v3.12.3 Release Notes

v3.12.3 Release Notes

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions <github-actions@github.com>
Co-authored-by: Dennis Bartlett <bartlett.dc.1@gmail.com>

.changeset/brown-fans-hammer.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-context management for file reads

.changeset/famous-snakes-deliver.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Fix auto approve state out of sync

.changeset/few-seals-grab.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-TS errors

.changeset/fluffy-toys-punch.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Reduce system prompt size by letting Cline dynamically load MCP documentation only when needed

.changeset/hungry-buckets-serve.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Feature to open basic settings & scroll a section into view with a highlight animation

.changeset/many-keys-pay.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Disable notifications in browser

.changeset/mean-pears-eat.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-Add modal for toggling MCP servers to the chat area

.changeset/nervous-ties-try.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Features to Relaunch browser in debug, test connection

.changeset/ninety-pots-rescue.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Added option to auto-approve all commands

.changeset/polite-flowers-lick.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Remote browser control via devtools protocol

.changeset/polite-lions-deliver.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-factor out servers list

.changeset/red-worms-turn.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Fix tool toggle auto approve

.changeset/selfish-baboons-tickle.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-allow enabling prompt caching for LiteLLM + Claude

.changeset/serious-hornets-behave.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Added feature to detect installed versions of chromium and display them as a placeholder if not already explicitly configured by the user

.changeset/swift-frogs-learn.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Fix for headless browser mode

.changeset/thirty-news-speak.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-Drag and drop of file/folders into cline chat

.vscode/launch.json

@@ -16,6 +16,19 @@
 				"IS_DEV": "true",
 				"DEV_WORKSPACE_FOLDER": "${workspaceFolder}"
 			}
+		},
+		{
+			"name": "Run Extension (Test Mode)",
+			"type": "extensionHost",
+			"request": "launch",
+			"args": ["--extensionDevelopmentPath=${workspaceFolder}"],
+			"outFiles": ["${workspaceFolder}/dist/**/*.js"],
+			"preLaunchTask": "${defaultBuildTask}",
+			"env": {
+				"IS_DEV": "true",
+				"IS_TEST": "true",
+				"DEV_WORKSPACE_FOLDER": "${workspaceFolder}"
+			}
 		}
 	]
 }

.vscode/tasks.json

@@ -14,6 +14,14 @@
 				"isDefault": true
 			}
 		},
+		{
+			"label": "watch:test",
+			"dependsOn": ["npm: build:webview:test", "npm: dev:webview", "npm: watch:tsc", "npm: watch:esbuild:test"],
+			"presentation": {
+				"reveal": "never"
+			},
+			"group": "build"
+		},
 		{
 			"type": "npm",
 			"script": "build:webview",
@@ -32,6 +40,25 @@
 				}
 			}
 		},
+		{
+			"type": "npm",
+			"script": "build:webview:test",
+			"group": "build",
+			"problemMatcher": [],
+			"isBackground": true,
+			"label": "npm: build:webview:test",
+			"presentation": {
+				"group": "watch",
+				"reveal": "never",
+				"close": true
+			},
+			"options": {
+				"env": {
+					"IS_DEV": "true",
+					"IS_TEST": "true"
+				}
+			}
+		},
 		{
 			"type": "npm",
 			"script": "dev:webview",
@@ -77,6 +104,30 @@
 				"group": "watch",
 				"reveal": "never",
 				"close": true
+			},
+			"options": {
+				"env": {
+					"IS_DEV": "true"
+				}
+			}
+		},
+		{
+			"type": "npm",
+			"script": "watch:esbuild:test",
+			"group": "build",
+			"problemMatcher": "$esbuild-watch",
+			"isBackground": true,
+			"label": "npm: watch:esbuild:test",
+			"presentation": {
+				"group": "watch",
+				"reveal": "never",
+				"close": true
+			},
+			"options": {
+				"env": {
+					"IS_DEV": "true",
+					"IS_TEST": "true"
+				}
 			}
 		},
 		{

CHANGELOG.md

@@ -1,5 +1,58 @@
 # Changelog
 
+## [3.12.3]
+
+-   Add copy button to MermaidBlock component (Thanks @cacosub7!)
+-   Add the ability to fetch from global cline rules files
+-   Add icon to indicate when a file outside of the users workspace is edited
+
+## [3.12.2]
+
+-   Add gpt-4.1
+
+## [3.12.1]
+
+-   Use visual checkpoint indicator to make it clear when checkpoints are created
+-   Big shoutout to @samuel871211 for numerous code quality improvements, refactoring contributions, and webview performance improvements!
+-   Use improved context manager
+
+## [3.12.0]
+
+-   Add favorite toggles for models when using the Cline & OpenRouter providers
+-   Add auto-approve options for edits/reads outside of the workspace
+-   Improve diff editing animation for large files
+-   Add indicator showing number of diff edits when Cline edits a file
+-   Add streaming support and reasoning effort option to xAI's Grok 3 Mini
+-   Add settings button to MCP popover to easily modify installed servers
+-   Fix bug where browser tool actions would show unparsed results in the chat view
+-   Fix issue with new checkpoints popover hiding too quickly
+-   Fix duplicate checkpoints bug
+-   Improve Ollama provider with retry mechanism, timeout handling, and improved error handling (thanks suvarchal!)
+
+## [3.11.0]
+
+-   Redesign checkpoint UI to declutter chat view by using a subtle indicator line that expands to a popover on hover, with a new date indicator for when it was created
+-   Add support for xAI's provider's Grok 3 models
+-   Add more robust error tracking for users opted in to telemetry (thank you for helping us make Cline better!)
+
+## [3.10.1]
+
+-   Add CMD+' keyboard shortcut to add selected text to Cline
+-   Cline now auto focuses the text field when using 'Add to Cline' shortcut
+-   Add new 'Create New Task' tool to let Cline start a new task autonomously!
+-   Fix Mermaid diagram issues
+-   Fix Gemini provider cost calculation to take new tiered pricing structure into account
+
+## [3.10.0]
+
+-   Add setting to let browser tool use local Chrome via remote debugging, enabling session-based browsing. Replaces sessionless Chromium, unlocking debugging and productivity workflows tied to your real browser state.
+-   Add new auto-approve option to approve _ALL_ commands (use at your own risk!)
+-   Add modal in the chat area to more easily enable or disable MCP servers
+-   Add drag and drop of file/folders into cline chat (Thanks eljapi!)
+-   Add prompt caching for LiteLLM + Claude (Thanks sammcj!)
+-   Add Improved context management
+-   Fix MCP auto approve toggle issues being out of sync with settings
+
 ## [3.9.2]
 
 -   Add recommended models for Cline provider

README.md

@@ -2,7 +2,7 @@
 English | <a href="https://github.com/cline/cline/blob/main/locales/es/README.md" target="_blank">Español</a> | <a href="https://github.com/cline/cline/blob/main/locales/de/README.md" target="_blank">Deutsch</a> | <a href="https://github.com/cline/cline/blob/main/locales/ja/README.md" target="_blank">日本語</a> | <a href="https://github.com/cline/cline/blob/main/locales/zh-cn/README.md" target="_blank">简体中文</a> | <a href="https://github.com/cline/cline/blob/main/locales/zh-tw/README.md" target="_blank">繁體中文</a> | <a href="https://github.com/cline/cline/blob/main/locales/ko/README.md" target="_blank">한국어</a>
 </sub></div>
 
-# Cline – \#1 on OpenRouter
+# Cline
 
 <p align="center">
   <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

docs/tools/cline-tools-guide.md

@@ -82,6 +82,7 @@ Cline has access to the following tools for various tasks:
 4. **Interaction Tools**
     - `ask_followup_question`: Ask user for clarification
     - `attempt_completion`: Present final results
+    - `new_task`: Start a new task with preloaded context
 
 Each tool has specific parameters and usage patterns. Here are some examples:
 
@@ -114,6 +115,21 @@ Each tool has specific parameters and usage patterns. Here are some examples:
     </execute_command>
     ```
 
+-   Start a new task with context (new_task):
+    ```xml
+    <new_task>
+    <context>
+    We've completed the backend API with these endpoints:
+    - GET /api/tasks
+    - POST /api/tasks
+    - PUT /api/tasks/:id
+    - DELETE /api/tasks/:id
+
+    Now we need to implement the React frontend.
+    </context>
+    </new_task>
+    ```
+
 ## Common Tasks
 
 1. **Create a New Component**

esbuild.js

@@ -4,6 +4,7 @@ const path = require("path")
 
 const production = process.argv.includes("--production")
 const watch = process.argv.includes("--watch")
+const test = process.env.IS_TEST === "true"
 
 /**
  * @type {import('esbuild').Plugin}
@@ -68,6 +69,10 @@ const extensionConfig = {
 	minify: production,
 	sourcemap: !production,
 	logLevel: "silent",
+	define: {
+		"process.env.IS_DEV": JSON.stringify(!production),
+		"process.env.IS_TEST": JSON.stringify(test),
+	},
 	plugins: [
 		copyWasmFiles,
 		/* add to the end of plugins array */

evals/.gitignore

@@ -0,0 +1,3 @@
+repositories
+
+results/evals.db

evals/README.md

@@ -0,0 +1,162 @@
+# Cline Evaluation System
+
+This directory contains the evaluation system for benchmarking Cline against various coding evaluation frameworks.
+
+## Overview
+
+The Cline Evaluation System allows you to:
+
+1. Run Cline against standardized coding benchmarks
+2. Collect comprehensive metrics on performance
+3. Generate detailed reports on evaluation results
+4. Compare performance across different models and benchmarks
+
+## Architecture
+
+The evaluation system consists of two main components:
+
+1. **Test Server**: Enhanced HTTP server in `src/services/test/TestServer.ts` that provides detailed task results
+2. **CLI Tool**: Command-line interface in `evals/cli/` for orchestrating evaluations
+
+## Directory Structure
+
+```
+cline-repo/
+├── src/
+│   ├── services/
+│   │   ├── test/
+│   │   │   ├── TestServer.ts         # Enhanced HTTP server for task execution
+│   │   │   ├── GitHelper.ts          # Git utilities for file tracking
+│   │   │   └── ...
+│   │   └── ...
+│   └── ...
+├── evals/                            # Main directory for evaluation system
+│   ├── cli/                          # CLI tool for orchestrating evaluations
+│   │   ├── src/
+│   │   │   ├── index.ts              # CLI entry point
+│   │   │   ├── commands/             # CLI commands (setup, run, report)
+│   │   │   ├── adapters/             # Benchmark adapters
+│   │   │   ├── db/                   # Database management
+│   │   │   └── utils/                # Utility functions
+│   │   ├── package.json
+│   │   └── tsconfig.json
+│   ├── repositories/                 # Cloned benchmark repositories
+│   │   ├── exercism/                 # Modified Exercism (from pashpashpash/evals)
+│   │   ├── swe-bench/                # SWE-Bench repository
+│   │   ├── swelancer/                # SWELancer repository
+│   │   └── multi-swe/                # Multi-SWE-Bench repository
+│   ├── results/                      # Evaluation results storage
+│   │   ├── runs/                     # Individual run results
+│   │   └── reports/                  # Generated reports
+│   └── README.md                     # This file
+└── ...
+```
+
+## Getting Started
+
+### Prerequisites
+
+- Node.js 16+
+- VSCode with Cline extension installed
+- Git
+
+### Installation
+
+1. Build the CLI tool:
+
+```bash
+cd evals/cli
+npm install
+npm run build
+```
+
+### Usage
+
+#### Setting Up Benchmarks
+
+```bash
+cd evals/cli
+node dist/index.js setup
+```
+
+This will clone and set up all benchmark repositories. You can specify specific benchmarks:
+
+```bash
+node dist/index.js setup --benchmarks exercism
+```
+
+#### Running Evaluations
+
+```bash
+node dist/index.js run --model claude-3-opus-20240229 --benchmark exercism
+```
+
+Options:
+- `--model`: The model to evaluate (default: claude-3-opus-20240229)
+- `--benchmark`: Specific benchmark to run (default: all)
+- `--count`: Number of tasks to run (default: all)
+
+#### Generating Reports
+
+```bash
+node dist/index.js report
+```
+
+Options:
+- `--format`: Report format (json, markdown) (default: markdown)
+- `--output`: Output path for the report
+
+## Benchmarks
+
+### Exercism
+
+Modified Exercism exercises from the [pashpashpash/evals](https://github.com/pashpashpash/evals) repository. These are small, focused programming exercises in various languages.
+
+### SWE-Bench (Coming Soon)
+
+Real-world software engineering tasks from the [SWE-bench](https://github.com/SWE-bench/SWE-bench) repository.
+
+### SWELancer (Coming Soon)
+
+Freelance-style programming tasks from the SWELancer benchmark.
+
+### Multi-SWE-Bench (Coming Soon)
+
+Multi-file software engineering tasks from the Multi-SWE-Bench repository.
+
+## Metrics
+
+The evaluation system collects the following metrics:
+
+- **Token Usage**: Input and output tokens
+- **Cost**: Estimated cost of API calls
+- **Duration**: Time taken to complete tasks
+- **Tool Usage**: Number of tool calls and failures
+- **Success Rate**: Percentage of tasks completed successfully
+- **Functional Correctness**: Percentage of tests passed
+
+## Reports
+
+Reports are generated in Markdown or JSON format and include:
+
+- Overall summary
+- Benchmark-specific results
+- Model-specific results
+- Tool usage statistics
+- Charts and visualizations
+
+## Development
+
+### Adding a New Benchmark
+
+1. Create a new adapter in `evals/cli/src/adapters/`
+2. Implement the `BenchmarkAdapter` interface
+3. Register the adapter in `evals/cli/src/adapters/index.ts`
+
+### Extending Metrics
+
+To add new metrics:
+
+1. Update the database schema in `evals/cli/src/db/schema.ts`
+2. Add collection logic in `evals/cli/src/utils/results.ts`
+3. Update report generation in `evals/cli/src/commands/report.ts`

evals/cli/package.json

@@ -0,0 +1,39 @@
+{
+	"name": "cline-evaluation-cli",
+	"version": "0.1.0",
+	"description": "CLI tool for orchestrating Cline evaluations across multiple benchmarks",
+	"main": "dist/index.js",
+	"scripts": {
+		"build": "tsc",
+		"start": "node dist/index.js",
+		"dev": "ts-node src/index.ts",
+		"test": "echo \"Error: no test specified\" && exit 1"
+	},
+	"keywords": [
+		"cline",
+		"evaluation",
+		"benchmark"
+	],
+	"author": "",
+	"license": "MIT",
+	"dependencies": {
+		"better-sqlite3": "^8.0.0",
+		"chalk": "^4.1.2",
+		"commander": "^9.4.1",
+		"execa": "^5.1.1",
+		"node-fetch": "^2.7.0",
+		"ora": "^5.4.1",
+		"sqlite": "^4.1.2",
+		"uuid": "^9.0.0",
+		"yargs": "^17.6.2"
+	},
+	"devDependencies": {
+		"@types/better-sqlite3": "^7.6.3",
+		"@types/node": "^18.11.18",
+		"@types/node-fetch": "^2.6.12",
+		"@types/uuid": "^9.0.0",
+		"@types/yargs": "^17.0.19",
+		"ts-node": "^10.9.1",
+		"typescript": "^4.9.4"
+	}
+}

