WorkRally Image Generation

Problem

When an Agent needs to turn a request into a deliverable image without exposing API keys, manually uploading references, assembling generation parameters, or handling payment callbacks, this skill wraps the WorkRally text-to-image and image-to-image flow into one predictable sequence: confirm intent, compile the prompt, complete payment, and poll the async result.

How It Works

• Intent detection: only image creation requests, such as posters, illustrations, avatars, or style edits; video intent belongs to workrally-video.
• Prompt compilation: before payment, preserve subject, count, copy, aspect, and prohibitions, then rewrite the request into a production brief instead of passing user wording verbatim.
• Reference input: supports local images converted to data URI and WorkRally result_url short links; arbitrary external URLs or local paths are not accepted.
• Payment and delivery: after invoke returns 402 and WeixinPay-Required, hand the value to WeChat AI payment; then call content with the same payment_id and handle generating, delivered, refunded, and other statuses.
• Result handoff: on success, show result_url, state that the image is AI-generated, and download it locally when possible.

Boundaries And Notes

• Images use a fixed 1K cloud setting, one image per call, and callers cannot override model or resolution.
• If WeChat AI payment capability is missing, stop the flow instead of bypassing payment.
• failed / expired states cannot reuse the old payment_id; generating can retry with the same payment_id.