[Feature] Console Go (OpenCode Go): pass through image content parts on the Anthropic-compatible endpoint (/v1/messages)

Summary

The OpenCode Go ("Console Go") Anthropic-compatible endpoint (/v1/messages) rejects any request containing a standard Anthropic image content block with HTTP 400 and a raw serde error — even though the backend model (e.g. deepseek-v4-flash / deepseek-v4-pro) does support image input on DeepSeek's official API. The gateway parses content blocks as an enum that only knows text, so image_url/image parts are rejected before they ever reach the model.

Client path

Claude Code (Anthropic wire format) → cc-switch local proxy → Console Go /v1/messages

Error returned

{"error":{"type":"invalid_request_error","message":"Error from provider (Console Go): Upstream request failed: [invalid_request_error] Failed to deserialize the JSON body into the target type: messages[345]: unknown variant `image_url`, expected `text` at line 1 column 745497"}}

The identical request with text-only content succeeds. The failure is in the gateway's request-schema parsing (serde), before the model is reached.

How this differs from #40811

#40811 covers the OpenAI-compatible /chat/completions path (third-party OpenAI-compatible clients like Codex CLI). This is the Anthropic-compatible /v1/messages path used by Claude Code — same root cause (gateway enum has no image variant), different endpoint and wire format. Both should be fixed together.

Expected behavior

• Accept and pass through standard Anthropic image content blocks ({"type":"image","source":{"type":"base64","media_type":"...","data":"..."}}) to the backend model on /v1/messages — since the routed models' official APIs (DeepSeek) already tolerate/accept images, the gateway should too (pass-through or ignore, per DeepSeek docs); OR
• At minimum, do not reject the entire request because one message contains an image part; return a proper OpenAI-style error envelope instead of echoing a serde error that includes the full request body.

Related

• #40811 — same unknown variant 'image_url', expected 'text' on the /chat/completions path (open)
• #40632 — deepseek-v4-flash is marked text-only in models.dev but DeepSeek's official API supports image input
• #31870 / #33942 — related image-content rejection issues on other endpoints

Start at the Anthropic-compatible /v1/messages request parser and its content-block schema, then compare the related /chat/completions path described in #40811. Done means standard Anthropic image blocks no longer cause a raw serde rejection: they are passed through to supported models or produce a proper error envelope without exposing the full request body.