evals/cli/src/adapters/exercism.ts

@@ -0,0 +1,190 @@
+import * as path from "path"
+import * as fs from "fs"
+import execa from "execa"
+import { BenchmarkAdapter, Task, VerificationResult } from "./types"
+
+const EVALS_DIR = path.resolve(__dirname, "../../../")
+
+/**
+ * Adapter for the modified Exercism benchmark
+ */
+export class ExercismAdapter implements BenchmarkAdapter {
+	name = "exercism"
+
+	/**
+	 * Set up the Exercism benchmark repository
+	 */
+	async setup(): Promise<void> {
+		// Clone repository if needed
+		const exercismDir = path.join(EVALS_DIR, "repositories", "exercism")
+
+		if (!fs.existsSync(exercismDir)) {
+			console.log(`Cloning Exercism repository to ${exercismDir}...`)
+			await execa("git", ["clone", "https://github.com/pashpashpash/evals.git", exercismDir])
+			console.log("Exercism repository cloned successfully")
+		} else {
+			console.log(`Exercism repository already exists at ${exercismDir}`)
+
+			// Pull latest changes
+			console.log("Pulling latest changes...")
+			await execa("git", ["pull"], { cwd: exercismDir })
+			console.log("Repository updated successfully")
+		}
+	}
+
+	/**
+	 * List all available tasks in the Exercism benchmark
+	 */
+	async listTasks(): Promise<Task[]> {
+		const tasks: Task[] = []
+		const exercisesDir = path.join(EVALS_DIR, "repositories", "exercism")
+
+		// Ensure the repository exists
+		if (!fs.existsSync(exercisesDir)) {
+			throw new Error(`Exercism repository not found at ${exercisesDir}. Run setup first.`)
+		}
+
+		// Read language directories
+		const languages = fs
+			.readdirSync(exercisesDir)
+			.filter((dir) => fs.statSync(path.join(exercisesDir, dir)).isDirectory())
+			.filter((dir) => !dir.startsWith(".") && !["node_modules", ".git"].includes(dir))
+
+		for (const language of languages) {
+			const languageDir = path.join(exercisesDir, language)
+
+			// Read exercise directories
+			const exercises = fs.readdirSync(languageDir).filter((dir) => fs.statSync(path.join(languageDir, dir)).isDirectory())
+
+			for (const exercise of exercises) {
+				const exerciseDir = path.join(languageDir, exercise)
+
+				// Read instructions
+				let description = ""
+				const instructionsPath = path.join(exerciseDir, "docs", "instructions.md")
+				if (fs.existsSync(instructionsPath)) {
+					description = fs.readFileSync(instructionsPath, "utf-8")
+				}
+
+				// Determine test commands based on language
+				let testCommands: string[] = []
+				switch (language) {
+					case "javascript":
+						testCommands = ["npm install", "npm test"]
+						break
+					case "python":
+						testCommands = ["python -m pytest -o markers=task *_test.py"]
+						break
+					case "go":
+						testCommands = ["go test"]
+						break
+					case "java":
+						testCommands = ["./gradlew test"]
+						break
+					case "rust":
+						testCommands = ["cargo test"]
+						break
+					default:
+						testCommands = []
+				}
+
+				tasks.push({
+					id: `exercism-${language}-${exercise}`,
+					name: exercise,
+					description,
+					workspacePath: exerciseDir,
+					setupCommands: [],
+					verificationCommands: testCommands,
+					metadata: {
+						language,
+						type: "exercism",
+					},
+				})
+			}
+		}
+
+		return tasks
+	}
+
+	/**
+	 * Prepare a specific task for execution
+	 * @param taskId The ID of the task to prepare
+	 */
+	async prepareTask(taskId: string): Promise<Task> {
+		const tasks = await this.listTasks()
+		const task = tasks.find((t) => t.id === taskId)
+
+		if (!task) {
+			throw new Error(`Task ${taskId} not found`)
+		}
+
+		// Check if Git repository is already initialized
+		const gitDirExists = fs.existsSync(path.join(task.workspacePath, ".git"))
+
+		try {
+			// Initialize Git repository if needed
+			if (!gitDirExists) {
+				await execa("git", ["init"], { cwd: task.workspacePath })
+			}
+
+			// Create a dummy file to ensure there's something to commit
+			const dummyFilePath = path.join(task.workspacePath, ".eval-timestamp")
+			fs.writeFileSync(dummyFilePath, new Date().toISOString())
+
+			// Add all files and commit
+			await execa("git", ["add", "."], { cwd: task.workspacePath })
+
+			try {
+				await execa("git", ["commit", "-m", "Initial commit"], { cwd: task.workspacePath })
+			} catch (error: any) {
+				// If commit fails because there are no changes, that's okay
+				if (!error.stderr?.includes("nothing to commit")) {
+					throw error
+				}
+			}
+		} catch (error: any) {
+			console.warn(`Warning: Git operations failed: ${error.message}`)
+			console.warn("Continuing without Git initialization")
+		}
+
+		return task
+	}
+
+	/**
+	 * Verify the result of a task execution
+	 * @param task The task that was executed
+	 * @param result The result of the task execution
+	 */
+	async verifyResult(task: Task, result: any): Promise<VerificationResult> {
+		// Run verification commands
+		let success = true
+		let output = ""
+
+		for (const command of task.verificationCommands) {
+			try {
+				const [cmd, ...args] = command.split(" ")
+				const { stdout } = await execa(cmd, args, { cwd: task.workspacePath })
+				output += stdout + "\n"
+			} catch (error: any) {
+				success = false
+				if (error.stdout) output += error.stdout + "\n"
+				if (error.stderr) output += error.stderr + "\n"
+			}
+		}
+
+		// Parse test results
+		const testsPassed = (output.match(/PASS/g) || []).length
+		const testsFailed = (output.match(/FAIL/g) || []).length
+		const testsTotal = testsPassed + testsFailed
+
+		return {
+			success,
+			metrics: {
+				testsPassed,
+				testsFailed,
+				testsTotal,
+				functionalCorrectness: testsTotal > 0 ? testsPassed / testsTotal : 0,
+			},
+		}
+	}
+}

evals/cli/src/adapters/index.ts

@@ -0,0 +1,47 @@
+import { BenchmarkAdapter } from "./types"
+import { ExercismAdapter } from "./exercism"
+import { SWEBenchAdapter } from "./swe-bench"
+import { SWELancerAdapter } from "./swelancer"
+import { MultiSWEAdapter } from "./multi-swe"
+
+// Registry of all available adapters
+const adapters: Record<string, BenchmarkAdapter> = {
+	// Exercism is the primary adapter with real implementation
+	exercism: new ExercismAdapter(),
+
+	// Dummy adapters for testing
+	"swe-bench": new SWEBenchAdapter(),
+	swelancer: new SWELancerAdapter(),
+	"multi-swe": new MultiSWEAdapter(),
+}
+
+/**
+ * Get a specific adapter by name
+ * @param name The name of the adapter to get
+ * @returns The requested adapter
+ * @throws Error if the adapter is not found
+ */
+export function getAdapter(name: string): BenchmarkAdapter {
+	const adapter = adapters[name]
+	if (!adapter) {
+		throw new Error(`Adapter for benchmark '${name}' not found`)
+	}
+	return adapter
+}
+
+/**
+ * Get all available adapters
+ * @returns Array of all registered adapters
+ */
+export function getAllAdapters(): BenchmarkAdapter[] {
+	return Object.values(adapters)
+}
+
+/**
+ * Register a new adapter
+ * @param name The name to register the adapter under
+ * @param adapter The adapter to register
+ */
+export function registerAdapter(name: string, adapter: BenchmarkAdapter): void {
+	adapters[name] = adapter
+}

evals/cli/src/adapters/multi-swe.ts

@@ -0,0 +1,192 @@
+import * as path from "path"
+import * as fs from "fs"
+import execa from "execa"
+import { BenchmarkAdapter, Task, VerificationResult } from "./types"
+
+const EVALS_DIR = path.resolve(__dirname, "../../../")
+
+/**
+ * Dummy adapter for the Multi-SWE-Bench benchmark
+ */
+export class MultiSWEAdapter implements BenchmarkAdapter {
+	name = "multi-swe"
+
+	/**
+	 * Set up the Multi-SWE-Bench benchmark repository (dummy implementation)
+	 */
+	async setup(): Promise<void> {
+		console.log("Multi-SWE-Bench dummy setup completed")
+
+		// Create repositories directory if it doesn't exist
+		const repoDir = path.join(EVALS_DIR, "repositories", "multi-swe")
+		if (!fs.existsSync(repoDir)) {
+			fs.mkdirSync(repoDir, { recursive: true })
+			console.log(`Created dummy Multi-SWE-Bench directory at ${repoDir}`)
+		}
+	}
+
+	/**
+	 * List all available tasks in the Multi-SWE-Bench benchmark (dummy implementation)
+	 */
+	async listTasks(): Promise<Task[]> {
+		return [
+			{
+				id: "multi-swe-task-1",
+				name: "Multi-Language API Integration",
+				description:
+					"Implement a system that integrates a Python backend with a TypeScript frontend and a Rust processing service.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "multi-swe"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					languages: ["python", "typescript", "rust"],
+					complexity: "high",
+					type: "multi-swe",
+				},
+			},
+			{
+				id: "multi-swe-task-2",
+				name: "Cross-Platform Mobile App",
+				description: "Create a cross-platform mobile app using React Native with native modules in Swift and Kotlin.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "multi-swe"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					languages: ["javascript", "swift", "kotlin"],
+					complexity: "medium",
+					type: "multi-swe",
+				},
+			},
+			{
+				id: "multi-swe-task-3",
+				name: "Microservice Architecture",
+				description: "Design and implement a microservice architecture with services written in Go, Node.js, and Java.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "multi-swe"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					languages: ["go", "javascript", "java"],
+					complexity: "high",
+					type: "multi-swe",
+				},
+			},
+		]
+	}
+
+	/**
+	 * Prepare a specific task for execution (dummy implementation)
+	 * @param taskId The ID of the task to prepare
+	 */
+	async prepareTask(taskId: string): Promise<Task> {
+		const tasks = await this.listTasks()
+		const task = tasks.find((t) => t.id === taskId)
+
+		if (!task) {
+			throw new Error(`Task ${taskId} not found`)
+		}
+
+		// Create a dummy workspace for the task
+		const taskDir = path.join(task.workspacePath, taskId)
+		if (!fs.existsSync(taskDir)) {
+			fs.mkdirSync(taskDir, { recursive: true })
+
+			// Create a dummy file for the task
+			fs.writeFileSync(
+				path.join(taskDir, "README.md"),
+				`# ${task.name}\n\n${task.description}\n\nThis is a dummy task for testing purposes.`,
+			)
+
+			// Create additional dummy files based on task type
+			if (task.id === "multi-swe-task-1") {
+				// Python backend
+				fs.mkdirSync(path.join(taskDir, "backend"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "backend", "app.py"),
+					`# TODO: Implement Python backend\nfrom flask import Flask\n\napp = Flask(__name__)\n\n@app.route('/')\ndef hello():\n    return "Hello, World!"\n`,
+				)
+
+				// TypeScript frontend
+				fs.mkdirSync(path.join(taskDir, "frontend"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "frontend", "app.ts"),
+					`// TODO: Implement TypeScript frontend\nconsole.log('Frontend starting...');\n`,
+				)
+
+				// Rust processing service
+				fs.mkdirSync(path.join(taskDir, "processor"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "processor", "main.rs"),
+					`// TODO: Implement Rust processing service\nfn main() {\n    println!("Processor starting...");\n}\n`,
+				)
+			} else if (task.id === "multi-swe-task-2") {
+				// React Native app
+				fs.mkdirSync(path.join(taskDir, "app"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "app", "App.js"),
+					`// TODO: Implement React Native app\nimport React from 'react';\nimport { View, Text } from 'react-native';\n\nexport default function App() {\n  return (\n    <View>\n      <Text>Hello, World!</Text>\n    </View>\n  );\n}\n`,
+				)
+
+				// Swift native module
+				fs.mkdirSync(path.join(taskDir, "ios"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "ios", "NativeModule.swift"),
+					`// TODO: Implement Swift native module\nimport Foundation\n\n@objc(NativeModule)\nclass NativeModule: NSObject {\n  @objc\n  func hello() -> String {\n    return "Hello from Swift"\n  }\n}\n`,
+				)
+
+				// Kotlin native module
+				fs.mkdirSync(path.join(taskDir, "android"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "android", "NativeModule.kt"),
+					`// TODO: Implement Kotlin native module\npackage com.example.app\n\nclass NativeModule {\n  fun hello(): String {\n    return "Hello from Kotlin"\n  }\n}\n`,
+				)
+			} else if (task.id === "multi-swe-task-3") {
+				// Go service
+				fs.mkdirSync(path.join(taskDir, "service-go"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "service-go", "main.go"),
+					`// TODO: Implement Go service\npackage main\n\nimport "fmt"\n\nfunc main() {\n\tfmt.Println("Go service starting...")\n}\n`,
+				)
+
+				// Node.js service
+				fs.mkdirSync(path.join(taskDir, "service-node"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "service-node", "server.js"),
+					`// TODO: Implement Node.js service\nconsole.log('Node.js service starting...');\n`,
+				)
+
+				// Java service
+				fs.mkdirSync(path.join(taskDir, "service-java"), { recursive: true })
+				fs.writeFileSync(
+					path.join(taskDir, "service-java", "Main.java"),
+					`// TODO: Implement Java service\npublic class Main {\n    public static void main(String[] args) {\n        System.out.println("Java service starting...");\n    }\n}\n`,
+				)
+			}
+		}
+
+		// Update the task's workspace path to the task-specific directory
+		return {
+			...task,
+			workspacePath: taskDir,
+		}
+	}
+
+	/**
+	 * Verify the result of a task execution (dummy implementation)
+	 * @param task The task that was executed
+	 * @param result The result of the task execution
+	 */
+	async verifyResult(task: Task, result: any): Promise<VerificationResult> {
+		// Always return success for dummy implementation
+		return {
+			success: true,
+			metrics: {
+				testsPassed: 1,
+				testsFailed: 0,
+				testsTotal: 1,
+				functionalCorrectness: 1.0,
+				crossLanguageIntegration: 0.9, // Dummy metric specific to Multi-SWE
+				architectureQuality: 0.85, // Dummy metric specific to Multi-SWE
+			},
+		}
+	}
+}

evals/cli/src/adapters/swe-bench.ts

@@ -0,0 +1,125 @@
+import * as path from "path"
+import * as fs from "fs"
+import execa from "execa"
+import { BenchmarkAdapter, Task, VerificationResult } from "./types"
+
+const EVALS_DIR = path.resolve(__dirname, "../../../")
+
+/**
+ * Dummy adapter for the SWE-Bench benchmark
+ */
+export class SWEBenchAdapter implements BenchmarkAdapter {
+	name = "swe-bench"
+
+	/**
+	 * Set up the SWE-Bench benchmark repository (dummy implementation)
+	 */
+	async setup(): Promise<void> {
+		console.log("SWE-Bench dummy setup completed")
+
+		// Create repositories directory if it doesn't exist
+		const repoDir = path.join(EVALS_DIR, "repositories", "swe-bench")
+		if (!fs.existsSync(repoDir)) {
+			fs.mkdirSync(repoDir, { recursive: true })
+			console.log(`Created dummy SWE-Bench directory at ${repoDir}`)
+		}
+	}
+
+	/**
+	 * List all available tasks in the SWE-Bench benchmark (dummy implementation)
+	 */
+	async listTasks(): Promise<Task[]> {
+		return [
+			{
+				id: "swe-bench-task-1",
+				name: "Fix React Component Bug",
+				description: "Fix a bug in a React component where the state is not properly updated.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "swe-bench"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					repository: "facebook/react",
+					issue: "#12345",
+					type: "swe-bench",
+				},
+			},
+			{
+				id: "swe-bench-task-2",
+				name: "Optimize Database Query",
+				description: "Optimize a slow database query in a Django application.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "swe-bench"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					repository: "django/django",
+					issue: "#6789",
+					type: "swe-bench",
+				},
+			},
+			{
+				id: "swe-bench-task-3",
+				name: "Fix Memory Leak",
+				description: "Fix a memory leak in a Node.js application.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "swe-bench"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					repository: "nodejs/node",
+					issue: "#9876",
+					type: "swe-bench",
+				},
+			},
+		]
+	}
+
+	/**
+	 * Prepare a specific task for execution (dummy implementation)
+	 * @param taskId The ID of the task to prepare
+	 */
+	async prepareTask(taskId: string): Promise<Task> {
+		const tasks = await this.listTasks()
+		const task = tasks.find((t) => t.id === taskId)
+
+		if (!task) {
+			throw new Error(`Task ${taskId} not found`)
+		}
+
+		// Create a dummy workspace for the task
+		const taskDir = path.join(task.workspacePath, taskId)
+		if (!fs.existsSync(taskDir)) {
+			fs.mkdirSync(taskDir, { recursive: true })
+
+			// Create a dummy file for the task
+			fs.writeFileSync(
+				path.join(taskDir, "README.md"),
+				`# ${task.name}\n\n${task.description}\n\nThis is a dummy task for testing purposes.`,
+			)
+		}
+
+		// Update the task's workspace path to the task-specific directory
+		return {
+			...task,
+			workspacePath: taskDir,
+		}
+	}
+
+	/**
+	 * Verify the result of a task execution (dummy implementation)
+	 * @param task The task that was executed
+	 * @param result The result of the task execution
+	 */
+	async verifyResult(task: Task, result: any): Promise<VerificationResult> {
+		// Always return success for dummy implementation
+		return {
+			success: true,
+			metrics: {
+				testsPassed: 1,
+				testsFailed: 0,
+				testsTotal: 1,
+				functionalCorrectness: 1.0,
+				performanceImprovement: 0.25, // Dummy metric specific to SWE-Bench
+				codeQuality: 0.9, // Dummy metric specific to SWE-Bench
+			},
+		}
+	}
+}

