Release/v6.8.4

[code-crusher]

No description provided.

[matterai-app]

Summary By MatterAI ### 🔄 What Changed This pull request (Release/v6.8.4) updates the default OSS model to deepseek/deepseek-v4-flash-0731, adds dynamic catalog registration tests, improves assistant message streaming parser robustness for missing tool IDs, fixes parallel read-only batch scheduling issues, and introduces new model usage settings and UI components. ### 🔍 Impact of the Change Enhances model selection reliability, prevents task loops by ensuring proper execution of trailing non-parallelizable tools, and provides users with granular visibility into plan and model usage windows. ### 📁 Total Files Changed

Click to Expand

File	ChangeLog
Schemas Config (apps/web-evals/src/lib/schemas.ts)	Updated MODEL_DEFAULT to DeepSeek v4 Flash.
Persistence Test (cli/src/config/__tests__/persistence-provider-merge.test.ts)	Added test assertion for kilocodeModel persistence.
CLI Defaults (cli/src/config/defaults.ts)	Updated default kilocodeModel configuration.
Model Constants (cli/src/constants/providers/models.ts)	Commented out alphabetized sorting on rest model IDs.
Provider Settings (cli/src/constants/providers/settings.ts)	Updated default provider settings and PROVIDER_DEFAULT_MODELS.
Browser Auth (cli/src/utils/browserAuth.ts)	Configured default kilocodeModel in OAuth persistence.
Model Types (packages/types/src/model.ts)	Added optional iconUrl property to model schema.
OpenRouter Types (packages/types/src/providers/openrouter.ts)	Updated default OpenRouter model ID and model info.
Kilocode Models Test (src/api/providers/__tests__/kilocode-models.spec.ts)	Added comprehensive test suite for dynamic model registration and pricing.
Kilocode OpenRouter Test (src/api/providers/__tests__/kilocode-openrouter.spec.ts)	Updated test cases for GLM 5.3 Flash model handler options.
Model Cache (src/api/providers/fetchers/modelCache.ts)	Enhanced fetch headers to include Authorization and Cache-Control.
OpenRouter Fetcher (src/api/providers/fetchers/openrouter.ts)	Updated dynamic model fetching and iconUrl parsing logic.
Kilocode Models (src/api/providers/kilocode-models.ts)	Refactored static catalog to support dynamic backend registration.
Assistant Parser (src/core/assistant-message/AssistantMessageParser.ts)	Synthesizes stable IDs for tool calls lacking IDs during streaming.
Parser Tests (src/core/assistant-message/__tests__/AssistantMessageParser.spec.ts)	Added test for tool calls missing ID deltas.
Presentation Tests (src/core/assistant-message/__tests__/presentAssistantMessage.spec.ts)	Added regression tests for trailing tool execution and failure recovery.
Native Tool Call (src/core/assistant-message/kilocode/native-tool-call.ts)	Added optional name property to function delta type.
Present Assistant Message (src/core/assistant-message/presentAssistantMessage.ts)	Fixed batch scheduler readiness flag to ensure trailing non-parallel tools execute.
Condense Tests (src/core/condense/__tests__/condense.spec.ts)	Updated summary prefix assertions.
Condense Index Tests (src/core/condense/__tests__/index.spec.ts)	Updated conversation summary expectations.
Condense Module (src/core/condense/index.ts)	Added SUMMARY_PREFIX context handoff constant.
Prompt Snapshots (src/core/prompts/__tests__/__snapshots__/)	Updated system and mode prompt snapshots with editing discipline guidelines.
System Prompts (src/core/prompts/system.ts)	Added prompt guidelines for early editing and iterative steps.
Task Manager (src/core/task/Task.ts)	Implemented context budget warning logic near the condense threshold.
Cline Provider (src/core/webview/ClineProvider.ts)	Added background model refresh intervals and stale model validation on startup.
Provider Tests (src/core/webview/__tests__/ClineProvider.spec.ts)	Seeded dynamic catalog in provider tests.
Webview Message Handler (src/core/webview/webviewMessageHandler.ts)	Added forceRefresh handling for model requests.
Package JSON (src/package.json)	Bumped version to 6.8.4.
Webview Messages (src/shared/WebviewMessage.ts)	Added AxonCodeModelUsage interface definition.
Shared API (src/shared/api.ts)	Added forceRefresh to common fetch parameters.
Chat Row (webview-ui/src/components/chat/ChatRow.tsx)	Adjusted border radius and overflow classes for code blocks.
Chat Tabs (webview-ui/src/components/chat/ChatTabs.tsx)	Refactored tab styling and active background colors.
Usage Dialog (webview-ui/src/components/chat/UsageDialog.tsx)	Added Model Usage section with progress bars and provider logos.
Chat Layout (webview-ui/src/components/chat/chatLayout.ts)	Added horizontal padding constant.
Sticky User Message (webview-ui/src/components/kilocode/StickyUserMessage.tsx)	Adjusted padding and margin styles.
Model Selector (webview-ui/src/components/kilocode/chat/ModelSelector.tsx)	Integrated provider logos and updated model refresh handling.
Provider Models Hook (webview-ui/src/components/kilocode/hooks/useProviderModels.ts)	Exposed refetchRouterModels.
Model Usage Settings (webview-ui/src/components/settings/ModelUsageSettings.tsx)	Created settings component for plan limits and per-model quotas.
Settings View (webview-ui/src/components/settings/SettingsView.tsx)	Added Model Usage settings tab.
Preferred Models Hook (webview-ui/src/components/ui/hooks/usePreferredModels.ts)	Commented out model ID sorting.
OpenRouter Providers Hook (webview-ui/src/components/ui/hooks/useOpenRouterModelProviders.ts)	Updated static fallback provider mapping.
Router Models Hook (webview-ui/src/components/ui/hooks/useRouterModels.ts)	Added window focus and visibility change listeners for query invalidation.
UI Index (webview-ui/src/components/ui/index.ts)	Exported ProviderLogo.
Provider Logo (webview-ui/src/components/ui/provider-logo.tsx)	Created wrapper component for provider SVG logos on white background.

