Expo App Assets

Create a coherent app-icon and splash-screen asset set, then configure it without damaging the project's existing Expo setup.

Inspect before proposing changes

Read the app config, Expo SDK, package.json, existing assets, and native directories. Determine whether the project uses app JSON, dynamic JS/TS config, Continuous Native Generation, or manually maintained native projects. Preserve dynamic config and existing plugins.

Ask only for missing brand decisions: app purpose, dominant symbol, primary/background colors, whether light/dark/tinted variants are wanted, and whether the splash uses the same mark. If the app already answers these questions, summarize the inferred direction instead. Do not make the user repeat information visible in the project.

If the user has not approved both a concept and a style, read references/icon-concept.md. Inspect the product and offer a diverse exploration board or one focused direction. Keep concept selection separate from rendering style and preserve explicit brand choices.

Before generating images or changing files, show no more than five short bullets covering the concept, planned assets, platform variants, config changes, and verification. Wait only when the concept or style is unapproved, generation was not requested, or a separate paid-provider charge may apply. Otherwise proceed from the user's existing approval without adding another gate.

Create and approve the source concept

Prefer an existing logo or user-supplied source when one exists. Otherwise propose an exploration board and let the user choose: generate several numbered concepts in one image when the user wants options, or generate one concept when the direction is already approved. A board is for selection only; regenerate the chosen concept as a clean standalone source instead of treating the board crop as the final asset.

When an image-capable model is available and the user requested generation, read references/icon-generation.md and use it directly with the inspected product as image-model context. Generate the image first; do not show, manufacture, or ask approval for an intermediate prompt. Briefly tell the user that using the skill directly in an image-capable model usually preserves more design context and produces better results than relaying a generated prompt.