Message image carousel (a run of images becomes one swipeable viewer)

What it is

What it is

When an AI reply includes several images in a row — for example an agent that made "5 versions" of a page and shows a full-page screenshot of each — Omniscio no longer stacks them into one tall column you scroll through. Instead, 3 or more consecutive images collapse into a single carousel: one image at a time, with a small count badge (like 3 / 12) in the corner.

• On a phone: swipe left/right through the images. The swipe stays inside the carousel — it will never flip you to the next/previous session the way a normal side-swipe does.
• On desktop: use the ‹ / › arrow buttons (the first/last arrow is greyed out at the ends).
• Either way: tap/click the current image to open it full-screen in the existing image viewer, exactly as before.

A run of 1 or 2 images is left exactly as it was (shown inline). Images that are broken up by text, or that sit inside a bulleted list, are also left alone — only a clean run of 3+ back-to-back images becomes a carousel.

Where to find it

In the chat, on an AI reply that shows a run of images. There is no setting — the carousel appears whenever a reply carries three or more images in a row.

How it behaves

Why it works this way

• The images sit in a fixed-height frame and are scaled to fit (centered, letterboxed if needed), so moving between a tall screenshot and a short one doesn't make the box jump around.
• The grouping is display-only — it changes how the message is drawn, not what's stored. Copying or exporting the message still gives you the original text and image links.
• There's no autoplay; the carousel only moves when you move it, and it respects your "reduce motion" setting.

Where the images come from

This works on both ways an AI reply can show images:

• Images the agent embeds in its message text (markdown images) — the large, roughly full-width ones.
• Images the agent attaches to the message — e.g. a tool render or a "here are the mockups" reply that comes back with a batch of screenshots. This is the common case, and it's what most "a dozen stacked images" replies actually are. 3 or more image attachments now collapse into the same carousel. (1–2 attached images stay as the small thumbnails, and any non-image attachment — a PDF or doc — stays a separate file chip you can open.)

Team Chat messages are unaffected (they use a separate renderer).

For agents

Under the hood (for agents with repo access)

• Grouping: a rehype plugin, rehype-group-images.ts, runs during markdown render. It finds a run of 3+ (CAROUSEL_MIN_IMAGES) consecutive root-level paragraphs that each hold one renderable image and replaces them with a single <amc-image-carousel> element. It's wired into the module-stable rehype arrays in AgentMarkdownImpl.tsx; the amc-image-carousel element is mapped in agent-markdown-components.tsx.
• Rendering: MessageImageCarousel.tsx — a native horizontal overflow-x scroll-snap track (each slide a MarkdownImage, so click-to-zoom is preserved), a count badge, and prev/next IconButtons. It stops touch-event propagation so the mobile session-swipe (useSwipeNavigation) never fires from inside it.
• Attachment path: MessageAttachments.tsx routes a message's image/* attachments through the SAME MessageImageCarousel at the same CAROUSEL_MIN_IMAGES threshold (each slide is the existing open-lightbox thumbnail button); non-image attachments stay FileText chips.
• Threshold: 3 is deliberate (owner decision), shared by both surfaces via one CAROUSEL_MIN_IMAGES constant. 1–2 images stay inline.
• Full invariants + the guarding tests: .claude/memory/contracts/message-image-carousel-contract.md.

Always on; no setting.

Related

• agent-message-display.md — how the rest of a reply renders around the images.
• agent-markdown-formatting.md — the other message shapes an agent can use, including showing a single image at a chosen width.
• mobile-swipe-dismiss.md — the phone gestures nearby, and why a swipe inside the carousel never flips your session.