evals/cli/src/adapters/swelancer.ts

@@ -0,0 +1,143 @@
+import * as path from "path"
+import * as fs from "fs"
+import execa from "execa"
+import { BenchmarkAdapter, Task, VerificationResult } from "./types"
+
+const EVALS_DIR = path.resolve(__dirname, "../../../")
+
+/**
+ * Dummy adapter for the SWELancer benchmark
+ */
+export class SWELancerAdapter implements BenchmarkAdapter {
+	name = "swelancer"
+
+	/**
+	 * Set up the SWELancer benchmark repository (dummy implementation)
+	 */
+	async setup(): Promise<void> {
+		console.log("SWELancer dummy setup completed")
+
+		// Create repositories directory if it doesn't exist
+		const repoDir = path.join(EVALS_DIR, "repositories", "swelancer")
+		if (!fs.existsSync(repoDir)) {
+			fs.mkdirSync(repoDir, { recursive: true })
+			console.log(`Created dummy SWELancer directory at ${repoDir}`)
+		}
+	}
+
+	/**
+	 * List all available tasks in the SWELancer benchmark (dummy implementation)
+	 */
+	async listTasks(): Promise<Task[]> {
+		return [
+			{
+				id: "swelancer-task-1",
+				name: "Create Landing Page",
+				description: "Create a responsive landing page for a new product using HTML, CSS, and JavaScript.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "swelancer"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					client: "TechStartup Inc.",
+					difficulty: "medium",
+					type: "swelancer",
+				},
+			},
+			{
+				id: "swelancer-task-2",
+				name: "Build REST API",
+				description: "Create a RESTful API for a blog application using Node.js and Express.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "swelancer"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					client: "BlogCo",
+					difficulty: "hard",
+					type: "swelancer",
+				},
+			},
+			{
+				id: "swelancer-task-3",
+				name: "Fix CSS Layout Issues",
+				description: "Fix layout issues in a responsive website across different screen sizes.",
+				workspacePath: path.join(EVALS_DIR, "repositories", "swelancer"),
+				setupCommands: [],
+				verificationCommands: [],
+				metadata: {
+					client: "DesignAgency",
+					difficulty: "easy",
+					type: "swelancer",
+				},
+			},
+		]
+	}
+
+	/**
+	 * Prepare a specific task for execution (dummy implementation)
+	 * @param taskId The ID of the task to prepare
+	 */
+	async prepareTask(taskId: string): Promise<Task> {
+		const tasks = await this.listTasks()
+		const task = tasks.find((t) => t.id === taskId)
+
+		if (!task) {
+			throw new Error(`Task ${taskId} not found`)
+		}
+
+		// Create a dummy workspace for the task
+		const taskDir = path.join(task.workspacePath, taskId)
+		if (!fs.existsSync(taskDir)) {
+			fs.mkdirSync(taskDir, { recursive: true })
+
+			// Create a dummy file for the task
+			fs.writeFileSync(
+				path.join(taskDir, "README.md"),
+				`# ${task.name}\n\n${task.description}\n\nThis is a dummy task for testing purposes.`,
+			)
+
+			// Create additional dummy files based on task type
+			if (task.id === "swelancer-task-1") {
+				fs.writeFileSync(
+					path.join(taskDir, "index.html"),
+					`<!DOCTYPE html>\n<html>\n<head>\n  <title>Landing Page</title>\n</head>\n<body>\n  <!-- TODO: Implement landing page -->\n</body>\n</html>`,
+				)
+			} else if (task.id === "swelancer-task-2") {
+				fs.writeFileSync(
+					path.join(taskDir, "server.js"),
+					`// TODO: Implement REST API\nconsole.log('Server starting...');`,
+				)
+			} else if (task.id === "swelancer-task-3") {
+				fs.writeFileSync(
+					path.join(taskDir, "styles.css"),
+					`/* TODO: Fix layout issues */\nbody {\n  margin: 0;\n  padding: 0;\n}`,
+				)
+			}
+		}
+
+		// Update the task's workspace path to the task-specific directory
+		return {
+			...task,
+			workspacePath: taskDir,
+		}
+	}
+
+	/**
+	 * Verify the result of a task execution (dummy implementation)
+	 * @param task The task that was executed
+	 * @param result The result of the task execution
+	 */
+	async verifyResult(task: Task, result: any): Promise<VerificationResult> {
+		// Always return success for dummy implementation
+		return {
+			success: true,
+			metrics: {
+				testsPassed: 1,
+				testsFailed: 0,
+				testsTotal: 1,
+				functionalCorrectness: 1.0,
+				clientSatisfaction: 0.95, // Dummy metric specific to SWELancer
+				timeEfficiency: 0.85, // Dummy metric specific to SWELancer
+			},
+		}
+	}
+}

evals/cli/src/adapters/types.ts

@@ -0,0 +1,31 @@
+/**
+ * Represents a task to be executed
+ */
+export interface Task {
+	id: string
+	name: string
+	description: string
+	workspacePath: string
+	setupCommands: string[]
+	verificationCommands: string[]
+	metadata: Record<string, any>
+}
+
+/**
+ * Result of verifying a task execution
+ */
+export interface VerificationResult {
+	success: boolean
+	metrics: Record<string, any>
+}
+
+/**
+ * Interface for benchmark adapters
+ */
+export interface BenchmarkAdapter {
+	name: string
+	setup(): Promise<void>
+	listTasks(): Promise<Task[]>
+	prepareTask(taskId: string): Promise<Task>
+	verifyResult(task: Task, result: any): Promise<VerificationResult>
+}

evals/cli/src/commands/report.ts

@@ -0,0 +1,237 @@
+import * as fs from "fs"
+import * as path from "path"
+import chalk from "chalk"
+import ora from "ora"
+import { ResultsDatabase } from "../db"
+import { generateMarkdownReport } from "../utils/markdown"
+
+interface ReportOptions {
+	format?: "json" | "markdown"
+	output?: string
+}
+
+/**
+ * Handler for the report command
+ * @param options Command options
+ */
+export async function reportHandler(options: ReportOptions): Promise<void> {
+	const format = options.format || "markdown"
+	const db = new ResultsDatabase()
+
+	try {
+		const spinner = ora("Generating report...").start()
+
+		// Get all runs
+		const runs = db.getRuns()
+
+		console.log(chalk.blue(`Found ${runs.length} evaluation runs`))
+
+		if (runs.length === 0) {
+			spinner.fail("No evaluation runs found")
+			return
+		}
+
+		// Generate summary report
+		const summary = {
+			runs: runs.length,
+			models: [...new Set(runs.map((run) => run.model))],
+			benchmarks: [...new Set(runs.map((run) => run.benchmark))],
+			tasks: 0,
+			successRate: 0,
+			averageTokens: 0,
+			averageCost: 0,
+			averageDuration: 0,
+			totalToolCalls: 0,
+			totalToolFailures: 0,
+			toolSuccessRate: 0,
+			toolUsage: {} as Record<string, { calls: number; failures: number }>,
+		}
+
+		let totalTasks = 0
+		let successfulTasks = 0
+		let totalTokens = 0
+		let totalCost = 0
+		let totalDuration = 0
+		let totalToolCalls = 0
+		let totalToolFailures = 0
+
+		for (const run of runs) {
+			const tasks = db.getRunTasks(run.id)
+			totalTasks += tasks.length
+
+			for (const task of tasks) {
+				if (task.success) {
+					successfulTasks++
+				}
+
+				const metrics = db.getTaskMetrics(task.id)
+
+				const tokensIn = metrics.find((m) => m.name === "tokensIn")?.value || 0
+				const tokensOut = metrics.find((m) => m.name === "tokensOut")?.value || 0
+				totalTokens += tokensIn + tokensOut
+
+				totalCost += metrics.find((m) => m.name === "cost")?.value || 0
+				totalDuration += metrics.find((m) => m.name === "duration")?.value || 0
+
+				// Collect tool call metrics
+				totalToolCalls += task.total_tool_calls || 0
+				totalToolFailures += task.total_tool_failures || 0
+
+				// Get detailed tool usage
+				const toolCalls = db.getTaskToolCalls(task.id)
+
+				for (const toolCall of toolCalls) {
+					if (!summary.toolUsage[toolCall.tool_name]) {
+						summary.toolUsage[toolCall.tool_name] = {
+							calls: 0,
+							failures: 0,
+						}
+					}
+
+					summary.toolUsage[toolCall.tool_name].calls += toolCall.call_count
+					summary.toolUsage[toolCall.tool_name].failures += toolCall.failure_count
+				}
+			}
+		}
+
+		// Calculate tool success rate
+		summary.totalToolCalls = totalToolCalls
+		summary.totalToolFailures = totalToolFailures
+		summary.toolSuccessRate = totalToolCalls > 0 ? 1 - totalToolFailures / totalToolCalls : 1.0
+
+		summary.tasks = totalTasks
+		summary.successRate = totalTasks > 0 ? successfulTasks / totalTasks : 0
+		summary.averageTokens = totalTasks > 0 ? totalTokens / totalTasks : 0
+		summary.averageCost = totalTasks > 0 ? totalCost / totalTasks : 0
+		summary.averageDuration = totalTasks > 0 ? totalDuration / totalTasks : 0
+
+		// Generate benchmark-specific reports
+		const benchmarkReports: Record<string, any> = {}
+
+		for (const benchmark of summary.benchmarks) {
+			const benchmarkRuns = runs.filter((run) => run.benchmark === benchmark)
+			const benchmarkSummary = {
+				runs: benchmarkRuns.length,
+				models: [...new Set(benchmarkRuns.map((run) => run.model))],
+				tasks: 0,
+				successRate: 0,
+				averageTokens: 0,
+				averageCost: 0,
+				averageDuration: 0,
+			}
+
+			let benchmarkTasks = 0
+			let benchmarkSuccessfulTasks = 0
+			let benchmarkTotalTokens = 0
+			let benchmarkTotalCost = 0
+			let benchmarkTotalDuration = 0
+
+			for (const run of benchmarkRuns) {
+				const tasks = db.getRunTasks(run.id)
+				benchmarkTasks += tasks.length
+
+				for (const task of tasks) {
+					if (task.success) {
+						benchmarkSuccessfulTasks++
+					}
+
+					const metrics = db.getTaskMetrics(task.id)
+
+					const tokensIn = metrics.find((m) => m.name === "tokensIn")?.value || 0
+					const tokensOut = metrics.find((m) => m.name === "tokensOut")?.value || 0
+					benchmarkTotalTokens += tokensIn + tokensOut
+
+					benchmarkTotalCost += metrics.find((m) => m.name === "cost")?.value || 0
+					benchmarkTotalDuration += metrics.find((m) => m.name === "duration")?.value || 0
+				}
+			}
+
+			benchmarkSummary.tasks = benchmarkTasks
+			benchmarkSummary.successRate = benchmarkTasks > 0 ? benchmarkSuccessfulTasks / benchmarkTasks : 0
+			benchmarkSummary.averageTokens = benchmarkTasks > 0 ? benchmarkTotalTokens / benchmarkTasks : 0
+			benchmarkSummary.averageCost = benchmarkTasks > 0 ? benchmarkTotalCost / benchmarkTasks : 0
+			benchmarkSummary.averageDuration = benchmarkTasks > 0 ? benchmarkTotalDuration / benchmarkTasks : 0
+
+			benchmarkReports[benchmark] = benchmarkSummary
+		}
+
+		// Generate model-specific reports
+		const modelReports: Record<string, any> = {}
+
+		for (const model of summary.models) {
+			const modelRuns = runs.filter((run) => run.model === model)
+			const modelSummary = {
+				runs: modelRuns.length,
+				benchmarks: [...new Set(modelRuns.map((run) => run.benchmark))],
+				tasks: 0,
+				successRate: 0,
+				averageTokens: 0,
+				averageCost: 0,
+				averageDuration: 0,
+			}
+
+			let modelTasks = 0
+			let modelSuccessfulTasks = 0
+			let modelTotalTokens = 0
+			let modelTotalCost = 0
+			let modelTotalDuration = 0
+
+			for (const run of modelRuns) {
+				const tasks = db.getRunTasks(run.id)
+				modelTasks += tasks.length
+
+				for (const task of tasks) {
+					if (task.success) {
+						modelSuccessfulTasks++
+					}
+
+					const metrics = db.getTaskMetrics(task.id)
+
+					const tokensIn = metrics.find((m) => m.name === "tokensIn")?.value || 0
+					const tokensOut = metrics.find((m) => m.name === "tokensOut")?.value || 0
+					modelTotalTokens += tokensIn + tokensOut
+
+					modelTotalCost += metrics.find((m) => m.name === "cost")?.value || 0
+					modelTotalDuration += metrics.find((m) => m.name === "duration")?.value || 0
+				}
+			}
+
+			modelSummary.tasks = modelTasks
+			modelSummary.successRate = modelTasks > 0 ? modelSuccessfulTasks / modelTasks : 0
+			modelSummary.averageTokens = modelTasks > 0 ? modelTotalTokens / modelTasks : 0
+			modelSummary.averageCost = modelTasks > 0 ? modelTotalCost / modelTasks : 0
+			modelSummary.averageDuration = modelTasks > 0 ? modelTotalDuration / modelTasks : 0
+
+			modelReports[model] = modelSummary
+		}
+
+		// Save reports
+		const reportDir = path.join(path.resolve(__dirname, "../../../"), "results", "reports")
+		fs.mkdirSync(reportDir, { recursive: true })
+
+		const timestamp = new Date().toISOString().replace(/:/g, "-")
+
+		if (format === "json") {
+			// Save JSON reports
+			fs.writeFileSync(path.join(reportDir, `summary-${timestamp}.json`), JSON.stringify(summary, null, 2))
+
+			fs.writeFileSync(path.join(reportDir, `benchmarks-${timestamp}.json`), JSON.stringify(benchmarkReports, null, 2))
+
+			fs.writeFileSync(path.join(reportDir, `models-${timestamp}.json`), JSON.stringify(modelReports, null, 2))
+
+			spinner.succeed(`JSON reports generated in ${reportDir}`)
+		} else {
+			// Generate markdown report
+			const outputPath = options.output || path.join(reportDir, `report-${timestamp}.md`)
+
+			generateMarkdownReport(summary, benchmarkReports, modelReports, outputPath)
+
+			spinner.succeed(`Markdown report generated at ${outputPath}`)
+		}
+	} catch (error: any) {
+		console.error(chalk.red(`Error generating report: ${error.message}`))
+		console.error(error.stack)
+	} finally {
+		db.close()
+	}
+}

