visualize

Visualize

Choose a visual that explains the user's question: a labelled comparison, timeline, diagram, simulation, or focused mockup. Prefer interaction only when it makes a meaningful variable or consequence easier to understand. State units, assumptions, and data sources; do not invent measurements or scores.

Create standalone HTML in the task workspace. Include its CSS, JavaScript, and data locally. Do not assume framework globals, external CDNs, host styles, or window.openai APIs. Use responsive layout, readable labels, keyboard-operable controls, visible focus, sufficient contrast, and a text equivalent of key data. Test initial state, control changes, reset when provided, small widths, and empty/boundary values. Do not use color as the only encoding.

Read ${SKILL_DIR}/references/unified.md when the host exposes app_control. Publish through its discovered canvas schema and check the returned result. Use persistent canvas apps for iterative shared-state work when available and simple HTML snapshots for a standalone visual. Keep user selections visible to the agent through the supported bridge; do not promise unsupported follow-up messages, persistence, or automatic turn initiation.

Without a canvas, return a link to the standalone HTML. Inspect in an available browser and verify behavior before reporting visual acceptance. File creation, publication, successful rendering, and correct interaction are separate claims.