Merge remote-tracking branch 'upstream/main' into sbc_add_subtasks

.changeset/chilly-bugs-pay.md

@@ -1,5 +0,0 @@
----
-"roo-cline": patch
----
-
-Delete task confirmation enhancements

.changeset/tasty-grapes-suffer.md

@@ -1,5 +0,0 @@
----
-"roo-cline": patch
----
-
-Fix maxTokens defaults for Claude 3.7 Sonnet models

.changeset/young-hornets-taste.md

@@ -1,5 +0,0 @@
----
-"roo-cline": patch
----
-
-Prettier thinking blocks

.github/workflows/marketplace-publish.yml

@@ -29,10 +29,7 @@ jobs:
       - name: Install Dependencies
         run: |
           npm install -g vsce ovsx
-          npm install
-          cd webview-ui
-          npm install
-          cd ..
+          npm run install:ci
       - name: Package and Publish Extension
         env:
           VSCE_PAT: ${{ secrets.VSCE_PAT }}

CHANGELOG.md

@@ -1,5 +1,15 @@
 # Roo Code Changelog
 
+## [3.7.9]
+
+- Delete task confirmation enhancements
+- Smarter context window management
+- Prettier thinking blocks
+- Fix maxTokens defaults for Claude 3.7 Sonnet models
+- Terminal output parsing improvements (thanks @KJ7LNW!)
+- UI fix to dropdown hover colors (thanks @SamirSaji!)
+- Add support for Claude Sonnet 3.7 thinking via Vertex AI (thanks @lupuletic!)
+
 ## [3.7.8]
 
 - Add Vertex AI prompt caching support for Claude models (thanks @aitoroses and @lupuletic!)

package-lock.json

@@ -1,12 +1,12 @@
 {
 	"name": "roo-cline",
-	"version": "3.7.8",
+	"version": "3.7.9",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "roo-cline",
-			"version": "3.7.8",
+			"version": "3.7.9",
 			"dependencies": {
 				"@anthropic-ai/bedrock-sdk": "^0.10.2",
 				"@anthropic-ai/sdk": "^0.37.0",

package.json

@@ -3,7 +3,7 @@
 	"displayName": "Roo Code (prev. Roo Cline)",
 	"description": "A whole dev team of AI agents in your editor.",
 	"publisher": "RooVeterinaryInc",
-	"version": "3.7.8",
+	"version": "3.7.9",
 	"icon": "assets/icons/rocket.png",
 	"galleryBanner": {
 		"color": "#617A91",
@@ -334,6 +334,7 @@
 		"get-folder-size": "^5.0.0",
 		"globby": "^14.0.2",
 		"isbinaryfile": "^5.0.2",
+		"js-tiktoken": "^1.0.19",
 		"mammoth": "^1.8.0",
 		"monaco-vscode-textmate-theme-converter": "^0.1.7",
 		"openai": "^4.78.1",
@@ -348,7 +349,6 @@
 		"sound-play": "^1.1.0",
 		"string-similarity": "^4.0.4",
 		"strip-ansi": "^7.1.0",
-		"js-tiktoken": "^1.0.19",
 		"tmp": "^0.2.3",
 		"tree-sitter-wasms": "^0.1.11",
 		"turndown": "^7.2.0",

src/core/sliding-window/__tests__/sliding-window.test.ts

@@ -3,7 +3,12 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 
 import { ModelInfo } from "../../../shared/api"
-import { estimateTokenCount, truncateConversation, truncateConversationIfNeeded } from "../index"
+import {
+	TOKEN_BUFFER_PERCENTAGE,
+	estimateTokenCount,
+	truncateConversation,
+	truncateConversationIfNeeded,
+} from "../index"
 
 /**
  * Tests for the truncateConversation function
@@ -121,10 +126,11 @@ describe("getMaxTokens", () => {
 		// Create messages with very small content in the last one to avoid token overflow
 		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
 
-		// Below max tokens - no truncation
+		// Account for the dynamic buffer which is 10% of context window (10,000 tokens)
+		// Below max tokens and buffer - no truncation
 		const result1 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 49999,
+			totalTokens: 39999, // Well below threshold + dynamic buffer
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -133,7 +139,7 @@ describe("getMaxTokens", () => {
 		// Above max tokens - truncate
 		const result2 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 50001,
+			totalTokens: 50001, // Above threshold
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -148,10 +154,11 @@ describe("getMaxTokens", () => {
 		// Create messages with very small content in the last one to avoid token overflow
 		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
 
-		// Below max tokens - no truncation
+		// Account for the dynamic buffer which is 10% of context window (10,000 tokens)
+		// Below max tokens and buffer - no truncation
 		const result1 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 79999,
+			totalTokens: 69999, // Well below threshold + dynamic buffer
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -160,7 +167,7 @@ describe("getMaxTokens", () => {
 		// Above max tokens - truncate
 		const result2 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 80001,
+			totalTokens: 80001, // Above threshold
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -175,10 +182,10 @@ describe("getMaxTokens", () => {
 		// Create messages with very small content in the last one to avoid token overflow
 		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
 