evals/cli/src/commands/run.ts

@@ -0,0 +1,133 @@
+import * as path from "path"
+import { v4 as uuidv4 } from "uuid"
+import chalk from "chalk"
+import ora from "ora"
+import { getAdapter } from "../adapters"
+import { ResultsDatabase } from "../db"
+import { spawnVSCode, cleanupVSCode } from "../utils/vscode"
+import { sendTaskToServer } from "../utils/task"
+import { storeTaskResult } from "../utils/results"
+
+interface RunOptions {
+	benchmark?: string
+	model: string
+	count?: number
+	apiKey?: string
+}
+
+/**
+ * Handler for the run command
+ * @param options Command options
+ */
+export async function runHandler(options: RunOptions): Promise<void> {
+	// Determine which benchmarks to run
+	const benchmarks = options.benchmark ? [options.benchmark] : ["exercism"] // Default to exercism for now
+	const model = options.model
+	const count = options.count || Infinity
+
+	console.log(chalk.blue(`Running evaluations for model: ${model}`))
+	console.log(chalk.blue(`Benchmarks: ${benchmarks.join(", ")}`))
+
+	// Create a run for each benchmark
+	for (const benchmark of benchmarks) {
+		const runId = uuidv4()
+		const db = new ResultsDatabase()
+
+		console.log(chalk.green(`\nStarting run for benchmark: ${benchmark}`))
+
+		// Create run in database
+		db.createRun(runId, model, benchmark)
+
+		// Get adapter for this benchmark
+		try {
+			const adapter = getAdapter(benchmark)
+
+			// List tasks
+			const spinner = ora("Listing tasks...").start()
+			const tasks = await adapter.listTasks()
+			spinner.succeed(`Found ${tasks.length} tasks for ${benchmark}`)
+
+			// Limit number of tasks if specified
+			const tasksToRun = tasks.slice(0, count)
+
+			console.log(chalk.blue(`Running ${tasksToRun.length} tasks...`))
+
+			// Run each task
+			for (let i = 0; i < tasksToRun.length; i++) {
+				const task = tasksToRun[i]
+
+				console.log(chalk.cyan(`\nTask ${i + 1}/${tasksToRun.length}: ${task.name}`))
+
+				// Prepare task
+				const prepareSpinner = ora("Preparing task...").start()
+				const preparedTask = await adapter.prepareTask(task.id)
+				prepareSpinner.succeed("Task prepared")
+
+				// Spawn VSCode
+				console.log("Spawning VSCode...")
+				await spawnVSCode(preparedTask.workspacePath)
+
+				// Send task to server
+				const sendSpinner = ora("Sending task to server...").start()
+				try {
+					const result = await sendTaskToServer(preparedTask.description, options.apiKey)
+					sendSpinner.succeed("Task completed")
+
+					// Verify result
+					const verifySpinner = ora("Verifying result...").start()
+					const verification = await adapter.verifyResult(preparedTask, result)
+
+					if (verification.success) {
+						verifySpinner.succeed(
+							`Verification successful: ${verification.metrics.testsPassed}/${verification.metrics.testsTotal} tests passed`,
+						)
+					} else {
+						verifySpinner.fail(
+							`Verification failed: ${verification.metrics.testsPassed}/${verification.metrics.testsTotal} tests passed`,
+						)
+					}
+
+					// Store result
+					const storeSpinner = ora("Storing result...").start()
+					await storeTaskResult(runId, preparedTask, result, verification)
+					storeSpinner.succeed("Result stored")
+
+					console.log(chalk.green(`Task completed. Success: ${verification.success}`))
+
+					// Clean up VS Code and temporary files
+					const cleanupSpinner = ora("Cleaning up...").start()
+					try {
+						await cleanupVSCode(preparedTask.workspacePath)
+						cleanupSpinner.succeed("Cleanup completed")
+					} catch (cleanupError: any) {
+						cleanupSpinner.fail(`Cleanup failed: ${cleanupError.message}`)
+						console.error(chalk.yellow(cleanupError.stack))
+					}
+				} catch (error: any) {
+					sendSpinner.fail(`Task failed: ${error.message}`)
+					console.error(chalk.red(error.stack))
+
+					// Clean up VS Code and temporary files even if the task failed
+					const cleanupSpinner = ora("Cleaning up...").start()
+					try {
+						await cleanupVSCode(preparedTask.workspacePath)
+						cleanupSpinner.succeed("Cleanup completed")
+					} catch (cleanupError: any) {
+						cleanupSpinner.fail(`Cleanup failed: ${cleanupError.message}`)
+						console.error(chalk.yellow(cleanupError.stack))
+					}
+				}
+			}
+
+			// Mark run as complete
+			db.completeRun(runId)
+
+			console.log(chalk.green(`\nRun complete for benchmark: ${benchmark}`))
+		} catch (error: any) {
+			console.error(chalk.red(`Error running benchmark ${benchmark}: ${error.message}`))
+			console.error(error.stack)
+		}
+	}
+
+	console.log(chalk.green("\nAll evaluations complete"))
+}

evals/cli/src/commands/setup.ts

@@ -0,0 +1,72 @@
+import * as path from "path"
+import * as fs from "fs"
+import execa from "execa"
+import chalk from "chalk"
+import ora from "ora"
+import { getAllAdapters } from "../adapters/index"
+import { BenchmarkAdapter } from "../adapters/types"
+
+interface SetupOptions {
+	benchmarks: string
+}
+
+/**
+ * Handler for the setup command
+ * @param options Command options
+ */
+export async function setupHandler(options: SetupOptions): Promise<void> {
+	const benchmarks = options.benchmarks.split(",")
+
+	console.log(chalk.blue(`Setting up benchmarks: ${benchmarks.join(", ")}`))
+
+	// Create directories
+	const evalsDir = path.resolve(__dirname, "../../../")
+	const reposDir = path.join(evalsDir, "repositories")
+	const resultsDir = path.join(evalsDir, "results")
+
+	const spinner = ora("Creating directory structure").start()
+
+	try {
+		fs.mkdirSync(reposDir, { recursive: true })
+		fs.mkdirSync(resultsDir, { recursive: true })
+		fs.mkdirSync(path.join(resultsDir, "runs"), { recursive: true })
+		fs.mkdirSync(path.join(resultsDir, "reports"), { recursive: true })
+		spinner.succeed("Directory structure created")
+	} catch (error) {
+		spinner.fail(`Failed to create directory structure: ${(error as Error).message}`)
+		throw error
+	}
+
+	// Set up each benchmark
+	try {
+		const adapters = getAllAdapters().filter((adapter: BenchmarkAdapter) => benchmarks.includes(adapter.name))
+
+		if (adapters.length === 0) {
+			console.warn(chalk.yellow("No valid benchmarks specified. Available benchmarks:"))
+			console.warn(
+				chalk.yellow(
+					getAllAdapters()
+						.map((a: BenchmarkAdapter) => a.name)
+						.join(", "),
+				),
+			)
+			return
+		}
+
+		for (const adapter of adapters) {
+			const setupSpinner = ora(`Setting up ${adapter.name}...`).start()
+			try {
+				await adapter.setup()
+				setupSpinner.succeed(`${adapter.name} setup complete`)
+			} catch (error) {
+				setupSpinner.fail(`Failed to set up ${adapter.name}: ${(error as Error).message}`)
+				throw error
+			}
+		}
+
+		console.log(chalk.green("Setup complete"))
+	} catch (error) {
+		console.error(chalk.red(`Setup failed: ${(error as Error).message}`))
+		throw error
+	}
+}

evals/cli/src/db/index.ts

@@ -0,0 +1,211 @@
+import * as path from "path"
+import * as fs from "fs"
+import Database from "better-sqlite3"
+import { SCHEMA } from "./schema"
+
+const EVALS_DIR = path.resolve(__dirname, "../../../")
+
+/**
+ * Database class for storing evaluation results
+ */
+export class ResultsDatabase {
+	db: Database.Database
+
+	constructor() {
+		// Ensure results directory exists
+		const resultsDir = path.join(EVALS_DIR, "results")
+		fs.mkdirSync(resultsDir, { recursive: true })
+
+		// Create database file
+		const dbPath = path.join(resultsDir, "evals.db")
+		this.db = new Database(dbPath)
+
+		// Initialize schema
+		this.initSchema()
+	}
+
+	/**
+	 * Initialize the database schema
+	 */
+	private initSchema(): void {
+		this.db.exec(SCHEMA)
+	}
+
+	/**
+	 * Create a new evaluation run
+	 * @param id Run ID
+	 * @param model Model name
+	 * @param benchmark Benchmark name
+	 */
+	createRun(id: string, model: string, benchmark: string): void {
+		const stmt = this.db.prepare(`
+      INSERT INTO runs (id, timestamp, model, benchmark)
+      VALUES (?, ?, ?, ?)
+    `)
+
+		stmt.run(id, Date.now(), model, benchmark)
+	}
+
+	/**
+	 * Mark a run as completed
+	 * @param id Run ID
+	 */
+	completeRun(id: string): void {
+		const stmt = this.db.prepare(`
+      UPDATE runs SET completed = 1 WHERE id = ?
+    `)
+
+		stmt.run(id)
+	}
+
+	/**
+	 * Create a new task
+	 * @param id Task ID
+	 * @param runId Run ID
+	 * @param taskId Original task ID
+	 */
+	createTask(id: string, runId: string, taskId: string): void {
+		const stmt = this.db.prepare(`
+      INSERT INTO tasks (id, run_id, task_id, timestamp)
+      VALUES (?, ?, ?, ?)
+    `)
+
+		stmt.run(id, runId, taskId, Date.now())
+	}
+
+	/**
+	 * Mark a task as completed
+	 * @param id Task ID
+	 * @param success Whether the task was successful
+	 * @param toolCalls Total tool calls
+	 * @param toolFailures Total tool failures
+	 */
+	completeTask(id: string, success: boolean, toolCalls: number = 0, toolFailures: number = 0): void {
+		const stmt = this.db.prepare(`
+      UPDATE tasks 
+      SET success = ?, total_tool_calls = ?, total_tool_failures = ? 
+      WHERE id = ?
+    `)
+
+		stmt.run(success ? 1 : 0, toolCalls, toolFailures, id)
+	}
+
+	/**
+	 * Add a metric to a task
+	 * @param taskId Task ID
+	 * @param name Metric name
+	 * @param value Metric value
+	 */
+	addMetric(taskId: string, name: string, value: number): void {
+		const stmt = this.db.prepare(`
+      INSERT INTO metrics (task_id, name, value)
+      VALUES (?, ?, ?)
+    `)
+
+		stmt.run(taskId, name, value)
+	}
+
+	/**
+	 * Add a tool call record
+	 * @param taskId Task ID
+	 * @param toolName Tool name
+	 * @param callCount Number of calls
+	 * @param failureCount Number of failures
+	 */
+	addToolCall(taskId: string, toolName: string, callCount: number, failureCount: number): void {
+		const stmt = this.db.prepare(`
+      INSERT INTO tool_calls (task_id, tool_name, call_count, failure_count)
+      VALUES (?, ?, ?, ?)
+    `)
+
+		stmt.run(taskId, toolName, callCount, failureCount)
+	}
+
+	/**
+	 * Add a file record
+	 * @param taskId Task ID
+	 * @param filePath File path
+	 * @param status File status (created, modified, deleted)
+	 */
+	addFile(taskId: string, filePath: string, status: "created" | "modified" | "deleted"): void {
+		const stmt = this.db.prepare(`
+      INSERT INTO files (task_id, path, status)
+      VALUES (?, ?, ?)
+    `)
+
+		stmt.run(taskId, filePath, status)
+	}
+
+	/**
+	 * Get all runs
+	 * @returns Array of runs
+	 */
+	getRuns(): any[] {
+		const stmt = this.db.prepare(`
+      SELECT * FROM runs ORDER BY timestamp DESC
+    `)
+
+		return stmt.all()
+	}
+
+	/**
+	 * Get all tasks for a run
+	 * @param runId Run ID
+	 * @returns Array of tasks
+	 */
+	getRunTasks(runId: string): any[] {
+		const stmt = this.db.prepare(`
+      SELECT * FROM tasks WHERE run_id = ? ORDER BY timestamp ASC
+    `)
+
+		return stmt.all(runId)
+	}
+
+	/**
+	 * Get all metrics for a task
+	 * @param taskId Task ID
+	 * @returns Array of metrics
+	 */
+	getTaskMetrics(taskId: string): any[] {
+		const stmt = this.db.prepare(`
+      SELECT name, value FROM metrics WHERE task_id = ?
+    `)
+
+		return stmt.all(taskId)
+	}
+
+	/**
+	 * Get all tool calls for a task
+	 * @param taskId Task ID
+	 * @returns Array of tool calls
+	 */
+	getTaskToolCalls(taskId: string): any[] {
+		const stmt = this.db.prepare(`
+      SELECT tool_name, call_count, failure_count 
+      FROM tool_calls 
+      WHERE task_id = ?
+    `)
+
+		return stmt.all(taskId)
+	}
+
+	/**
+	 * Get all files for a task
+	 * @param taskId Task ID
+	 * @returns Array of files
+	 */
+	getTaskFiles(taskId: string): any[] {
+		const stmt = this.db.prepare(`
+      SELECT path, status FROM files WHERE task_id = ?
+    `)
+
+		return stmt.all(taskId)
+	}
+
+	/**
+	 * Close the database connection
+	 */
+	close(): void {
+		this.db.close()
+	}
+}

evals/cli/src/db/schema.ts

@@ -0,0 +1,48 @@
+/**
+ * SQL schema for the evaluation database
+ */
+export const SCHEMA = `
+CREATE TABLE IF NOT EXISTS runs (
+  id TEXT PRIMARY KEY,
+  timestamp INTEGER NOT NULL,
+  model TEXT NOT NULL,
+  benchmark TEXT NOT NULL,
+  completed INTEGER NOT NULL DEFAULT 0
+);
+
+CREATE TABLE IF NOT EXISTS tasks (
+  id TEXT PRIMARY KEY,
+  run_id TEXT NOT NULL,
+  task_id TEXT NOT NULL,
+  timestamp INTEGER NOT NULL,
+  success INTEGER NOT NULL DEFAULT 0,
+  total_tool_calls INTEGER DEFAULT 0,
+  total_tool_failures INTEGER DEFAULT 0,
+  FOREIGN KEY (run_id) REFERENCES runs(id)
+);
+
+CREATE TABLE IF NOT EXISTS metrics (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  task_id TEXT NOT NULL,
+  name TEXT NOT NULL,
+  value REAL NOT NULL,
+  FOREIGN KEY (task_id) REFERENCES tasks(id)
+);
+
+CREATE TABLE IF NOT EXISTS tool_calls (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  task_id TEXT NOT NULL,
+  tool_name TEXT NOT NULL,
+  call_count INTEGER NOT NULL,
+  failure_count INTEGER NOT NULL,
+  FOREIGN KEY (task_id) REFERENCES tasks(id)
+);
+
+CREATE TABLE IF NOT EXISTS files (
+  id INTEGER PRIMARY KEY AUTOINCREMENT,
+  task_id TEXT NOT NULL,
+  path TEXT NOT NULL,
+  status TEXT NOT NULL,
+  FOREIGN KEY (task_id) REFERENCES tasks(id)
+);
+`

evals/cli/src/index.ts

@@ -0,0 +1,70 @@
+#!/usr/bin/env node
+import { Command } from "commander"
+import chalk from "chalk"
+import { setupHandler } from "./commands/setup"
+import { runHandler } from "./commands/run"
+import { reportHandler } from "./commands/report"
+
+// Create the CLI program
+const program = new Command()
+
+// Set up CLI metadata
+program.name("cline-eval").description("CLI tool for orchestrating Cline evaluations across multiple benchmarks").version("0.1.0")
+
+// Setup command
+program
+	.command("setup")
+	.description("Clone and set up benchmark repositories")
+	.option(
+		"-b, --benchmarks <benchmarks>",
+		"Comma-separated list of benchmarks to set up",
+		"exercism,swe-bench,swelancer,multi-swe",
+	)
+	.action(async (options) => {
+		try {
+			await setupHandler(options)
+		} catch (error) {
+			console.error(chalk.red(`Error during setup: ${error instanceof Error ? error.message : String(error)}`))
+			process.exit(1)
+		}
+	})
+
+// Run command
+program
+	.command("run")
+	.description("Run evaluations")
+	.option("-b, --benchmark <benchmark>", "Specific benchmark to run")
+	.option("-m, --model <model>", "Model to evaluate", "claude-3-opus-20240229")
+	.option("-c, --count <count>", "Number of tasks to run", parseInt)
+	.option("-k, --api-key <apiKey>", "Cline API key to use for evaluations")
+	.action(async (options) => {
+		try {
+			await runHandler(options)
+		} catch (error) {
+			console.error(chalk.red(`Error during run: ${error instanceof Error ? error.message : String(error)}`))
+			process.exit(1)
+		}
+	})
+
+// Report command
+program
+	.command("report")
+	.description("Generate reports")
+	.option("-f, --format <format>", "Report format (json, markdown)", "markdown")
+	.option("-o, --output <path>", "Output path for the report")
+	.action(async (options) => {
+		try {
+			await reportHandler(options)
+		} catch (error) {
+			console.error(chalk.red(`Error generating report: ${error instanceof Error ? error.message : String(error)}`))
+			process.exit(1)
+		}
+	})
+
+// Parse command line arguments
+program.parse(process.argv)
+
+// If no arguments provided, show help
+if (process.argv.length === 2) {
+	program.help()
+}