### 🧪 Test Added/Recommended #### Added - Unit tests for dynamic KiloCode model catalog registration, pricing coercion, and sparse entry fallbacks in `src/api/providers/__tests__/kilocode-models.spec.ts`. - Regression tests for trailing non-parallelizable tool execution and failure recovery in `src/core/assistant-message/__tests__/presentAssistantMessage.spec.ts`. - Streaming parser test for tool calls lacking an explicit ID in `src/core/assistant-message/__tests__/AssistantMessageParser.spec.ts`. #### Recommended - Add integration tests for background model refresh intervals and focus-based cache invalidation in `ClineProvider`. ### 🔒Security Vulnerabilities N/A

[matterai-app]

🧪 PR Review is completed: Release v6.8.4 introduces dynamic model catalogs, context budget warnings, and model usage UI. Found a pricing/description mismatch on the new default model info, a redundant double-fetch on window focus in the webview router-models hook, debug console.log in Task.ts, and an as any bypassing the new response payload type. Reviewed src/api/providers/fetchers/modelCache.ts: no issues found (auth headers composed correctly, forceRefresh plumbed through). Reviewed src/api/providers/fetchers/openrouter.ts: no issues found (dynamic fetch is try/catch guarded; the ~55-line duplication between the two fetch paths is acceptable for now). Reviewed src/api/providers/kilocode-models.ts: no issues found (scientific-notation price coercion is documented as lossless via parseFloat). Reviewed src/core/assistant-message/AssistantMessageParser.ts: no issues found (synthesized tool-call id stays stable via the index→id map). Reviewed src/core/assistant-message/presentAssistantMessage.ts: no issues found (readiness gating and batch continuation are correct; cleanup of lock before recursion verified). Reviewed src/core/webview/ClineProvider.ts: no issues found (modelsRefreshInterval is unref'd and cleared on dispose; focus refresh is throttled). Reviewed src/core/webview/webviewMessageHandler.ts: no issues found. Reviewed src/core/condense/index.ts: no issues found (SUMMARY_PREFIX is exported and consumed consistently). Reviewed src/core/prompts/system.ts and snapshot files: intentional prompt updates, no issues. Reviewed webview-ui/src/components/kilocode/chat/ModelSelector.tsx: no issues found. Reviewed webview-ui/src/components/ui/hooks/useOpenRouterModelProviders.ts: no issues found (static catalog intentionally emptied; fallback provider is display-only). Reviewed test files (kilocode-models.spec.ts, kilocode-openrouter.spec.ts, AssistantMessageParser.spec.ts, presentAssistantMessage.spec.ts, ClineProvider.spec.ts, condense specs): no issues found.

