Proposal: recoverable handling of unsupported tool-result image formats

What do you want to change?

Validate unsupported image MIME types before submitting tool-result images to a provider, with a model-visible text fallback instead of an invalid image block. Is this appropriate for core? I can contribute a narrow PR if approved; SVG rasterization can stay in the producing extension.

Why?

pi-image-gen-replicate returned image/svg+xml as image content. With Pi 1.4.0 and openai-codex/gpt-6.1-sol, the next request failed with Codex error: input_image image_url unsupported image format: unknown. Continuing failed again while the attachment remained in the active history; backing up past it restored the session. This is extension-triggered, not built-in read treating SVG as text. The producer is being fixed separately.

How? (optional)

Prefer an actionable text replacement identifying the unsupported format; no core SVG renderer is needed. Check request-bound images, including replayed history, so the model can recover. Minimal isolated reproduction and AI-assistance disclosure follow in a comment.