evals/cli/src/utils/extensions.ts

@@ -0,0 +1,131 @@
+import execa from "execa"
+import * as fs from "fs"
+import * as path from "path"
+import * as os from "os"
+
+/**
+ * List of VSCode extensions to install for evaluation environments
+ * These extensions provide language support and other useful features
+ */
+export const REQUIRED_EXTENSIONS = [
+	"golang.go", // Go language support
+	"dbaeumer.vscode-eslint", // ESLint support
+	"redhat.java", // Java support
+	"ms-python.python", // Python support
+	"rust-lang.rust-analyzer", // Rust support
+	"ms-vscode.cpptools", // C/C++ support
+]
+
+/**
+ * Install required VSCode extensions in the specified extensions directory
+ * @param extensionsDir The directory where extensions should be installed
+ * @returns Promise that resolves when all extensions are installed
+ */
+export async function installRequiredExtensions(extensionsDir: string): Promise<void> {
+	console.log("Installing required VSCode extensions...")
+
+	// Create the extensions directory if it doesn't exist
+	if (!fs.existsSync(extensionsDir)) {
+		fs.mkdirSync(extensionsDir, { recursive: true })
+	}
+
+	// Install each extension
+	for (const extension of REQUIRED_EXTENSIONS) {
+		try {
+			console.log(`Installing extension: ${extension}...`)
+			await execa("code", ["--extensions-dir", extensionsDir, "--install-extension", extension, "--force"])
+			console.log(`✅ Extension ${extension} installed successfully`)
+		} catch (error: any) {
+			console.warn(`⚠️ Failed to install extension ${extension}: ${error.message}`)
+			// Continue with other extensions even if one fails
+		}
+	}
+
+	console.log("✅ All required extensions installed")
+}
+
+/**
+ * Check if a VSCode extension is installed in the specified directory
+ * @param extensionsDir The directory to check for installed extensions
+ * @param extensionId The ID of the extension to check
+ * @returns True if the extension is installed, false otherwise
+ */
+export function isExtensionInstalled(extensionsDir: string, extensionId: string): boolean {
+	// Extensions are installed in directories named publisher.name-version
+	// We need to check if any directory starts with the extensionId
+	const extensionPrefix = extensionId.toLowerCase() + "-"
+
+	try {
+		const files = fs.readdirSync(extensionsDir)
+		return files.some((file) => {
+			const lowerCaseFile = file.toLowerCase()
+			return lowerCaseFile === extensionId.toLowerCase() || lowerCaseFile.startsWith(extensionPrefix)
+		})
+	} catch (error) {
+		return false
+	}
+}
+
+/**
+ * Get the path to the VSCode settings file in the specified user data directory
+ * @param userDataDir The VSCode user data directory
+ * @returns The path to the settings.json file
+ */
+export function getSettingsPath(userDataDir: string): string {
+	const settingsDir = path.join(userDataDir, "User")
+	fs.mkdirSync(settingsDir, { recursive: true })
+	return path.join(settingsDir, "settings.json")
+}
+
+/**
+ * Configure extension settings in the VSCode user data directory
+ * @param userDataDir The VSCode user data directory
+ */
+export function configureExtensionSettings(userDataDir: string): void {
+	const settingsPath = getSettingsPath(userDataDir)
+
+	// Read existing settings if they exist
+	let settings = {}
+	if (fs.existsSync(settingsPath)) {
+		try {
+			settings = JSON.parse(fs.readFileSync(settingsPath, "utf8"))
+		} catch (error) {
+			console.warn(`Error reading settings file: ${error}`)
+		}
+	}
+
+	// Add or update extension-specific settings
+	const updatedSettings = {
+		...settings,
+		// Go extension settings
+		"go.toolsManagement.autoUpdate": false,
+		"go.survey.prompt": false,
+
+		// ESLint settings
+		"eslint.enable": true,
+		"eslint.run": "onSave",
+
+		// Java settings
+		"java.configuration.checkProjectSettingsExclusions": false,
+		"java.configure.checkForOutdatedExtensions": false,
+		"java.help.firstView": false,
+
+		// Python settings
+		"python.experiments.enabled": false,
+		"python.showStartPage": false,
+
+		// Rust settings
+		"rust-analyzer.checkOnSave.command": "check",
+
+		// C/C++ settings
+		"C_Cpp.intelliSenseEngine": "default",
+
+		// General extension settings
+		"extensions.autoUpdate": false,
+		"extensions.ignoreRecommendations": true,
+	}
+
+	// Write updated settings
+	fs.writeFileSync(settingsPath, JSON.stringify(updatedSettings, null, 2))
+	console.log("✅ Extension settings configured")
+}

evals/cli/src/utils/markdown.ts

@@ -0,0 +1,109 @@
+import * as fs from "fs"
+import * as path from "path"
+
+/**
+ * Generate a markdown report from evaluation results
+ * @param summary Overall summary
+ * @param benchmarkReports Benchmark-specific reports
+ * @param modelReports Model-specific reports
+ * @param outputPath Output file path
+ */
+export function generateMarkdownReport(
+	summary: any,
+	benchmarkReports: Record<string, any>,
+	modelReports: Record<string, any>,
+	outputPath: string,
+): void {
+	let markdown = `# Cline Evaluation Report\n\n`
+
+	// Generate summary section
+	markdown += `## Summary\n\n`
+	markdown += `- **Total Runs:** ${summary.runs}\n`
+	markdown += `- **Models:** ${summary.models.join(", ")}\n`
+	markdown += `- **Benchmarks:** ${summary.benchmarks.join(", ")}\n`
+	markdown += `- **Total Tasks:** ${summary.tasks}\n`
+	markdown += `- **Success Rate:** ${(summary.successRate * 100).toFixed(2)}%\n`
+	markdown += `- **Average Tokens:** ${Math.round(summary.averageTokens)}\n`
+	markdown += `- **Average Cost:** $${summary.averageCost.toFixed(4)}\n`
+	markdown += `- **Average Duration:** ${(summary.averageDuration / 1000).toFixed(2)}s\n`
+	markdown += `- **Total Tool Calls:** ${summary.totalToolCalls}\n`
+	markdown += `- **Tool Success Rate:** ${(summary.toolSuccessRate * 100).toFixed(2)}%\n\n`
+
+	// Generate tool usage section
+	markdown += `## Tool Usage\n\n`
+	markdown += `| Tool | Calls | Failures | Success Rate |\n`
+	markdown += `| ---- | ----- | -------- | ------------ |\n`
+
+	for (const [toolName, metrics] of Object.entries(summary.toolUsage)) {
+		const calls = (metrics as any).calls
+		const failures = (metrics as any).failures
+		const successRate = calls > 0 ? (1 - failures / calls) * 100 : 100
+
+		markdown += `| ${toolName} | ${calls} | ${failures} | ${successRate.toFixed(2)}% |\n`
+	}
+
+	// Generate benchmark results section
+	markdown += `\n## Benchmark Results\n\n`
+
+	for (const [benchmark, report] of Object.entries(benchmarkReports)) {
+		markdown += `### ${benchmark}\n\n`
+		markdown += `- **Runs:** ${report.runs}\n`
+		markdown += `- **Models:** ${report.models.join(", ")}\n`
+		markdown += `- **Tasks:** ${report.tasks}\n`
+		markdown += `- **Success Rate:** ${(report.successRate * 100).toFixed(2)}%\n`
+		markdown += `- **Average Tokens:** ${Math.round(report.averageTokens)}\n`
+		markdown += `- **Average Cost:** $${report.averageCost.toFixed(4)}\n`
+		markdown += `- **Average Duration:** ${(report.averageDuration / 1000).toFixed(2)}s\n\n`
+	}
+
+	// Generate model results section
+	markdown += `## Model Results\n\n`
+
+	for (const [model, report] of Object.entries(modelReports)) {
+		markdown += `### ${model}\n\n`
+		markdown += `- **Runs:** ${report.runs}\n`
+		markdown += `- **Benchmarks:** ${report.benchmarks.join(", ")}\n`
+		markdown += `- **Tasks:** ${report.tasks}\n`
+		markdown += `- **Success Rate:** ${(report.successRate * 100).toFixed(2)}%\n`
+		markdown += `- **Average Tokens:** ${Math.round(report.averageTokens)}\n`
+		markdown += `- **Average Cost:** $${report.averageCost.toFixed(4)}\n`
+		markdown += `- **Average Duration:** ${(report.averageDuration / 1000).toFixed(2)}s\n\n`
+	}
+
+	// Add charts using Mermaid
+	markdown += `## Charts\n\n`
+
+	// Success rate by benchmark chart
+	markdown += `### Success Rate by Benchmark\n\n`
+	markdown += "```mermaid\n"
+	markdown += "graph TD\n"
+	markdown += "  title[Success Rate by Benchmark]\n"
+	markdown += "  style title fill:none,stroke:none\n\n"
+
+	for (const [benchmark, report] of Object.entries(benchmarkReports)) {
+		const successRate = (report.successRate * 100).toFixed(2)
+		markdown += `  ${benchmark}[${benchmark}: ${successRate}%]\n`
+	}
+
+	markdown += "```\n\n"
+
+	// Success rate by model chart
+	markdown += `### Success Rate by Model\n\n`
+	markdown += "```mermaid\n"
+	markdown += "graph TD\n"
+	markdown += "  title[Success Rate by Model]\n"
+	markdown += "  style title fill:none,stroke:none\n\n"
+
+	for (const [model, report] of Object.entries(modelReports)) {
+		const successRate = (report.successRate * 100).toFixed(2)
+		markdown += `  ${model.replace(/[-\.]/g, "_")}[${model}: ${successRate}%]\n`
+	}
+
+	markdown += "```\n\n"
+
+	// Add timestamp
+	markdown += `\n\n---\n\nReport generated on ${new Date().toISOString()}\n`
+
+	// Write markdown to file
+	fs.writeFileSync(outputPath, markdown)
+}

evals/cli/src/utils/results.ts

@@ -0,0 +1,79 @@
+import { v4 as uuidv4 } from "uuid"
+import { ResultsDatabase } from "../db"
+import { Task } from "../adapters/types"
+
+/**
+ * Store task result in the database
+ * @param runId The run ID
+ * @param task The task that was executed
+ * @param result The result from the test server
+ * @param verification The verification result
+ */
+export async function storeTaskResult(runId: string, task: Task, result: any, verification: any): Promise<void> {
+	const db = new ResultsDatabase()
+	const taskId = uuidv4()
+
+	try {
+		// Extract metrics from the result
+		const { metrics } = result
+		const totalToolCalls = metrics?.totalToolCalls || 0
+		const totalToolFailures = metrics?.totalToolFailures || 0
+
+		// Create task with tool metrics
+		db.createTask(taskId, runId, task.id)
+		db.completeTask(taskId, verification.success, totalToolCalls, totalToolFailures)
+
+		// Store metrics
+		if (metrics) {
+			// Store token metrics
+			if (metrics.tokensIn) db.addMetric(taskId, "tokensIn", metrics.tokensIn)
+			if (metrics.tokensOut) db.addMetric(taskId, "tokensOut", metrics.tokensOut)
+			if (metrics.cost) db.addMetric(taskId, "cost", metrics.cost)
+			if (metrics.duration) db.addMetric(taskId, "duration", metrics.duration)
+
+			// Store tool call metrics
+			if (metrics.toolCalls) {
+				for (const [toolName, callCount] of Object.entries(metrics.toolCalls)) {
+					const failureCount = metrics.toolFailures?.[toolName] || 0
+					db.addToolCall(taskId, toolName, callCount as number, failureCount)
+				}
+			}
+		}
+
+		// Store verification metrics
+		if (verification.metrics) {
+			for (const [key, value] of Object.entries(verification.metrics)) {
+				if (typeof value === "number") {
+					db.addMetric(taskId, key, value)
+				}
+			}
+		}
+
+		// Store file changes
+		if (result.files) {
+			// Store created files
+			if (result.files.created) {
+				for (const file of result.files.created) {
+					db.addFile(taskId, file, "created")
+				}
+			}
+
+			// Store modified files
+			if (result.files.modified) {
+				for (const file of result.files.modified) {
+					db.addFile(taskId, file, "modified")
+				}
+			}
+
+			// Store deleted files
+			if (result.files.deleted) {
+				for (const file of result.files.deleted) {
+					db.addFile(taskId, file, "deleted")
+				}
+			}
+		}
+	} finally {
+		// Close the database connection
+		db.close()
+	}
+}

evals/cli/src/utils/task.ts

@@ -0,0 +1,52 @@
+import fetch from "node-fetch"
+import chalk from "chalk"
+
+/**
+ * Send a task to the Cline test server
+ * @param task The task description to send
+ * @param apiKey Optional Cline API key to use for the task
+ * @returns The result of the task execution
+ */
+export async function sendTaskToServer(task: string, apiKey?: string): Promise<any> {
+	const SERVER_URL = "http://localhost:9876/task"
+
+	try {
+		console.log(chalk.blue(`Sending task to server: ${task.substring(0, 100)}${task.length > 100 ? "..." : ""}`))
+
+		const response = await fetch(SERVER_URL, {
+			method: "POST",
+			headers: {
+				"Content-Type": "application/json",
+			},
+			body: JSON.stringify({
+				task,
+				apiKey,
+			}),
+		})
+
+		if (!response.ok) {
+			const errorText = await response.text()
+			throw new Error(`Server responded with status ${response.status}: ${errorText}`)
+		}
+
+		const result = await response.json()
+
+		if (!result.success) {
+			throw new Error(`Task execution failed: ${result.error || "Unknown error"}`)
+		}
+
+		if (result.timeout) {
+			throw new Error("Task execution timed out")
+		}
+
+		return result
+	} catch (error: any) {
+		if (error.code === "ECONNREFUSED") {
+			throw new Error(
+				"Could not connect to the test server. Make sure VSCode is running with the Cline extension and the test server is active.",
+			)
+		}
+
+		throw error
+	}
+}

