show per-message assistant identity from metadata instead of props

[morgmart]

Category: improvement
User Impact: Each assistant message now shows the correct provider name and icon for the model that actually generated it, even when switching providers mid-conversation.

Problem: Assistant identity (name, avatar, icon) was passed down as props from the chat view, meaning every message displayed the current agent's identity rather than the identity of the model that actually generated that message. Switching providers mid-conversation would retroactively relabel old messages.

Solution: Identity is now derived per-message from metadata (providerId, personaId) stamped at send time. A pendingAssistantProviderId in the chat runtime captures the provider before the backend confirms the message, ensuring the correct provider is recorded even if the session provider changes between send and message creation. The loading indicator was also simplified from animated icons to a shimmer-based "Thinking..." / "Responding..." status.

Note: Session replay will fallback to providerID at the session level, not per message. Following up with separate PRs for storing per message provider in backend.

File changes

src/shared/types/chat.ts
Added pendingAssistantProviderId to SessionChatRuntime so the provider chosen at send time is available before the backend creates the assistant message.

src/shared/types/messages.ts
Added providerId to MessageMetadata so each message records which provider generated it.

src/features/chat/stores/chatStore.ts
Added setPendingAssistantProvider action to set and clear the pending provider ID on the session runtime.

src/features/chat/hooks/acpStreamTypes.ts
Extracted ACP event payload type definitions from useAcpStream.ts to keep the hook file under the size limit.

src/features/chat/hooks/useAcpStream.ts
Stamps providerId on assistant messages at creation time using the pending provider, with a fallback to the session's provider. Clears the pending provider after message creation and on completion.

src/features/chat/hooks/useChat.ts
Sets pendingAssistantProviderId before sending, and clears it on error, stop, and reset paths.

src/features/chat/ui/ChatView.tsx
Removed agentName and agentAvatarUrl props that were being threaded through to child components. Wrapped the loading indicator in AnimatePresence for smooth enter/exit transitions.

src/features/chat/ui/LoadingGoose.tsx
Replaced animated icon indicators with a shimmer-based status. Simplified from four state-specific copy strings to two ("Thinking..." / "Responding..."). Removed the agentName prop since the indicator is now agent-agnostic. Added proper useReducedMotion support.

src/features/chat/ui/MessageBubble.tsx
Derives assistant identity (name, icon, avatar) from per-message metadata instead of props. Shows the provider icon and display name from the provider catalog, with persona name taking priority when present. Migrated the fallback Bot icon from lucide to @tabler/icons-react.

src/features/chat/ui/MessageTimeline.tsx
Removed agentName and agentAvatarUrl props. Added filtering for empty in-progress assistant shell messages.

src/shared/ui/ai-elements/shimmer.tsx
Added delay, repeatDelay, and tone props to support the loading indicator's softer shimmer variant.

src/shared/i18n/locales/en/chat.json
Replaced four loading state keys with two simplified ones: thinking and responding.

src/shared/i18n/locales/es/chat.json
Updated Spanish translations to match the simplified loading state keys.

scripts/check-file-sizes.mjs
Bumped the test file size limit for useAcpStream.test.ts to accommodate provider identity tests.

src/features/chat/hooks/tests/useAcpStream.test.ts
Added tests verifying that providerId is stamped on created messages and that the pending provider is preserved when the session provider changes before message creation.

src/features/chat/ui/tests/LoadingGoose.test.tsx
New test file covering thinking/responding copy, idle state rendering, with assertions driven from translation JSON.

src/features/chat/ui/tests/MessageBubble.test.tsx
Updated tests to verify provider-based identity rendering, persona name priority, and empty in-progress message handling.

Reproduction Steps

1. Run just dev to start the app.
2. Start a chat with any provider (e.g., Claude). Send a message and observe the assistant response shows the provider icon and name (e.g., the Claude icon + "Claude Code").
3. Switch to a different provider mid-conversation and send another message. Verify the new response shows the new provider's identity, while previous messages retain their original provider identity.
4. Create an agent with a persona. Chat with it and verify the persona name and avatar take priority over the provider label.
5. Observe the loading indicator shows "Thinking..." initially, then "Responding..." during streaming — no agent name prefix.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: d511c7e21a

ℹ️ About Codex in GitHub

Your team has set up Codex to 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 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".