[ybai08]

What

The thinking trace in reply-thinking.tsx renders as markdown: headings, lists, emphasis and fenced code inside the existing box. It uses Streamdown with the shared streamdownPlugins (code and math), linkSafety on, and the answer's restrictions from message.tsx: http, https and mailto links only, no images, no data images. Default mode, since a trace streams in. whitespace-pre-wrap is dropped from the viewport class; markdown handles line breaks.

One thing worth knowing: the plain Streamdown component has no security prop; that belongs to the assistant-ui primitive, which turns it into a rehype plugin list. So the trace passes rehypePlugins built from Streamdown's exported defaultRehypePlugins: its raw and sanitize passes as they are, and its harden pass re-run with the answer's options. Same passes, same order, stricter limits.

The box keeps everything it had: the fold, the fade, following the newest line, role="region", its accessible name and tabIndex={0}. The trace still rides in metadata.custom, so Copy still takes the answer alone.

docs/architecture/chat.md: the Known gaps line is cut down to the thinking switch, since #1950 has not landed, and the trace paragraph says it renders as markdown with the answer's restrictions.

Why

The answer directly below the trace is rendered, so a trace full of literal ## and ** read as broken rather than plain. Now the same model's output gets the same treatment on both sides of the fold.

Fixes #1949

How to test

cd surfsense_local/frontend
pnpm test
cd ../.. && python scripts/check_docs.py

Two new tests in reply-thinking.test.tsx, written first: a trace with a heading, a list, emphasis and a code fence renders as elements with no literal syntax left, and a javascript: link, a data-URI image and an https link come out as blocked text, blocked image and the one link. The two scroll-following tests now find the viewport by its region role, since the text no longer sits directly in it. Locally: chat feature 27/27, whole frontend suite green, eslint, prettier --check, pnpm typecheck and check_docs.py clean.

High-level PR Summary

This PR enhances the thinking trace display in the chat interface by rendering it as markdown instead of plain text. The trace now supports headings, lists, emphasis, and code blocks using the Streamdown component with the same security restrictions as the answer (allowing only http, https, and mailto links, no images). The implementation reuses the shared streamdownPlugins and carefully reconstructs the rehype plugin chain to apply the answer's stricter link and image policies while maintaining all existing functionality like scroll-following, folding, and accessibility features.

⏱️ Estimated Review Time: 5-15 minutes

💡 Review Order Suggestion

Order	File Path
1	surfsense_local/frontend/src/features/chat/reply-thinking.test.tsx
2	surfsense_local/frontend/src/features/chat/reply-thinking.tsx
3	docs/architecture/chat.md

[vercel]

@ybai08 is attempting to deploy a commit to the Rohan Verma's projects Team on Vercel.

A member of the Team first needs to authorize it.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on base/target branches other than the default branch.

Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 6d04d083-a247-433a-8205-26a6393e9850

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}