evals/cli/src/utils/vscode.ts

@@ -0,0 +1,570 @@
+import execa from "execa"
+import * as path from "path"
+import * as fs from "fs"
+import fetch from "node-fetch"
+import * as os from "os"
+import * as child_process from "child_process"
+import { installRequiredExtensions, configureExtensionSettings } from "./extensions"
+
+// Store temporary directories for cleanup
+interface VSCodeResources {
+	tempUserDataDir: string
+	tempExtensionsDir: string
+	vscodePid?: number
+}
+
+// Global map to track resources for each workspace
+const workspaceResources = new Map<string, VSCodeResources>()
+
+/**
+ * Spawn a VSCode instance with the Cline extension
+ * @param workspacePath The workspace path to open
+ * @param vsixPath Optional path to a VSIX file to install
+ * @returns The resources created for this VS Code instance
+ */
+export async function spawnVSCode(workspacePath: string, vsixPath?: string): Promise<VSCodeResources> {
+	// Ensure the workspace path exists
+	if (!fs.existsSync(workspacePath)) {
+		throw new Error(`Workspace path does not exist: ${workspacePath}`)
+	}
+
+	// If no VSIX path is provided, build one with IS_TEST=true
+	if (!vsixPath) {
+		try {
+			// Build the VSIX with IS_TEST=true
+			console.log("Building test VSIX...")
+			const clineRoot = path.resolve(process.cwd(), "..", "..")
+			await execa("npx", ["vsce", "package"], {
+				cwd: clineRoot,
+				env: {
+					IS_TEST: "true",
+				},
+				stdio: "inherit",
+			})
+
+			// Find the generated VSIX file
+			const files = fs.readdirSync(clineRoot)
+			const vsixFile = files.find((file) => file.endsWith(".vsix"))
+			if (vsixFile) {
+				vsixPath = path.join(clineRoot, vsixFile)
+				console.log(`Using built VSIX: ${vsixPath}`)
+			} else {
+				console.warn("Could not find generated VSIX file")
+			}
+		} catch (error) {
+			console.warn("Failed to build test VSIX:", error)
+		}
+	}
+
+	// Create a temporary user data directory for this VS Code instance
+	const tempUserDataDir = path.join(os.tmpdir(), `vscode-cline-eval-${Date.now()}`)
+	fs.mkdirSync(tempUserDataDir, { recursive: true })
+	console.log(`Created temporary user data directory: ${tempUserDataDir}`)
+
+	// Create a temporary extensions directory to ensure no other extensions are loaded
+	const tempExtensionsDir = path.join(os.tmpdir(), `vscode-cline-eval-ext-${Date.now()}`)
+	fs.mkdirSync(tempExtensionsDir, { recursive: true })
+	console.log(`Created temporary extensions directory: ${tempExtensionsDir}`)
+
+	// Create settings.json in the temporary user data directory to disable workspace trust
+	// and configure Cline to auto-open on startup
+	const settingsDir = path.join(tempUserDataDir, "User")
+	fs.mkdirSync(settingsDir, { recursive: true })
+	const settingsPath = path.join(settingsDir, "settings.json")
+	const settings = {
+		// Disable workspace trust
+		"security.workspace.trust.enabled": false,
+		"security.workspace.trust.startupPrompt": "never",
+		"security.workspace.trust.banner": "never",
+		"security.workspace.trust.emptyWindow": true,
+
+		// Configure startup behavior
+		"workbench.startupEditor": "none",
+
+		// Auto-open Cline on startup
+		"cline.autoOpenOnStartup": true,
+
+		// Show the activity bar and sidebar
+		"workbench.activityBar.visible": true,
+		"workbench.sideBar.visible": true,
+		"workbench.view.extension.saoudrizwan.claude-dev-ActivityBar.visible": true,
+		"workbench.view.alwaysShowHeaderActions": true,
+		"workbench.editor.openSideBySideDirection": "right",
+
+		// Disable GitLens from opening automatically
+		"gitlens.views.repositories.autoReveal": false,
+		"gitlens.views.fileHistory.autoReveal": false,
+		"gitlens.views.lineHistory.autoReveal": false,
+		"gitlens.views.compare.autoReveal": false,
+		"gitlens.views.search.autoReveal": false,
+		"gitlens.showWelcomeOnInstall": false,
+		"gitlens.showWhatsNewAfterUpgrades": false,
+
+		// Disable other extensions that might compete for startup focus
+		"extensions.autoUpdate": false,
+	}
+	fs.writeFileSync(settingsPath, JSON.stringify(settings, null, 2))
+	console.log(`Created settings.json to disable workspace trust and auto-open Cline`)
+
+	// Create keybindings.json to automatically open Cline on startup
+	const keybindingsPath = path.join(settingsDir, "keybindings.json")
+	const keybindings = [
+		{
+			key: "alt+c",
+			command: "workbench.view.extension.saoudrizwan.claude-dev-ActivityBar",
+			when: "viewContainer.workbench.view.extension.saoudrizwan.claude-dev-ActivityBar.enabled",
+		},
+		{
+			key: "alt+shift+c",
+			command: "cline.openInNewTab",
+			when: "viewContainer.workbench.view.extension.saoudrizwan.claude-dev-ActivityBar.enabled",
+		},
+	]
+	fs.writeFileSync(keybindingsPath, JSON.stringify(keybindings, null, 2))
+	console.log(`Created keybindings.json to help with Cline activation`)
+
+	// Build the command arguments with custom user data directory
+	const args = [
+		// Use a custom user data directory to isolate this instance
+		"--user-data-dir",
+		tempUserDataDir,
+		// Use a custom extensions directory to ensure only our extension is loaded
+		"--extensions-dir",
+		tempExtensionsDir,
+		// Disable workspace trust
+		"--disable-workspace-trust",
+		"-n",
+		workspacePath,
+		// Force the extension to be activated on startup
+		"--start-up-extension",
+		"saoudrizwan.claude-dev",
+		// Run a command on startup to open Cline
+		"--command",
+		"workbench.view.extension.saoudrizwan.claude-dev-ActivityBar",
+		// Additional flags to help with extension activation
+		"--disable-gpu=false",
+		"--max-memory=4096",
+	]
+
+	// Create a startup script to run commands after VS Code launches
+	const startupScriptPath = path.join(settingsDir, "startup.js")
+	const startupScript = `
+		// This script will be executed when VS Code starts
+		setTimeout(() => {
+			// Try to open Cline in the sidebar
+			require('vscode').commands.executeCommand('workbench.view.extension.saoudrizwan.claude-dev-ActivityBar');
+			
+			// Also try to open Cline in a tab as a fallback
+			setTimeout(() => {
+				require('vscode').commands.executeCommand('cline.openInNewTab');
+			}, 5000);
+		}, 5000);
+	`
+	fs.writeFileSync(startupScriptPath, startupScript)
+	console.log(`Created startup script to activate Cline`)
+
+	// If a VSIX is provided, install it
+	if (vsixPath) {
+		if (!fs.existsSync(vsixPath)) {
+			throw new Error(`VSIX file does not exist: ${vsixPath}`)
+		}
+		args.unshift("--install-extension", vsixPath)
+	}
+
+	// Install required extensions
+	console.log("Installing required VSCode extensions...")
+	await installRequiredExtensions(tempExtensionsDir)
+
+	// Configure extension settings
+	console.log("Configuring extension settings...")
+	configureExtensionSettings(tempUserDataDir)
+
+	// Execute the command
+	try {
+		// We don't need to install extensions globally anymore since we're using a custom user data directory
+		// The VSIX will be installed in the isolated environment if provided in the args
+
+		// Launch VS Code
+		console.log("Launching VS Code...")
+		await execa("code", args, {
+			stdio: "inherit",
+		})
+
+		// Wait longer for VSCode to initialize and extension to load
+		console.log("Waiting for VS Code to initialize...")
+		await new Promise((resolve) => setTimeout(resolve, 30000))
+
+		// Create a JavaScript file that will be loaded as a VS Code extension
+		const extensionDir = path.join(tempExtensionsDir, "cline-activator")
+		fs.mkdirSync(extensionDir, { recursive: true })
+
+		// Create package.json for the extension
+		const packageJsonPath = path.join(extensionDir, "package.json")
+		const packageJson = {
+			name: "cline-activator",
+			displayName: "Cline Activator",
+			description: "Activates Cline and starts the test server",
+			version: "0.0.1",
+			engines: {
+				vscode: "^1.60.0",
+			},
+			main: "./extension.js",
+			activationEvents: ["*"],
+			contributes: {
+				commands: [
+					{
+						command: "cline-activator.activate",
+						title: "Activate Cline",
+					},
+				],
+			},
+		}
+		fs.writeFileSync(packageJsonPath, JSON.stringify(packageJson, null, 2))
+
+		// Create extension.js
+		const extensionJsPath = path.join(extensionDir, "extension.js")
+		const extensionJs = `
+			const vscode = require('vscode');
+			
+			/**
+			 * @param {vscode.ExtensionContext} context
+			 */
+			function activate(context) {
+				console.log('Cline Activator is now active!');
+				
+				// Register the command to activate Cline
+				let disposable = vscode.commands.registerCommand('cline-activator.activate', async function () {
+					try {
+						// Make sure the Cline extension is activated
+						const extension = vscode.extensions.getExtension('saoudrizwan.claude-dev');
+						if (!extension) {
+							console.error('Cline extension not found');
+							return;
+						}
+						
+						if (!extension.isActive) {
+							console.log('Activating Cline extension...');
+							await extension.activate();
+						}
+						
+						// Show the Cline sidebar
+						console.log('Opening Cline sidebar...');
+						await vscode.commands.executeCommand('workbench.view.extension.saoudrizwan.claude-dev-ActivityBar');
+						
+						// Wait a moment for the sidebar to initialize
+						await new Promise(resolve => setTimeout(resolve, 2000));
+						
+						// Also open Cline in a tab as a fallback
+						console.log('Opening Cline in a tab...');
+						await vscode.commands.executeCommand('cline.openInNewTab');
+						
+						// Wait a moment for the tab to initialize
+						await new Promise(resolve => setTimeout(resolve, 2000));
+						
+						// Create the test server if it doesn't exist
+						console.log('Creating test server...');
+						
+						// Get the visible webview instance
+						const clineRootPath = '${path.resolve(process.cwd(), "..", "..")}';
+						const visibleWebview = require(path.join(clineRootPath, 'src', 'core', 'webview')).WebviewProvider.getVisibleInstance();
+						if (visibleWebview) {
+							require(path.join(clineRootPath, 'src', 'services', 'test', 'TestServer')).createTestServer(visibleWebview);
+							console.log('Test server created successfully');
+						} else {
+							console.error('No visible webview instance found');
+						}
+					} catch (error) {
+						console.error('Error activating Cline:', error);
+					}
+				});
+				
+				context.subscriptions.push(disposable);
+				
+				// Automatically run the command after a delay
+				setTimeout(() => {
+					vscode.commands.executeCommand('cline-activator.activate');
+				}, 5000);
+			}
+			
+			function deactivate() {}
+			
+			module.exports = {
+				activate,
+				deactivate
+			}
+		`
+		fs.writeFileSync(extensionJsPath, extensionJs)
+		console.log(`Created Cline Activator extension`)
+
+		// Try multiple approaches to activate the extension
+		let serverStarted = false
+
+		// Create an activation script to run in VS Code
+		const activationScriptPath = path.join(settingsDir, "activate-cline.js")
+		const activationScript = `
+			// This script will be executed to activate Cline and start the test server
+			const vscode = require('vscode');
+			
+			// Execute the cline-activator.activate command
+			vscode.commands.executeCommand('cline-activator.activate');
+		`
+		fs.writeFileSync(activationScriptPath, activationScript)
+		console.log(`Created activation script to run in VS Code`)
+
+		// Execute the activation script
+		try {
+			console.log("Executing activation script to start Cline and test server...")
+			await execa(
+				"code",
+				[
+					"--user-data-dir",
+					tempUserDataDir,
+					"--extensions-dir",
+					tempExtensionsDir,
+					"--folder-uri",
+					`file://${workspacePath}`,
+					"--execute",
+					activationScriptPath,
+				],
+				{
+					stdio: "inherit",
+				},
+			)
+
+			// Wait for the test server to start
+			console.log("Waiting for test server to start...")
+			for (let i = 0; i < 30; i++) {
+				try {
+					// Try to connect to the test server
+					const response = await fetch("http://localhost:9876/task", {
+						method: "OPTIONS",
+						headers: {
+							"Content-Type": "application/json",
+						},
+					})
+
+					if (response.status === 204) {
+						console.log("Test server is running!")
+						serverStarted = true
+						break
+					}
+				} catch (error) {
+					// Server not started yet, wait and try again
+					await new Promise((resolve) => setTimeout(resolve, 1000))
+				}
+			}
+		} catch (error) {
+			console.warn("Failed to execute activation script:", error)
+		}
+
+		if (!serverStarted) {
+			console.warn("Test server did not start after multiple attempts")
+			console.log("You may need to manually open the Cline extension in VS Code")
+		}
+
+		// Store the resources for this workspace
+		const resources: VSCodeResources = {
+			tempUserDataDir,
+			tempExtensionsDir,
+		}
+
+		// Store in the global map
+		workspaceResources.set(workspacePath, resources)
+
+		// Return the resources
+		return resources
+	} catch (error: any) {
+		throw new Error(`Failed to spawn VSCode: ${error.message}`)
+	}
+}
+
+/**
+ * Clean up VS Code resources and shut down the test server
+ * @param workspacePath The workspace path to clean up resources for
+ */
+export async function cleanupVSCode(workspacePath: string): Promise<void> {
+	console.log(`Cleaning up VS Code resources for workspace: ${workspacePath}`)
+
+	// Get the resources for this workspace
+	const resources = workspaceResources.get(workspacePath)
+	if (!resources) {
+		console.log(`No resources found for workspace: ${workspacePath}`)
+		return
+	}
+
+	// Try to shut down the test server
+	try {
+		console.log("Shutting down test server...")
+		await fetch("http://localhost:9876/shutdown", {
+			method: "POST",
+			headers: {
+				"Content-Type": "application/json",
+			},
+		}).catch(() => {
+			// Ignore errors, the server might already be down
+		})
+	} catch (error) {
+		console.warn(`Error shutting down test server: ${error}`)
+	}
+
+	// Try to gracefully close VS Code instead of killing it
+	try {
+		console.log("Attempting to gracefully close VS Code...")
+
+		// Create a settings file that will disable the crash reporter and the exit confirmation dialog
+		const settingsDir = path.join(resources.tempUserDataDir, "User")
+		const settingsPath = path.join(settingsDir, "settings.json")
+
+		// Read existing settings if they exist
+		let settings = {}
+		if (fs.existsSync(settingsPath)) {
+			try {
+				settings = JSON.parse(fs.readFileSync(settingsPath, "utf8"))
+			} catch (error) {
+				console.warn(`Error reading settings file: ${error}`)
+			}
+		}
+
+		// Update settings to disable crash reporter and exit confirmation
+		settings = {
+			...settings,
+			"window.confirmBeforeClose": "never",
+			"telemetry.enableCrashReporter": false,
+			"window.restoreWindows": "none",
+			"window.newWindowDimensions": "default",
+		}
+
+		// Write updated settings
+		fs.writeFileSync(settingsPath, JSON.stringify(settings, null, 2))
+
+		// On macOS, use AppleScript to quit VS Code gracefully
+		if (process.platform === "darwin") {
+			try {
+				// First try AppleScript to quit VS Code gracefully
+				await execa("osascript", ["-e", 'tell application "Visual Studio Code" to quit'])
+
+				// Wait a moment for VS Code to close
+				await new Promise((resolve) => setTimeout(resolve, 2000))
+			} catch (appleScriptError) {
+				console.warn(`Error using AppleScript to quit VS Code: ${appleScriptError}`)
+			}
+		} else if (process.platform === "win32") {
+			// On Windows, try to use taskkill without /F first
+			try {
+				await execa("taskkill", ["/IM", "code.exe"])
+
+				// Wait a moment for VS Code to close
+				await new Promise((resolve) => setTimeout(resolve, 2000))
+			} catch (taskkillError) {
+				console.warn(`Error using taskkill to quit VS Code: ${taskkillError}`)
+			}
+		} else {
+			// On Linux, try to use SIGTERM first
+			try {
+				// Find VS Code processes
+				const { stdout } = await execa("ps", ["aux"])
+				const lines = stdout.split("\n")
+
+				for (const line of lines) {
+					if (line.includes(resources.tempUserDataDir)) {
+						const parts = line.trim().split(/\s+/)
+						const pid = parseInt(parts[1])
+
+						if (pid && !isNaN(pid)) {
+							console.log(`Sending SIGTERM to VS Code process with PID: ${pid}`)
+							try {
+								// Use SIGTERM instead of SIGKILL for a graceful shutdown
+								process.kill(pid, "SIGTERM")
+							} catch (killError) {
+								console.warn(`Failed to terminate process ${pid}: ${killError}`)
+							}
+						}
+					}
+				}
+
+				// Wait a moment for VS Code to close
+				await new Promise((resolve) => setTimeout(resolve, 2000))
+			} catch (psError) {
+				console.warn(`Error listing processes: ${psError}`)
+			}
+		}
+
+		// If graceful methods failed, fall back to forceful termination as a last resort
+		// Check if VS Code is still running with the temp user data dir
+		let vsCodeStillRunning = false
+
+		if (process.platform !== "win32") {
+			try {
+				const { stdout } = await execa("ps", ["aux"])
+				vsCodeStillRunning = stdout.split("\n").some((line) => line.includes(resources.tempUserDataDir))
+			} catch (error) {
+				console.warn(`Error checking if VS Code is still running: ${error}`)
+			}
+		} else {
+			try {
+				const { stdout } = await execa("tasklist", ["/FI", `IMAGENAME eq code.exe`])
+				vsCodeStillRunning = stdout.includes("code.exe")
+			} catch (error) {
+				console.warn(`Error checking if VS Code is still running: ${error}`)
+			}
+		}
+
+		// If VS Code is still running, use forceful termination as a last resort
+		if (vsCodeStillRunning) {
+			console.log("Graceful shutdown failed, falling back to forceful termination...")
+
+			if (process.platform === "win32") {
+				try {
+					await execa("taskkill", ["/IM", "code.exe", "/F"])
+				} catch (error) {
+					console.warn(`Error forcefully terminating VS Code: ${error}`)
+				}
+			} else {
+				try {
+					const { stdout } = await execa("ps", ["aux"])
+					const lines = stdout.split("\n")
+
+					for (const line of lines) {
+						if (line.includes(resources.tempUserDataDir)) {
+							const parts = line.trim().split(/\s+/)
+							const pid = parseInt(parts[1])
+
+							if (pid && !isNaN(pid)) {
+								console.log(`Forcefully killing VS Code process with PID: ${pid}`)
+								try {
+									process.kill(pid, "SIGKILL")
+								} catch (killError) {
+									console.warn(`Failed to kill process ${pid}: ${killError}`)
+								}
+							}
+						}
+					}
+				} catch (error) {
+					console.warn(`Error forcefully terminating VS Code: ${error}`)
+				}
+			}
+		}
+	} catch (error) {
+		console.warn(`Error closing VS Code: ${error}`)
+	}
+
+	// Clean up temporary directories
+	try {
+		console.log(`Removing temporary user data directory: ${resources.tempUserDataDir}`)
+		fs.rmSync(resources.tempUserDataDir, { recursive: true, force: true })
+	} catch (error) {
+		console.warn(`Error removing temporary user data directory: ${error}`)
+	}
+
+	try {
+		console.log(`Removing temporary extensions directory: ${resources.tempExtensionsDir}`)
+		fs.rmSync(resources.tempExtensionsDir, { recursive: true, force: true })
+	} catch (error) {
+		console.warn(`Error removing temporary extensions directory: ${error}`)
+	}
+
+	// Remove from the global map
+	workspaceResources.delete(workspacePath)
+
+	console.log("Cleanup completed")
+}

evals/cli/tsconfig.json

@@ -0,0 +1,17 @@
+{
+	"compilerOptions": {
+		"target": "ES2020",
+		"module": "commonjs",
+		"lib": ["ES2020"],
+		"declaration": true,
+		"outDir": "dist",
+		"rootDir": "src",
+		"strict": true,
+		"esModuleInterop": true,
+		"skipLibCheck": true,
+		"forceConsistentCasingInFileNames": true,
+		"resolveJsonModule": true
+	},
+	"include": ["src/**/*"],
+	"exclude": ["node_modules", "dist"]
+}

locales/ar-sa/README.md

@@ -2,7 +2,7 @@
 العربية | <a href="https://github.com/cline/cline/blob/main/locales/es/README.md" target="_blank">الإسبانية</a> | <a href="https://github.com/cline/cline/blob/main/locales/de/README.md" target="_blank">الألمانية</a> | <a href="https://github.com/cline/cline/blob/main/locales/ja/README.md" target="_blank">اليابانية</a> | <a href="https://github.com/cline/cline/blob/main/locales/zh-cn/README.md" target="_blank">الصينية المبسطة</a> | <a href="https://github.com/cline/cline/blob/main/locales/zh-tw/README.md" target="_blank">الصينية التقليدية</a> | <a href="https://github.com/cline/cline/blob/main/locales/pt-BR/README.md" target="_blank">البرتغالية</a>
 </sub></div>
 
