feat(tui): render inline images for ReadMediaFile tool results

[linxinhong]

Problem

 ReadMediaFile tool results containing `image_url` ContentPart are currently rendered as plain text metadata

(image · 45.2 KB) in the TUI transcript. Users cannot visually inspect images without switching to the vis
web interface or running an external viewer.

 ## What changed

 Enhanced the `readMediaSummary` renderer in `media.ts` to detect `image_url` data URLs and, on terminals that

support the Kitty or iTerm2 inline graphics protocol, render the actual image using pi-tui's Image component.

 - Parse and preserve the base64 payload from `image_url` data URLs.
 - Extract original pixel dimensions from ReadMediaFile's `original size NxMpx` text when available.
 - Use `getCapabilities()` to gate image rendering: only active on `kitty` / `iterm2` terminals.
 - Fall back to the existing text summary on unsupported terminals.
 - Image display is capped at 12 rows × 60 columns to avoid monopolizing the viewport.

 ## Checklist

 - [x] I have read the CONTRIBUTING document.
 - [x] Ran `pnpm typecheck` — passes.
 - [x] Changeset included.

[changeset-bot]

🦋 Changeset detected

Latest commit: 38142f3

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@moonshot-ai/kimi-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 38142f3346

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[chatgpt-codex-connector]

Avoid sending huge images to the terminal

When a supported Kitty/iTerm2 terminal expands a large ReadMediaFile image, this path passes the full data-URL payload directly into the inline image component even though ReadMediaFile permits files up to 100 MB (packages/agent-core/src/tools/builtin/file/read-media.ts). The maxHeightCells/maxWidthCells options cap the displayed cell size, but they do not reduce the base64 payload being emitted, so expanding a large screenshot/photo can push tens or hundreds of MB of escape-sequence data through the TUI and make it appear hung; gate inline rendering by summary.bytes or generate a smaller thumbnail before constructing Image.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Parse the dimensions format emitted by ReadMediaFile

For real ReadMediaFile results with known image dimensions, core emits the leading system text as Original dimensions: WxH pixels. (packages/agent-core/src/tools/builtin/file/read-media.ts), but this new inline-image path only looks for the older WxHpx summary format before passing dimensions to Image. That means the renderer almost always constructs the image without the original pixel size, so terminals cannot reliably reserve/scale the image with the intended aspect ratio; accept the current Original dimensions wording or source the dimensions from the actual output format before rendering.

Useful? React with 👍 / 👎.