Skipped files

• .changeset/axon-auto-default-model.md: Skipped file pattern
• .changeset/pasted-image-paths.md: Skipped file pattern
• .changeset/remove-fireworks-provider.md: Skipped file pattern
• CHANGELOG.md: Skipped file pattern
• README.md: Skipped file pattern
• webview-ui/src/i18n/locales/en/settings.json: Skipped file pattern

⬇️ Low Priority Suggestions (4)

packages/types/src/providers/openrouter.ts (1 suggestion)

Location: packages/types/src/providers/openrouter.ts (Lines 12-16)

🟡 Data Integrity / Business Logic

Issue: openRouterDefaultModelId is now "deepseek/deepseek-v4-flash-0731" (L4), but the accompanying openRouterDefaultModelInfo describes "GLM 5.3 Flash" and carries GLM 5.3 Flash pricing (input $0.15/M, output $0.50/M, cache read $0.03/M). Per the catalog in src/api/providers/kilocode-models.ts, DeepSeek V4 Flash is priced at $0.14/M input, $0.28/M output, $0.028/M cache read. Any cost/usage calculation or UI display that leans on this default ModelInfo will report the wrong model name and wrong prices for the default model.

Fix: Align the description and pricing fields with DeepSeek V4 Flash's published rates.

Impact: Correct cost estimation and model metadata for the default model.

-  	inputPrice: 0.00000015,
-  	outputPrice: 0.0000005,
-  	cacheWritesPrice: 0,
-  	cacheReadsPrice: 0.00000003,
-  	description: "GLM 5.3 Flash is a fast, low cost open model for everyday coding tasks.",
+  	inputPrice: 0.00000014,
+  	outputPrice: 0.00000028,
+  	cacheWritesPrice: 0,
+  	cacheReadsPrice: 0.000000028,
+  	description: "DeepSeek V4 Flash is a fast, low cost open model for low-effort day-to-day coding tasks.",

webview-ui/src/components/ui/hooks/useRouterModels.ts (1 suggestion)

Location: webview-ui/src/components/ui/hooks/useRouterModels.ts (Lines 83-88)

🟠 Performance

Issue: The new useQuery config sets refetchOnWindowFocus: true while the new custom focus listener in the same hook already calls queryClient.invalidateQueries({ queryKey: ["routerModels"] }) on the same window-focus event. Both fire per focus, so each focus triggers two requestRouterModels postMessage round-trips — and the extension host also force-refreshes models on window focus (ClineProvider.onDidChangeWindowState), making it up to three fetch cycles per focus. Note: in TanStack Query v5 refetchOnWindowFocus defaults to true, so simply deleting the line may not fix it.

Fix: Disable react-query's built-in focus refetch and keep the throttled custom listener (which also covers the didBecomeVisible message) as the single focus trigger.

Impact: Halves the webview-side fetch chatter on every window focus; reduces load on the OpenRouter/MatterAI model endpoints.

-  	return useQuery({
-  		queryKey: ["routerModels", queryKey],
-  		queryFn: () => getRouterModels(),
-  		refetchInterval: 10 * 60 * 1000,
-  		refetchOnWindowFocus: true,
-  	})
+  	return useQuery({
+  		queryKey: ["routerModels", queryKey],
+  		queryFn: () => getRouterModels(),
+  		refetchInterval: 10 * 60 * 1000,
+  		refetchOnWindowFocus: false,
+  	})

src/core/task/Task.ts (1 suggestion)

Location: src/core/task/Task.ts (Lines 3970-3973)

🔵 Logging

Issue: Debug console.log statement left in the new context-budget warning path. Task.ts routes diagnostics through its logging facilities; a raw console.log fires on every warning event in the extension host output.

Fix: Remove the console.log block; the <context_window_warning> text pushed to userMessageContent already carries the same information to the model/UI.

Impact: Cleaner extension host output; no behavioral change.

-  					console.log(
-  						`[Task#${this.taskId}] Context budget warning: ${contextPercent.toFixed(1)}% used ` +
-  							`(~${tokensRemaining} of ${contextWindow} tokens remain), condense threshold ${effectiveThreshold}%.`,
-  					)
+  

webview-ui/src/components/settings/ModelUsageSettings.tsx (1 suggestion)

Location: webview-ui/src/components/settings/ModelUsageSettings.tsx (Lines 111-111)

🔵 Type Safety

Issue: The profileDataResponse handler casts message.payload with as any, bypassing the ProfileDataResponsePayload interface that already exists in @roo/WebviewMessage (and is used by the backend response shape). This discards type checking on payload.success / payload.data in a brand-new file where the types are readily available.

Fix: Cast to the exported payload type instead. Also add ProfileDataResponsePayload to the existing type import on line 5.

Impact: Compile-time safety for the profile response contract; no runtime change.

-  				const payload = message.payload as any
+  			const payload = message.payload as ProfileDataResponsePayload | undefined