-# Cline – \#1 على OpenRouter
+# Cline
 
 <p align="center">
   <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/de/README.md

@@ -1,4 +1,4 @@
-# Cline – \#1 auf OpenRouter
+# Cline
 
 <p align="center">
         <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/es/README.md

@@ -1,4 +1,4 @@
-# Cline – #1 en OpenRouter
+# Cline
 
 <p align="center">
         <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/ja/CONTRIBUTING.md

@@ -1,4 +1,4 @@
-# Clineへの貢献
+# Cline
 
 Clineへの貢献に興味をお持ちいただきありがとうございます。
 

locales/ja/README.md

@@ -1,4 +1,4 @@
-# Cline – OpenRouterでのナンバーワン
+# Cline
 
 <p align="center">
     <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/ko/CONTRIBUTING.md

@@ -1,4 +1,4 @@
-# Cline에 기여하기
+# Cline
 
 Cline에 기여하는 것에 관심을 가져주셔서 감사합니다! 버그 수정, 기능 추가, 문서 개선 등 모든 기여는 Cline을 더욱 스마트하게 만드는 데 기여합니다. 활기차고 환영하는 커뮤니티를 유지하기 위해 모든 구성원은 [행동 강령](CODE_OF_CONDUCT.md)을 준수해야 합니다.
 

locales/ko/README.md

@@ -1,4 +1,4 @@
-# Cline - 최고의 OpenRouter
+# Cline
 
 <p align="center">
     <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/pt-BR/README.md

@@ -1,4 +1,4 @@
-# Cline – #1 no OpenRouter
+# Cline
 
 <p align="center">
         <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/zh-cn/README.md

@@ -1,4 +1,4 @@
-# Cline – OpenRouter 排名第一的 AI 工具
+# Cline
 
 <p align="center">
     <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

locales/zh-tw/README.md

@@ -2,7 +2,7 @@
 <a href="https://github.com/cline/cline/blob/main/README.md" target="_blank">English</a> | <a href="https://github.com/cline/cline/blob/main/locales/es/README.md" target="_blank">Español</a> | <a href="https://github.com/cline/cline/blob/main/locales/de/README.md" target="_blank">Deutsch</a> | <a href="https://github.com/cline/cline/blob/main/locales/ja/README.md" target="_blank">日本語</a> | <a href="https://github.com/cline/cline/blob/main/locales/zh-cn/README.md" target="_blank">简体中文</a> | 繁體中文 | <a href="https://github.com/cline/cline/blob/main/locales/ko/README.md" target="_blank">한국어</a>
 </sub></div>
 
-# Cline – OpenRouter 第一名的 AI 工具
+# Cline
 
 <p align="center">
     <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />

package-lock.json

@@ -1,12 +1,12 @@
 {
 	"name": "claude-dev",
-	"version": "3.9.2",
+	"version": "3.12.0",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "claude-dev",
-			"version": "3.9.2",
+			"version": "3.12.0",
 			"license": "Apache-2.0",
 			"dependencies": {
 				"@anthropic-ai/bedrock-sdk": "^0.12.4",
@@ -23,6 +23,7 @@
 				"@opentelemetry/sdk-node": "^0.39.1",
 				"@opentelemetry/sdk-trace-node": "^1.30.1",
 				"@opentelemetry/semantic-conventions": "^1.30.0",
+				"@sentry/browser": "^9.12.0",
 				"@vscode/codicons": "^0.0.36",
 				"axios": "^1.8.2",
 				"cheerio": "^1.0.0",
@@ -6856,6 +6857,81 @@
 			"integrity": "sha512-831qok9r2t8AlxLko40y2ebgSDhenenCatLVeW/uBtnHPyhHOvG0C7TvfgecV+wHzIm5KUICgzmVpWS+IMEAeg==",
 			"license": "MIT"
 		},
+		"node_modules/@sentry-internal/browser-utils": {
+			"version": "9.12.0",
+			"resolved": "https://registry.npmjs.org/@sentry-internal/browser-utils/-/browser-utils-9.12.0.tgz",
+			"integrity": "sha512-GXuDEG2Ix8DmVtTkjsItWdusk2CvJ6EPWKYVqFKifxt+IAT3ZbhGZd99Rg3wdRmt9xhCNuS4QrDzDTPMPgfdCw==",
+			"license": "MIT",
+			"dependencies": {
+				"@sentry/core": "9.12.0"
+			},
+			"engines": {
+				"node": ">=18"
+			}
+		},
+		"node_modules/@sentry-internal/feedback": {
+			"version": "9.12.0",
+			"resolved": "https://registry.npmjs.org/@sentry-internal/feedback/-/feedback-9.12.0.tgz",
+			"integrity": "sha512-3+UxoT97QIXNSUQS4ATL1FFws0RkUb6PeaQN8CPndI6mFlqTW5tuVVLNg9Eo1seNg7R/dfk6WHCWrYN1NbFFKQ==",
+			"license": "MIT",
+			"dependencies": {
+				"@sentry/core": "9.12.0"
+			},
+			"engines": {
+				"node": ">=18"
+			}
+		},
+		"node_modules/@sentry-internal/replay": {
+			"version": "9.12.0",
+			"resolved": "https://registry.npmjs.org/@sentry-internal/replay/-/replay-9.12.0.tgz",
+			"integrity": "sha512-njEQosFeO/UX+gG+DMRANkPUuz6OIJLb+A1GVylhq9adUgFQydQ9Ay3v7/x1gMhdfHVP6Jeb27qkti0BWYbzBQ==",
+			"license": "MIT",
+			"dependencies": {
+				"@sentry-internal/browser-utils": "9.12.0",
+				"@sentry/core": "9.12.0"
+			},
+			"engines": {
+				"node": ">=18"
+			}
+		},
+		"node_modules/@sentry-internal/replay-canvas": {
+			"version": "9.12.0",
+			"resolved": "https://registry.npmjs.org/@sentry-internal/replay-canvas/-/replay-canvas-9.12.0.tgz",
+			"integrity": "sha512-p8LuKZgWT/CoQBbDOXkSGjWWnc8WsnAayWgna8M/ZFWNITCNEM2rCuqZOyWOElIlrni+M7qoEA3jS7MZe8Ejxw==",
+			"license": "MIT",
+			"dependencies": {
+				"@sentry-internal/replay": "9.12.0",
+				"@sentry/core": "9.12.0"
+			},
+			"engines": {
+				"node": ">=18"
+			}
+		},
+		"node_modules/@sentry/browser": {
+			"version": "9.12.0",
+			"resolved": "https://registry.npmjs.org/@sentry/browser/-/browser-9.12.0.tgz",
+			"integrity": "sha512-4xQYoZqi+VVhNvlhWiwRd57+SMr3Og4sLjuayAA+zIp1Wx/bDcIld697cugLwml/BR+mVJI2eokkgh1CBl6zag==",
+			"license": "MIT",
+			"dependencies": {
+				"@sentry-internal/browser-utils": "9.12.0",
+				"@sentry-internal/feedback": "9.12.0",
+				"@sentry-internal/replay": "9.12.0",
+				"@sentry-internal/replay-canvas": "9.12.0",
+				"@sentry/core": "9.12.0"
+			},
+			"engines": {
+				"node": ">=18"
+			}
+		},
+		"node_modules/@sentry/core": {
+			"version": "9.12.0",
+			"resolved": "https://registry.npmjs.org/@sentry/core/-/core-9.12.0.tgz",
+			"integrity": "sha512-jOqQK/90uzHmsBvkPTj/DAEFvA5poX4ZRyC7LE1zjg4F5jdOp3+M4W3qCy0CkSTu88Zu5VWBoppCU2Bs34XEqg==",
+			"license": "MIT",
+			"engines": {
+				"node": ">=18"
+			}
+		},
 		"node_modules/@sindresorhus/merge-streams": {
 			"version": "4.0.0",
 			"resolved": "https://registry.npmjs.org/@sindresorhus/merge-streams/-/merge-streams-4.0.0.tgz",

package.json

@@ -2,7 +2,7 @@
 	"name": "claude-dev",
 	"displayName": "Cline",
 	"description": "Autonomous coding agent right in your IDE, capable of creating/editing files, running commands, using the browser, and more with your permission every step of the way.",
-	"version": "3.9.2",
+	"version": "3.12.3",
 	"icon": "assets/icons/icon.png",
 	"engines": {
 		"vscode": "^1.84.0"
@@ -118,6 +118,16 @@
 				"category": "Cline"
 			}
 		],
+		"keybindings": [
+			{
+				"command": "cline.addToChat",
+				"key": "cmd+'",
+				"mac": "cmd+'",
+				"win": "ctrl+'",
+				"linux": "ctrl+'",
+				"when": "editorHasSelection"
+			}
+		],
 		"menus": {
 			"view/title": [
 				{
@@ -277,8 +287,12 @@
 		"compile": "npm run check-types && npm run lint && node esbuild.js",
 		"watch": "npm-run-all -p watch:*",
 		"watch:esbuild": "node esbuild.js --watch",
+		"watch:esbuild:test": "IS_TEST=true node esbuild.js --watch",
 		"watch:tsc": "tsc --noEmit --watch --project tsconfig.json",
 		"package": "npm run build:webview && npm run check-types && npm run lint && node esbuild.js --production",
+		"package:test": "IS_TEST=true npm run build:webview:test && npm run check-types && npm run lint && IS_TEST=true node esbuild.js --production",
+		"build:webview:test": "cd webview-ui && IS_TEST=true npm run build",
+		"watch:test": "IS_TEST=true npm-run-all -p watch:tsc watch:esbuild:test",
 		"compile-tests": "tsc -p ./tsconfig.test.json --outDir out",
 		"watch-tests": "tsc -p . -w --outDir out",
 		"pretest": "npm run compile-tests && npm run compile && npm run lint",
@@ -346,6 +360,7 @@
 		"@opentelemetry/sdk-node": "^0.39.1",
 		"@opentelemetry/sdk-trace-node": "^1.30.1",
 		"@opentelemetry/semantic-conventions": "^1.30.0",
+		"@sentry/browser": "^9.12.0",
 		"@vscode/codicons": "^0.0.36",
 		"axios": "^1.8.2",
 		"cheerio": "^1.0.0",

src/api/providers/__tests__/ollama.test.ts

@@ -0,0 +1,228 @@
+import { describe, it, beforeEach, afterEach, before } from "mocha"
+import "should"
+import sinon from "sinon"
+import { Anthropic } from "@anthropic-ai/sdk"
+import { OllamaHandler } from "../ollama"
+import { ApiHandlerOptions } from "../../../shared/api"
+import axios from "axios"
+
+describe("OllamaHandler", () => {
+	let ollamaAvailable = false
+
+	// Check if Ollama is running before running tests
+	before(async function () {
+		this.timeout(5000)
+		try {
+			await axios.get("http://localhost:11434/api/version", { timeout: 2000 })
+			ollamaAvailable = true
+		} catch (error) {
+			console.log("Ollama server not available, skipping tests")
+			ollamaAvailable = false
+		}
+	})
+	let handler: OllamaHandler
+	let options: ApiHandlerOptions
+	let clock: sinon.SinonFakeTimers
+
+	beforeEach(() => {
+		options = {
+			ollamaModelId: "llama2",
+			ollamaBaseUrl: "http://localhost:11434",
+		}
+		handler = new OllamaHandler(options)
+		// Use fake timers for testing timeouts
+		clock = sinon.useFakeTimers()
+	})
+
+	afterEach(() => {
+		clock.restore()
+		sinon.restore()
+	})
+
+	describe("createMessage", () => {
+		it("should handle successful responses", async function () {
+			if (!ollamaAvailable) {
+				this.skip()
+			}
+			this.timeout(5000)
+			// Mock the Ollama client's chat method
+			const chatStub = sinon.stub(handler["client"], "chat").resolves({
+				[Symbol.asyncIterator]: async function* () {
+					yield {
+						message: { content: "Hello, world!" },
+						eval_count: 10,
+						prompt_eval_count: 20,
+					}
+				},
+			} as any)
+
+			const systemPrompt = "You are a helpful assistant."
+			const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Hello" }]
+
+			const result = []
+			const usageInfo = []
+
+			// Collect the results
+			for await (const chunk of handler.createMessage(systemPrompt, messages)) {
+				if (chunk.type === "text") {
+					result.push(chunk.text)
+				} else if (chunk.type === "usage") {
+					usageInfo.push({
+						inputTokens: chunk.inputTokens,
+						outputTokens: chunk.outputTokens,
+					})
+				}
+			}
+
+			// Verify the results
+			result.should.deepEqual(["Hello, world!"])
+			usageInfo.should.deepEqual([{ inputTokens: 20, outputTokens: 10 }])
+			chatStub.calledOnce.should.be.true()
+		})
+
+		it("should handle timeout errors", async function () {
+			if (!ollamaAvailable) {
+				this.skip()
+			}
+			this.timeout(10000)
+			// Restore real timers for this test
+			clock.restore()
+
+			// Create a handler with a very short timeout for testing
+			const testHandler = new OllamaHandler(options)
+
+			// Replace the createMessage method with one that has a shorter timeout
+			testHandler.createMessage = async function* (systemPrompt, messages) {
+				try {
+					// Create a promise that rejects after a short timeout
+					const timeoutPromise = new Promise<never>((_, reject) => {
+						setTimeout(() => reject(new Error("Ollama request timed out after 30 seconds")), 100)
+					})
+
+					// Create a promise that never resolves
+					const neverPromise = new Promise(() => {})
+
+					// Race them
+					await Promise.race([timeoutPromise, neverPromise])
+				} catch (error: any) {
+					// Enhance error reporting
+					console.error(`Ollama API error: ${error.message}`)
+					throw error
+				}
+			}
+
+			const systemPrompt = "You are a helpful assistant."
+			const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Hello" }]
+
+			// Start the request and catch the error
+			let errorMessage = ""
+			try {
+				for await (const _ of testHandler.createMessage(systemPrompt, messages)) {
+					// This should not be reached
+				}
+			} catch (error: any) {
+				errorMessage = error.message
+			}
+
+			// Check the result
+			errorMessage.should.equal("Ollama request timed out after 30 seconds")
+
+			// Restore the fake timers for other tests
+			clock = sinon.useFakeTimers()
+		})
+
+		it("should retry on errors when using the withRetry decorator", async function () {
+			if (!ollamaAvailable) {
+				this.skip()
+			}
+			this.timeout(10000)
+			// Restore real timers for this test
+			clock.restore()
+
+			// Mock the Ollama client's chat method to fail on first call and succeed on second
+			const chatStub = sinon.stub(handler["client"], "chat")
+
+			// First call throws an error
+			chatStub.onFirstCall().rejects(new Error("API Error"))
+
+			// Second call succeeds
+			chatStub.onSecondCall().resolves({
+				[Symbol.asyncIterator]: async function* () {
+					yield {
+						message: { content: "Success after retry" },
+					}
+				},
+			} as any)
+
+			const systemPrompt = "You are a helpful assistant."
+			const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Hello" }]
+
+			const result = []
+
+			// Add a small delay to ensure the retry mechanism has time to work
+			await new Promise((resolve) => setTimeout(resolve, 100))
+
+			// Collect the results
+			for await (const chunk of handler.createMessage(systemPrompt, messages)) {
+				if (chunk.type === "text") {
+					result.push(chunk.text)
+				}
+			}
+
+			// Verify the results
+			result.should.deepEqual(["Success after retry"])
+			chatStub.calledTwice.should.be.true()
+
+			// Restore the fake timers for other tests
+			clock = sinon.useFakeTimers()
+		})
+
+		it("should handle stream processing errors", async function () {
+			if (!ollamaAvailable) {
+				this.skip()
+			}
+			this.timeout(10000)
+			// Restore real timers for this test
+			clock.restore()
+
+			// Create a handler with a custom implementation for testing
+			const testHandler = new OllamaHandler(options)
+
+			// Replace the createMessage method with one that simulates a stream error
+			testHandler.createMessage = async function* (systemPrompt, messages) {
+				// First yield a successful chunk
+				yield {
+					type: "text",
+					text: "Partial response",
+				}
+
+				// Then throw an error in the stream
+				throw new Error("Ollama stream processing error: Stream error")
+			}
+
+			const systemPrompt = "You are a helpful assistant."
+			const messages: Anthropic.Messages.MessageParam[] = [{ role: "user", content: "Hello" }]
+
+			const result = []
+
+			// Collect the results and catch the error
+			let errorMessage = ""
+			try {
+				for await (const chunk of testHandler.createMessage(systemPrompt, messages)) {
+					if (chunk.type === "text") {
+						result.push(chunk.text)
+					}
+				}
+			} catch (error: any) {
+				errorMessage = error.message
+			}
+
+			// Verify the results
+			errorMessage.should.equal("Ollama stream processing error: Stream error")
+			result.should.deepEqual(["Partial response"])
+
+			// Restore the fake timers for other tests
+			clock = sinon.useFakeTimers()
+		})
+	})
+})

