Chanling Global Model Gateway

Problem

When integrating image generation into an agent workflow, the issue is often not whether a single request can be sent, but how endpoints, model IDs, task polling, result URLs, and error codes are kept consistent. Ad hoc curl calls or one-off scripts can mix synchronous and asynchronous interfaces, or miss consistent handling for 401, 402, and 429 responses.

How It Works

This skill codifies the xdhdancer.top platform conventions into a Markdown entry point:

• Shared auth: all sub-skills use one Bearer Token, configured as xdhdancer_api_key.
• Intent routing: the top-level entry selects a sub-skill by user intent; chanling-image currently covers text-to-image and image-to-image.
• Async tasks: image generation submits to POST /v1/videos, then polls by task_id, accepting either data.task_id or data.id.
• Result reading: there is no synchronous image endpoint, and result URL fields may differ across sub-skills, so follow the documented fallback chain.
• Error handling: rules cover 401, 402, 404, 400, 429, and 5xx, including token checks, balance guidance, model-ID validation, parameter checks, backoff, and retries.
• Model selection: it defaults to the recommended model, requires user approval before switching, and avoids presenting specific prices.

Scope And Caveats

The skill is tightly bound to https://www.xdhdancer.top. Model IDs, task protocol, and helper endpoints depend on that platform, so changing the base URL, token name, or model IDs will break existing calls. The current focus is chanling-image; future video, music, and text sub-skills should be checked in their respective docs.