v0.13.0 — data_visualization skill + render_chart (Vega-Lite)

Data visualization, the honest way. Studied plotly / bokeh / seaborn / altair / matplotlib — all are frameworks or JS-backed renderers an agent should call, not reimplement (bokeh is ~half TypeScript, plotly wraps plotly.js, matplotlib's renderer is C++, seaborn wraps matplotlib). Two complementary capabilities, neither a reimplementation:

Added

• data_visualization skill — sibling of data_analysis: writes a self-contained chart script for the execute_code sandbox (matplotlib/seaborn → static PNG/SVG, plotly → interactive HTML), with the headless-backend gotcha (matplotlib.use("Agg") before pyplot) and save-to-workspace-then-print-the-path discipline baked into its prompt. Covers arbitrary/custom charts.
• render_chart tool — Altair's insight applied to an agent: a Vega-Lite spec is inert, inspectable JSON data, not code — a stronger governance story than executing generated plotting code. Renders HTML with zero extra deps (embeds the spec + the Vega CDN); PNG/SVG via the optional viz-vega extra (vl-convert-python). Shape-validated before render.
• Optional extras: viz (matplotlib + seaborn + plotly) and viz-vega (vl-convert-python).

Honest scope

The code sandbox already imports matplotlib/plotly/seaborn — the skill just names the capability and handles the headless discipline; it does not vendor or reimplement any of them. Vega-Lite earns a dedicated tool only because its artifact is safe declarative data.

Gate: ruff clean, 1224 passed / 2 skipped, mypy clean.

PyPI: `pip install chimera-agent==0.13.0`