claw-code 대조 기반 AX Agent 품질 향상 3트랙 계획을 수립한다

- 사용자 체감 UI/UX, LLM·작업 처리, 유지보수·추가기능 구조의 3트랙으로 남은 품질 향상 과제를 재정리했다.

- docs/claw-code-parity-plan.md에 참조 파일, AX 적용 위치, 완료 조건, 품질 기준, 권장 실행 순서를 고정했다.

- README와 DEVELOPMENT 문서에 2026-04-06 09:27 (KST) 기준 계획 변경 이력을 반영했다.

- 이번 변경은 문서화 작업으로 코드 변경이 없어 별도 빌드는 생략했다.

docs/claw-code-parity-plan.md

@@ -218,6 +218,97 @@
 - Remaining quality target:
   - move more tool-result and permission-result presentation into smaller message-type-specific helpers, closer to `claw-code` component separation
 
+## Focused Quality Tracks
+- Updated: 2026-04-06 09:27 (KST)
+- The remaining improvement work should now be managed in three parallel tracks so UX polish, runtime quality, and maintainability do not get mixed together.
+
+### Track 1. User-Facing UI/UX Quality
+- Reference:
+  - `src/components/Messages.tsx`
+  - `src/components/MessageRow.tsx`
+  - `src/components/StatusLine.tsx`
+  - `src/components/PromptInput/PromptInput.tsx`
+  - `src/components/PromptInput/PromptInputFooter.tsx`
+  - `src/components/SessionPreview.tsx`
+- AX apply target:
+  - `src/AxCopilot/Views/ChatWindow.xaml`
+  - `src/AxCopilot/Views/ChatWindow.MessageInteractions.cs`
+  - `src/AxCopilot/Views/ChatWindow.StatusPresentation.cs`
+  - `src/AxCopilot/Views/ChatWindow.FooterPresentation.cs`
+  - `src/AxCopilot/Views/ChatWindow.PreviewPresentation.cs`
+- Focus:
+  - keep transcript text dominant and metadata secondary
+  - make footer controls read as a task bar, not a settings strip
+  - unify preview surfaces, chooser popups, and approval cards under one visual language
+  - reduce visual noise from queue/status/diagnostic surfaces unless the state is actionable
+- Completion criteria:
+  - message rows, footer, preview, and inline approval/question cards feel visually coherent
+  - chooser popups share the same spacing, hover behavior, and summary-row structure
+  - footer/status elements appear only when they convey useful state
+- Quality criteria:
+  - a user can understand “what is happening now” from the transcript and footer without opening extra panels
+  - the interface remains readable under narrow widths without text clipping or layout jitter
+
+### Track 2. LLM / Task Handling Quality
+- Reference:
+  - `src/bootstrap/state.ts`
+  - `src/bridge/initReplBridge.ts`
+  - `src/bridge/sessionRunner.ts`
+  - `src/components/messages/AssistantToolUseMessage.tsx`
+  - `src/components/messages/PlanApprovalMessage.tsx`
+  - `src/components/permissions/*`
+- AX apply target:
+  - `src/AxCopilot/Services/Agent/AxAgentExecutionEngine.cs`
+  - `src/AxCopilot/Services/Agent/AgentLoopService.cs`
+  - `src/AxCopilot/Services/Agent/AgentTranscriptDisplayCatalog.cs`
+  - `src/AxCopilot/Services/Agent/PermissionRequestPresentationCatalog.cs`
+  - `src/AxCopilot/Services/Agent/ToolResultPresentationCatalog.cs`
+  - `src/AxCopilot/Views/ChatWindow.InlineInteractions.cs`
+  - `src/AxCopilot/Views/ChatWindow.TranscriptPolicy.cs`
+- Focus:
+  - keep all entry routes (`send`, `retry`, `regenerate`, `queue`, `slash`) on the same prepared execution path
+  - distinguish `success / error / reject / cancel / needs-approval` tool results more clearly
+  - keep plan approval and user-question flows transcript-native by default
+  - minimize mismatches between execution state and what the user sees in the timeline
+- Completion criteria:
+  - plan/permission/tool-result/question events all have consistent transcript-native lifecycles
+  - reopen/retry/queue/compact flows preserve the same visible runtime state
+  - tool failures and permission rejections are clearly distinguishable in transcript rendering
+- Quality criteria:
+  - the same prompt under the same tab/settings uses the same execution route
+  - users can tell whether the agent succeeded, failed, was blocked, or is waiting for approval without reading raw diagnostics
+
+### Track 3. Maintainability / Extensibility Structure
+- Reference:
+  - `src/components/*` split by role in `claw-code`
+  - `src/components/messages/*`
+  - `src/components/permissions/*`
+  - `src/components/PromptInput/*`
+- AX apply target:
+  - `src/AxCopilot/Views/ChatWindow.xaml.cs`
+  - `src/AxCopilot/Views/ChatWindow.*.cs`
+  - `src/AxCopilot/Services/AppStateService.cs`
+  - `src/AxCopilot/Services/Agent/*.cs`
+- Focus:
+  - continue shrinking `ChatWindow.xaml.cs` toward orchestration-only responsibility
+  - keep renderer, popup, footer, message-interaction, status, and preview logic in dedicated partials
+  - centralize presentation rules in catalogs/models instead of scattered UI string/visibility branches
+  - prepare the codebase for new permission types, new tool classes, and new transcript card types without re-bloating the main window file
+- Completion criteria:
+  - `ChatWindow.xaml.cs` owns orchestration and runtime coordination more than direct UI element construction
+  - new message/permission/tool card types can be added via presentation catalogs or dedicated partials
+  - runtime summary and footer/status visibility derive from presentation models rather than ad-hoc branching
+- Quality criteria:
+  - adding a new tool-result or approval type should mostly affect one catalog/renderer area
+  - future UI polish work should land in dedicated presentation files rather than expanding the main window file again
+
+## Recommended Execution Order
+- Updated: 2026-04-06 09:27 (KST)
+1. Finish Track 2 consistency first whenever a UX issue is caused by runtime truth mismatch.
+2. Apply Track 1 visual cleanup only after the state/message lifecycle is stable for that surface.
+3. Fold each stable surface into Track 3 structure immediately so later changes do not reintroduce `ChatWindow.xaml.cs` sprawl.
+4. Keep validating against `docs/AX_AGENT_REGRESSION_PROMPTS.md` after each change set, especially for `plan / permission / queue / compact / reopen`.
+
 ## Current Snapshot
 - Updated: 2026-04-05 19:42 (KST)
 - Estimated parity: