Purpose
The plugin connects Codex OAuth text, vision input, image generation, and reference-image editing to AstrBot’s Provider Source interface so higher-level features can select and call them through one provider shape.
Structure
The project contains authorization-state handling, model capability descriptions, text-response adaptation, streamed image-response parsing, reference-image input conversion, error normalization, and timeout control. The image execution interface is separated from its concrete backend so implementations can be replaced without changing the caller contract.
How it works
Each request is converted to a common input according to its capability, then passed to a text or image executor. The adapter converts different response forms into one result type. Refresh failures, remote rejection, malformed responses, and timeouts become stable error categories. Every call owns a request-level deadline; any retry stays within that request and its total time budget.
Current state
The project is a public experiment. The current implementation covers text and image capabilities, replaceable image backends, normalized errors, request-level timeouts, and regression tests for streamed responses and reference-image handling.