-		// Below max tokens - no truncation
+		// Below max tokens and buffer - no truncation
 		const result1 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 39999,
+			totalTokens: 34999, // Well below threshold + buffer
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -187,7 +194,7 @@ describe("getMaxTokens", () => {
 		// Above max tokens - truncate
 		const result2 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 40001,
+			totalTokens: 40001, // Above threshold
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -202,10 +209,11 @@ describe("getMaxTokens", () => {
 		// Create messages with very small content in the last one to avoid token overflow
 		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
 
-		// Below max tokens - no truncation
+		// Account for the dynamic buffer which is 10% of context window (20,000 tokens for this test)
+		// Below max tokens and buffer - no truncation
 		const result1 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 169999,
+			totalTokens: 149999, // Well below threshold + dynamic buffer
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -214,7 +222,7 @@ describe("getMaxTokens", () => {
 		// Above max tokens - truncate
 		const result2 = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
-			totalTokens: 170001,
+			totalTokens: 170001, // Above threshold
 			contextWindow: modelInfo.contextWindow,
 			maxTokens: modelInfo.maxTokens,
 		})
@@ -244,7 +252,8 @@ describe("truncateConversationIfNeeded", () => {
 	it("should not truncate if tokens are below max tokens threshold", () => {
 		const modelInfo = createModelInfo(100000, true, 30000)
 		const maxTokens = 100000 - 30000 // 70000
-		const totalTokens = 69999 // Below threshold
+		const dynamicBuffer = modelInfo.contextWindow * TOKEN_BUFFER_PERCENTAGE // 10000
+		const totalTokens = 70000 - dynamicBuffer - 1 // Just below threshold - buffer
 
 		// Create messages with very small content in the last one to avoid token overflow
 		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
@@ -337,8 +346,9 @@ describe("truncateConversationIfNeeded", () => {
 			{ role: messages[messages.length - 1].role, content: smallContent },
 		]
 
-		// Set base tokens so total is below threshold even with small content added
-		const baseTokensForSmall = availableTokens - smallContentTokens - 10
+		// Set base tokens so total is well below threshold + buffer even with small content added
+		const dynamicBuffer = modelInfo.contextWindow * TOKEN_BUFFER_PERCENTAGE
+		const baseTokensForSmall = availableTokens - smallContentTokens - dynamicBuffer - 10
 		const resultWithSmall = truncateConversationIfNeeded({
 			messages: messagesWithSmallContent,
 			totalTokens: baseTokensForSmall,
@@ -388,7 +398,30 @@ describe("truncateConversationIfNeeded", () => {
 		})
 		expect(resultWithVeryLarge).not.toEqual(messagesWithVeryLargeContent) // Should truncate
 	})
+
+	it("should truncate if tokens are within TOKEN_BUFFER_PERCENTAGE of the threshold", () => {
+		const modelInfo = createModelInfo(100000, true, 30000)
+		const maxTokens = 100000 - 30000 // 70000
+		const dynamicBuffer = modelInfo.contextWindow * TOKEN_BUFFER_PERCENTAGE // 10% of 100000 = 10000
+		const totalTokens = 70000 - dynamicBuffer + 1 // Just within the dynamic buffer of threshold (70000)
+
+		// Create messages with very small content in the last one to avoid token overflow
+		const messagesWithSmallContent = [...messages.slice(0, -1), { ...messages[messages.length - 1], content: "" }]
+
+		// When truncating, always uses 0.5 fraction
+		// With 4 messages after the first, 0.5 fraction means remove 2 messages
+		const expectedResult = [messagesWithSmallContent[0], messagesWithSmallContent[3], messagesWithSmallContent[4]]
+
+		const result = truncateConversationIfNeeded({
+			messages: messagesWithSmallContent,
+			totalTokens,
+			contextWindow: modelInfo.contextWindow,
+			maxTokens: modelInfo.maxTokens,
+		})
+		expect(result).toEqual(expectedResult)
+	})
 })
+
 /**
  * Tests for the estimateTokenCount function
  */

src/core/sliding-window/index.ts

@@ -3,7 +3,11 @@ import { Anthropic } from "@anthropic-ai/sdk"
 import { Tiktoken } from "js-tiktoken/lite"
 import o200kBase from "js-tiktoken/ranks/o200k_base"
 
-const TOKEN_FUDGE_FACTOR = 1.5
+export const TOKEN_FUDGE_FACTOR = 1.5
+/**
+ * Default percentage of the context window to use as a buffer when deciding when to truncate
+ */
+export const TOKEN_BUFFER_PERCENTAGE = 0.1
 
 /**
  * Counts tokens for user content using tiktoken for text
@@ -107,8 +111,9 @@ export function truncateConversationIfNeeded({
 	const effectiveTokens = totalTokens + lastMessageTokens
 
 	// Calculate available tokens for conversation history
-	const allowedTokens = contextWindow - reservedTokens
+	// Truncate if we're within TOKEN_BUFFER_PERCENTAGE of the context window
+	const allowedTokens = contextWindow * (1 - TOKEN_BUFFER_PERCENTAGE) - reservedTokens
 
 	// Determine if truncation is needed and apply if necessary
-	return effectiveTokens < allowedTokens ? messages : truncateConversation(messages, 0.5)
+	return effectiveTokens > allowedTokens ? truncateConversation(messages, 0.5) : messages
 }