Per-chat reasoning effort in contextual sheet

[YoussefKeyrouz]

Task/Issue URL: https://app.asana.com/1/137249556945/project/1212810093780571/task/1215062229903137?focus=true

Description

Wires the contextual Duck.ai sheet into the per-chat reasoning flow introduced by #8666 and fixes two correctness bugs

• Minor cleanup in ReasoningModePickerViewModel

Steps to test this PR

Per-chat scope restored on reopen

• Start a chat in the contextual sheet on tab A; pick a non-default model and reasoning effort; submit a prompt.
• Close the contextual sheet.
• Open new duck ai on another tab or submit a prompt via the native input field using a new model (global default changes)
• Reopen the contextual sheet on tab A (within the session window). Verify the picker reflects the reasoning effort, not the global default)
• Submit another prompt; confirm that the submission carries the chat's modelId and reasoningEffort.
• Open another tab and start a new chat in the contextual sheet on tab B
• Switching between tabs properly restore the reasoning effort of the corresponding chat.

Globals untouched

• After any per-chat submission above, dismiss the sheet and open Duck.ai from the address bar (or a fresh tab). Verify your global defaults are unchanged.

---

Note

Medium Risk
Medium risk because it changes contextual sheet state management and URL/chat-id tracking, which can affect session restore behavior and what model/reasoning settings get applied to submissions.

Overview
Wires the contextual Duck.ai sheet to a per-chat identifier by extracting chatID from the webview URL and exposing it as a StateFlow on DuckChatContextualViewModel.

The contextual native input widget now receives this chatIdFlow (via ContextualNativeInputManager.init) and binds it into the native-input state, enabling chat-scoped reasoning/model selection to follow the active chat across sheet reopen/tab switches. The view model centralizes URL/chat-id updates via setSheetUrl/clearSheetUrl and adds tests to ensure chatId is set/cleared correctly; ReasoningModePickerViewModel is lightly refactored to use a single snapshot of modelState, with tests adjusted accordingly.

^{Reviewed by Cursor Bugbot for commit 666691b. Bugbot is set up for automated code reviews on this repo. Configure here.}

[YoussefKeyrouz]

• Per-chat reasoning effort in contextual sheet #8674 👈 (View in Graphite)
• Per-chat model and reasoning restoration on existing Duck.ai chats. #8666
• Publish active chatId to NativeInputState (per-tab) #8664
• develop

This stack of pull requests is managed by Graphite. Learn more about stacking.

[malmstein]

tested locally and confirmed it works as expected.

clean wire-up of contextual into the per-chat reasoning flow. nice that this PR also fixes the fullModeUrl correctness gap on the existing-chat-restore and new-chat-clear paths — easy to miss those were never being assigned before. the ReasoningModePickerViewModel.onModeTapped snapshot of modelState is a small but real win against the chance of a fetchModels emission landing mid-handler.

main note is the fullModeUrl / _chatId pair-assignment drift risk below. separately worth keeping on the radar: PR #8631 (still open) adds Uri.toChatIdOrNull(duckChat) in duckchat-api doing similar parsing — once it lands, both this PR's extractChatId and any other in-flight extractors could converge on the api extension. not blocking either way.

note for outcome tracking: the model-fallback asymmetry I flagged on #8666 (L161, getSelectedModelId vs getResolvedReasoningEffort when chat's model isn't in modelState) isn't touched here — that one's still pending on #8666's side.

[malmstein]

nit: fullModeUrl = ...; _chatId.value = extractChatId(...) is now repeated at L190-191, L243-244, and L289+292 (and a clearing pair at L649-650). a future addition that sets a new sheet URL but forgets the chatId update would silently break per-chat resolution. can we wrap this in a private helper like setSheetUrl(url: String?) that owns both fields, plus a clearSheetUrl() for the new-chat path? keeps the invariant in one place.

[YoussefKeyrouz]

Good catch. Addressed.

[cursor]

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

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

^{Reviewed by Cursor Bugbot for commit 9f586f9. Configure here.}