FE-1808: Separate stock and Brunch assistant presentations

[kostandinang]

🌟 What is the purpose of this PR?

The Brunch conversation stack (#9829) was meant to change only the Brunch assistant, but it also changed the stock assistant that every other host and npm consumer of @hashintel/petrinaut sees. Both presentations shared one large component, so Brunch changes leaked into stock.

This PR restores the stock assistant and gives each presentation its own components.

🔗 Related links

• FE-1808 (internal)
• The merged stack: #9829, #9856, #9857, #9836

🚫 Blocked by

Nothing. Targets main. #9886 is stacked on top.

🔍 What does this change?

• The stock assistant (presentation: "stock", the default) matches main from before FE-1793: Unify Chat and Voice conversation presentation #9829: message order, tool rows, styling, tabs, scrolling and composer. Two fixes stay: a tool stopped mid-run shows "Cancelled" instead of spinning, and reasoning stops looking live once its reply fails.
• The Brunch assistant is unchanged.
• The panel is split so neither presentation renders the other's components:

AiAssistantContents (container)
├── StockTranscript  ── StockToolList
├── BrunchTranscript ── BrunchToolList, work fold, voice input provenance
├── AiAssistantComposer
└── BrunchResponseStatus

shared/: tool model, interactive tool items, row styles

Pre-Merge Checklist 🚀

🚢 Has this modified a publishable library?

This PR:

• modifies an npm-publishable library and I have added a changeset file(s)

📜 Does this require a change to the docs?

The changes in this PR:

• are internal and do not require a docs change

🕸️ Does this require a change to the Turbo Graph?

The changes in this PR:

• do not affect the execution graph

⚠️ Known issues

• A few isBrunchChat checks remain in the container.

🐾 Next steps

• Move the remaining isBrunchChat checks out of the container.
• Stop ai-assistant-panel.tsx and the stories importing from the private ai-assistant-contents/ folder (they do on main too).

🛡 What tests cover this?

• The existing panel tests, now run once per presentation where the two differ.
• New tests for stock scrolling, stock message order and stock reasoning after an error.

❓ How to test this?

1. Open the Petrinaut website on this branch with the stock assistant selected, and open main from before FE-1793: Unify Chat and Voice conversation presentation #9829 (4561ee8e7b^) next to it.
2. Send a prompt that runs tools, then stop another prompt while a tool is running. Confirm the messages, tool rows, tabs, scrolling and composer match the old main, and the stopped tool shows "Cancelled".
3. Switch to the Brunch assistant and confirm it matches current main.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
hash	Ready	Preview	Oct 1, 2026 10:19am UTC
petrinaut	Ready	Preview	Oct 1, 2026 10:19am UTC
petrinaut-docs	Ready	Preview	Oct 1, 2026 10:19am UTC

1 Skipped Deployment

Project	Deployment	Actions	Updated
hashdotdesign-tokens	Ignored	Preview	Oct 1, 2026 10:19am UTC

[cursor]

PR Summary

Medium Risk
Touches the primary AI assistant UI and message rendering for all Petrinaut hosts; regressions could affect stock users or Brunch parity, though the change is largely a structural split with explicit presentation tests.

Overview
Splits the AI assistant panel so stock (default) and Brunch no longer share one transcript implementation, preventing Brunch conversation UI from leaking into npm hosts and the default assistant.

AiAssistantContents now routes to StockTranscript vs BrunchTranscript, with shared AiAssistantComposer, BrunchResponseStatus, and extracted shared helpers. Stock reads messages via getChronologicalRenderItems (stream order, including tool runs split at steps); Brunch keeps turn-based grouping. Stock-specific behavior is restored or clarified: chronological layout, stock tool-row chrome, tab styling without Brunch marks, auto-scroll that still follows new output when the reader has scrolled up (Brunch stops following past 96px), and stock reasoning labels (Thinking / heading in the trigger). Brunch-only tab marks and status row stay gated on presentation === "brunch".

Tests are reorganized: voice-dock coverage moves next to those components, and many assertions run as paired stock/brunch parameterised suites. Tab attention tests now look for aria-hidden markers instead of [data-attention]. Patch changeset notes that tools stopped mid-run show Cancelled instead of an endless spinner in stock.

^{Reviewed by Cursor Bugbot for commit 560d796. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit a822cee. Configure here.}