation, dual axes, or 3D.
4. Make scale choices explicit. Use truthful baselines, label log or truncated axes, disclose normalization and binning, preserve zero where magnitude is the claim, and never hide unfavorable or missing observations.
5. Design hierarchy, annotations, legends, labels, and color for the audience. Provide non-color distinctions, readable contrast, keyboard interaction where applicable, a text alternative, and a tabular fallback for material values.
6. Implement reproducibly from reviewed inputs. Keep transformations and chart specifications inspectable; bind filters and tooltips to the displayed population and units.
7. Verify totals and representative marks against source data, inspect empty, loading, error, narrow, dense, and extreme-value states, and review console, network, keyboard, responsive, and screenshot evidence for interactive work.
8. Deliver the visual artifact with its question, data source, transformation summary, uncertainty, limitations, accessibility alternative, and the conclusion the graphic does and does not support.

Verification

• Every displayed value, denominator, filter, unit, and scale is auditable.
• The encoding matches the analytical question without visual distortion.
• Uncertainty, missing data, and material limitations remain visible.
• Color, keyboard, text alternatives, and narrow layouts are usable.
• Interactive states and representative plotted values were verified.

Red Flags

• Choosing a chart library or decorative form before defining the question.
• Truncated axes, dual scales, 3D, area, or animation that exaggerate a result.
• Dashboards with many metrics but no decision hierarchy.
• Treating a visually striking association as causal evidence.
• Shipping a screenshot without checking data correctness or interaction states.