ADFA-6214 | Polish Chat UX, Input Scroll, Shortcuts, and Data-Loss Fix

[jatezzz]

Description

This PR addresses multiple UX pain points in the Agent chat interface to improve context visibility, interaction fluidity, and application reliability.

• Chat Header Context: The active chat title now dynamically displays in the header, truncating properly and showing a long-press tooltip. Titles are automatically generated by the LLM after the first reply, with a "Naming chat…" progress state shown during generation.
• Input Scrolling: Implemented PromptScrollGuard and PromptHeightCap to allow vertical scrolling inside a multiline chat input. This ensures the user can read long prompts without accidentally dragging the parent bottom sheet.
• Keyboard Shortcuts: Introduced a custom KeyboardShortcutHandler that allows sending messages via Shift + Enter on hardware keyboards, while explicitly preserving Enter for inserting newlines.
• Data-Loss Fix: The chat composer's sendPrompt logic now refuses the send and leaves the typed text intact if the API key is missing. Additionally, backend setup warnings (flagged via isSetupError) are automatically purged from the transcript once a valid API key is saved.
• Styling Polish: Added Material Design dividers to the Gemini and OpenAI settings panes. User messages longer than four lines are now collapsible via a toggle button, and feature an updated max-width layout constraint.

Details

• UI Updates: Added dynamic toolbar titles, fold/unfold toggles for user messages, and a new message copy action.
• Settings Styling: Included visual dividers (1dp height, @color/plugin_outline_variant) to the API settings XML layouts.
• Documentation: Updated the internal index.html guide to reflect the new shortcuts, collapsible messages, and automated chat naming behaviors.

DEMO:

https://drive.google.com/file/d/1V7wBrhiPVKrHgCcuPUDbV5aW_mWDBT7u/view?usp=sharing

Ticket

ADFA-6214

Observation

• The KeyboardShortcutHandler logic is completely decoupled from KeyEvent and Android views, making the shortcut rules fully unit-testable without a device runtime.
• Chat title generation runs asynchronously and includes a supersedeTitleRequest mechanism. If a user sends a new prompt rapidly, the system safely cancels the pending title generation so it doesn't hold up the next run.
• ChatViewModel manages the expanded/folded state of user bubbles so that message states survive view rebuilds.

[alome007]

IMPORTANT

• viewmodel/ChatViewModel.kt:1288 - title timeout cancels another plugin's generation
• fragments/ChatFragment.kt:352 - "Naming chat…" survives a rename during the first run

MINOR

• viewmodel/ChatViewModel.kt:1724 - cleared chat can still receive the old title (plausible, not reproduced)
• adapters/ChatAdapter.kt:282 - copy action on tool-activity rows
• plugin/AiCorePlugin.kt:268 - chat-title tooltip describes the pre-PR naming

Paths are under plugins/AI-Core/src/main/kotlin/com/itsaky/androidide/plugins/aicore/. Each finding was re-checked against 1f8fe70. This repo has no written approve/request-changes rule (CLAUDE.md covers the plugin-review skill only), so the default applied. Checked and clean: title settling across stop, session switch and retry; setup-notice clearing; the shortcut handler; the prompt height cap; removed resource and dependency references; Gemini/OpenAI pane styling.

[Daniel-ADFA]

IMPORTANT

• viewmodel/ChatViewModel.kt:1288 - title timeout cancels another plugin's generation
• fragments/ChatFragment.kt:352 - "Naming chat…" survives a rename during the first run

MINOR

• viewmodel/ChatViewModel.kt:1724 - cleared chat can still receive the old title (plausible, not reproduced)
• adapters/ChatAdapter.kt:282 - copy action on tool-activity rows
• plugin/AiCorePlugin.kt:268 - chat-title tooltip describes the pre-PR naming

Paths are under plugins/AI-Core/src/main/kotlin/com/itsaky/androidide/plugins/aicore/. Each finding was re-checked against 1f8fe70. This repo has no written approve/request-changes rule (CLAUDE.md covers the plugin-review skill only), so the default applied. Checked and clean: title settling across stop, session switch and retry; setup-notice clearing; the shortcut handler; the prompt height cap; removed resource and dependency references; Gemini/OpenAI pane styling.

[Daniel-ADFA]

Round 2, verified against 1de3b03. Each prior finding was re-read at the new head, not taken from the replies.

IMPORTANT

• viewmodel/ChatViewModel.kt:1294 - title timeout no longer stops the backend request (reply in its thread)
• viewmodel/ChatViewModel.kt:1731 - Clear Chat strands another chat's "Naming chat…" (reply in its thread)

Re-check of round 1

• Title timeout cancels other plugins' work: fixed as reported (no global cancel), but see the new IMPORTANT in that thread.
• "Naming chat…" survives a rename: fixed. renameSession settles when the new name is non-empty (ChatViewModel.kt, if (trimmed != null) settleTitle).
• clearMessages() leaves the title job running: fixed for the current chat, but see the new IMPORTANT in that thread.
• Copy action on tool rows: fixed. copyable is USER || AGENT (ChatAdapter.kt:283).
• Chat-title tooltip: fixed. Summary and detail describe the backend-written title and the fallback to it.

Both threads with live issues are unresolved. The CHANGES_REQUESTED from round 1 still stands for them.

[Daniel-ADFA]

Round 3, verified against 9783fd8. Approving: what remains is MINOR.

MINOR

• viewmodel/ChatViewModel.kt:1733 - clearing one chat still cancels another chat's title through the global cancel
• viewmodel/ChatViewModel.kt:1253 - Stop on a prompt queued behind a title cancels the title

Both are graded MINOR, not IMPORTANT, because they end on this PR's own failure path: the placeholder settles, the chat keeps its prompt as its title, and nothing is stuck or lost. The one cost is that the chat is not re-asked until the process restarts. They share one cause: cancelGeneration() cannot tell a chat run from a title request, so a follow-up that gives titles their own cancel handle fixes both.

Re-check of round 2

• Title timeout left the backend running: fixed. future.cancelJobOnCancel(job) is now wired in Gemini generate/generateWithHistory, OpenAI generate/generateWithHistory, and local runGeneration.
• Clear Chat stranded another chat's "Naming chat…": fixed. clearMessages() cancels and nulls titleRequest only for its own session, so the other chat's finally settles its placeholder.