src/api/providers/anthropic.ts

@@ -114,7 +114,7 @@ export class AnthropicHandler implements ApiHandler {
 				break
 			}
 			default: {
-				stream = (await this.client.messages.create({
+				stream = await this.client.messages.create({
 					model: modelId,
 					max_tokens: model.info.maxTokens || 8192,
 					temperature: 0,
@@ -123,7 +123,7 @@ export class AnthropicHandler implements ApiHandler {
 					// tools,
 					// tool_choice: { type: "auto" },
 					stream: true,
-				})) as any
+				})
 				break
 			}
 		}

src/api/providers/ollama.ts

@@ -4,6 +4,7 @@ import { ApiHandler } from "../"
 import { ApiHandlerOptions, ModelInfo, openAiModelInfoSaneDefaults } from "../../shared/api"
 import { convertToOllamaMessages } from "../transform/ollama-format"
 import { ApiStream } from "../transform/stream"
+import { withRetry } from "../retry"
 
 export class OllamaHandler implements ApiHandler {
 	private options: ApiHandlerOptions
@@ -14,24 +15,63 @@ export class OllamaHandler implements ApiHandler {
 		this.client = new Ollama({ host: this.options.ollamaBaseUrl || "http://localhost:11434" })
 	}
 
+	@withRetry({ retryAllErrors: true })
 	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
 		const ollamaMessages: Message[] = [{ role: "system", content: systemPrompt }, ...convertToOllamaMessages(messages)]
 
-		const stream = await this.client.chat({
-			model: this.getModel().id,
-			messages: ollamaMessages,
-			stream: true,
-			options: {
-				num_ctx: Number(this.options.ollamaApiOptionsCtxNum) || 32768,
-			},
-		})
-		for await (const chunk of stream) {
-			if (typeof chunk.message.content === "string") {
-				yield {
-					type: "text",
-					text: chunk.message.content,
+		try {
+			// Create a promise that rejects after timeout
+			const timeoutPromise = new Promise<never>((_, reject) => {
+				setTimeout(() => reject(new Error("Ollama request timed out after 30 seconds")), 30000)
+			})
+
+			// Create the actual API request promise
+			const apiPromise = this.client.chat({
+				model: this.getModel().id,
+				messages: ollamaMessages,
+				stream: true,
+				options: {
+					num_ctx: Number(this.options.ollamaApiOptionsCtxNum) || 32768,
+				},
+			})
+
+			// Race the API request against the timeout
+			const stream = (await Promise.race([apiPromise, timeoutPromise])) as Awaited<typeof apiPromise>
+
+			try {
+				for await (const chunk of stream) {
+					if (typeof chunk.message.content === "string") {
+						yield {
+							type: "text",
+							text: chunk.message.content,
+						}
+					}
+
+					// Handle token usage if available
+					if (chunk.eval_count !== undefined || chunk.prompt_eval_count !== undefined) {
+						yield {
+							type: "usage",
+							inputTokens: chunk.prompt_eval_count || 0,
+							outputTokens: chunk.eval_count || 0,
+						}
+					}
 				}
+			} catch (streamError: any) {
+				console.error("Error processing Ollama stream:", streamError)
+				throw new Error(`Ollama stream processing error: ${streamError.message || "Unknown error"}`)
 			}
+		} catch (error: any) {
+			// Check if it's a timeout error
+			if (error.message && error.message.includes("timed out")) {
+				throw new Error("Ollama request timed out after 30 seconds")
+			}
+
+			// Enhance error reporting
+			const statusCode = error.status || error.statusCode
+			const errorMessage = error.message || "Unknown error"
+
+			console.error(`Ollama API error (${statusCode || "unknown"}): ${errorMessage}`)
+			throw error
 		}
 	}
 

src/api/providers/openai-native.ts

@@ -12,7 +12,7 @@ import {
 import { convertToOpenAiMessages } from "../transform/openai-format"
 import { calculateApiCostOpenAI } from "../../utils/cost"
 import { ApiStream } from "../transform/stream"
-import { ChatCompletionReasoningEffort } from "openai/resources/chat/completions.mjs"
+import type { ChatCompletionReasoningEffort } from "openai/resources/chat/completions"
 
 export class OpenAiNativeHandler implements ApiHandler {
 	private options: ApiHandlerOptions

src/api/providers/openai.ts

@@ -6,7 +6,7 @@ import { ApiHandler } from "../index"
 import { convertToOpenAiMessages } from "../transform/openai-format"
 import { ApiStream } from "../transform/stream"
 import { convertToR1Format } from "../transform/r1-format"
-import { ChatCompletionReasoningEffort } from "openai/resources/chat/completions.mjs"
+import type { ChatCompletionReasoningEffort } from "openai/resources/chat/completions"
 
 export class OpenAiHandler implements ApiHandler {
 	private options: ApiHandlerOptions

src/api/providers/requesty.ts

@@ -1,9 +1,10 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import OpenAI from "openai"
-import { ApiHandlerOptions, ModelInfo, openAiModelInfoSaneDefaults } from "../../shared/api"
+import { ApiHandlerOptions, ModelInfo, requestyDefaultModelId, requestyDefaultModelInfo } from "../../shared/api"
 import { ApiHandler } from "../index"
 import { withRetry } from "../retry"
 import { convertToOpenAiMessages } from "../transform/openai-format"
+import { calculateApiCostOpenAI } from "../../utils/cost"
 import { ApiStream } from "../transform/stream"
 
 export class RequestyHandler implements ApiHandler {
@@ -24,21 +25,34 @@ export class RequestyHandler implements ApiHandler {
 
 	@withRetry()
 	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
-		const modelId = this.options.requestyModelId ?? ""
+		const model = this.getModel()
 
 		const openAiMessages: OpenAI.Chat.ChatCompletionMessageParam[] = [
 			{ role: "system", content: systemPrompt },
 			...convertToOpenAiMessages(messages),
 		]
 
+		const reasoningEffort = this.options.o3MiniReasoningEffort || "medium"
+		const reasoning = { reasoning_effort: reasoningEffort }
+		const reasoningArgs = model.id === "openai/o3-mini" ? reasoning : {}
+
+		const thinkingBudget = this.options.thinkingBudgetTokens || 0
+		const thinking =
+			thinkingBudget > 0
+				? { thinking: { type: "enabled", budget_tokens: thinkingBudget } }
+				: { thinking: { type: "disabled" } }
+		const thinkingArgs = model.id.includes("claude-3-7-sonnet") ? thinking : {}
+
 		// @ts-ignore-next-line
 		const stream = await this.client.chat.completions.create({
-			model: modelId,
+			model: model.id,
+			max_tokens: model.info.maxTokens || undefined,
 			messages: openAiMessages,
 			temperature: 0,
 			stream: true,
 			stream_options: { include_usage: true },
-			...(modelId === "openai/o3-mini" ? { reasoning_effort: this.options.o3MiniReasoningEffort || "medium" } : {}),
+			...reasoningArgs,
+			...thinkingArgs,
 		})
 
 		for await (const chunk of stream) {
@@ -73,7 +87,7 @@ export class RequestyHandler implements ApiHandler {
 				const outputTokens = usage.completion_tokens || 0
 				const cacheWriteTokens = usage.prompt_tokens_details?.caching_tokens || undefined
 				const cacheReadTokens = usage.prompt_tokens_details?.cached_tokens || undefined
-				const totalCost = 0 // TODO: Replace with calculateApiCostOpenAI(model.info, inputTokens, outputTokens, cacheWriteTokens, cacheReadTokens)
+				const totalCost = calculateApiCostOpenAI(model.info, inputTokens, outputTokens, cacheWriteTokens, cacheReadTokens)
 
 				yield {
 					type: "usage",
@@ -88,9 +102,11 @@ export class RequestyHandler implements ApiHandler {
 	}
 
 	getModel(): { id: string; info: ModelInfo } {
-		return {
-			id: this.options.requestyModelId ?? "",
-			info: openAiModelInfoSaneDefaults,
+		const modelId = this.options.requestyModelId
+		const modelInfo = this.options.requestyModelInfo
+		if (modelId && modelInfo) {
+			return { id: modelId, info: modelInfo }
 		}
+		return { id: requestyDefaultModelId, info: requestyDefaultModelInfo }
 	}
 }

src/api/providers/types.ts

@@ -1,6 +1,13 @@
 // For the following openrouter error type sources, see the docs here:
 // https://openrouter.ai/docs/api-reference/errors
 
+export interface LanguageModelChatSelector {
+	vendor?: string
+	family?: string
+	version?: string
+	id?: string
+}
+
 export type OpenRouterErrorResponse = {
 	error: {
 		message: string

src/api/providers/vscode-lm.ts

@@ -6,6 +6,7 @@ import { ApiStream } from "../transform/stream"
 import { convertToVsCodeLmMessages } from "../transform/vscode-lm-format"
 import { SELECTOR_SEPARATOR, stringifyVsCodeLmModelSelector } from "../../shared/vsCodeSelectorUtils"
 import { ApiHandlerOptions, ModelInfo, openAiModelInfoSaneDefaults } from "../../shared/api"
+import type { LanguageModelChatSelector as LanguageModelChatSelectorFromTypes } from "./types"
 
 // Cline does not update VSCode type definitions or engine requirements to maintain compatibility.
 // This declaration (as seen in src/integrations/TerminalManager.ts) provides types for the Language Model API in newer versions of VSCode.
@@ -19,12 +20,7 @@ declare module "vscode" {
 		Auto = 1,
 		Required = 2,
 	}
-	interface LanguageModelChatSelector {
-		vendor?: string
-		family?: string
-		version?: string
-		id?: string
-	}
+	interface LanguageModelChatSelector extends LanguageModelChatSelectorFromTypes {}
 	interface LanguageModelChatTool {
 		name: string
 		description: string

src/api/providers/xai.ts

@@ -4,6 +4,7 @@ import { ApiHandler } from "../"
 import { ApiHandlerOptions, XAIModelId, ModelInfo, xaiDefaultModelId, xaiModels } from "../../shared/api"
 import { convertToOpenAiMessages } from "../transform/openai-format"
 import { ApiStream } from "../transform/stream"
+import { ChatCompletionReasoningEffort } from "openai/resources/chat/completions.mjs"
 
 export class XAIHandler implements ApiHandler {
 	private options: ApiHandlerOptions
@@ -18,13 +19,23 @@ export class XAIHandler implements ApiHandler {
 	}
 
 	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
+		const modelId = this.getModel().id
+		// ensure reasoning effort is either "low" or "high" for grok-3-mini
+		let reasoningEffort: ChatCompletionReasoningEffort | undefined
+		if (modelId.includes("3-mini")) {
+			let reasoningEffort = this.options.reasoningEffort
+			if (reasoningEffort && !["low", "high"].includes(reasoningEffort)) {
+				reasoningEffort = undefined
+			}
+		}
 		const stream = await this.client.chat.completions.create({
-			model: this.getModel().id,
+			model: modelId,
 			max_completion_tokens: this.getModel().info.maxTokens,
 			temperature: 0,
 			messages: [{ role: "system", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
 			stream: true,
 			stream_options: { include_usage: true },
+			reasoning_effort: reasoningEffort,
 		})
 
 		for await (const chunk of stream) {
@@ -36,6 +47,14 @@ export class XAIHandler implements ApiHandler {
 				}
 			}
 
+			if ("reasoning_content" in delta && delta.reasoning_content) {
+				yield {
+					type: "reasoning",
+					// @ts-ignore-next-line
+					reasoning: delta.reasoning_content,
+				}
+			}
+
 			if (chunk.usage) {
 				yield {
 					type: "usage",

src/api/transform/gemini-format.ts

@@ -20,7 +20,7 @@ export function convertAnthropicContentToGemini(content: string | Anthropic.Cont
 					},
 				} as InlineDataPart
 			default:
-				throw new Error(`Unsupported content block type: ${(block as any).type}`)
+				throw new Error(`Unsupported content block type: ${block.type}`)
 		}
 	})
 }

src/api/transform/vscode-lm-format.test.ts

@@ -0,0 +1,211 @@
+// This file contains `declare module "vscode"` so we must import it.
+import "../providers/vscode-lm"
+import { describe, it } from "mocha"
+import "should"
+import * as vscode from "vscode"
+import { Anthropic } from "@anthropic-ai/sdk"
+import { asObjectSafe, convertToAnthropicRole, convertToVsCodeLmMessages, convertToAnthropicMessage } from "./vscode-lm-format"
+
+describe("asObjectSafe", () => {
+	it("should handle falsy values", () => {
+		asObjectSafe(0).should.deepEqual({})
+		asObjectSafe("").should.deepEqual({})
+		asObjectSafe(null).should.deepEqual({})
+		asObjectSafe(undefined).should.deepEqual({})
+	})
+
+	it("should parse valid JSON strings", () => {
+		asObjectSafe('{"key": "value"}').should.deepEqual({ key: "value" })
+	})
+
+	it("should return an empty object for invalid JSON strings", () => {
+		asObjectSafe("invalid json").should.deepEqual({})
+	})
+
+	it("should convert objects to plain objects", () => {
+		const input = { prop: "value" }
+		asObjectSafe(input).should.deepEqual(input)
+		asObjectSafe(input).should.not.equal(input) // Should be a new object
+	})
+
+	it("should convert arrays to plain objects", () => {
+		const input = ["hello world"]
+		asObjectSafe(input).should.deepEqual({ 0: "hello world" })
+	})
+})
+
+describe("convertToAnthropicRole", () => {
+	it("should convert VSCode roles to Anthropic roles", () => {
+		// @ts-expect-error（Testing with an invalid role）
+		const unknownRole = "unknown" as vscode.LanguageModelChatMessageRole
+		;(convertToAnthropicRole(vscode.LanguageModelChatMessageRole.Assistant) === "assistant").should.be.true()
+		;(convertToAnthropicRole(vscode.LanguageModelChatMessageRole.User) === "user").should.be.true()
+		;(convertToAnthropicRole(unknownRole) === null).should.be.true()
+	})
+})
+
+describe("convertToVsCodeLmMessages", () => {
+	it("should convert simple string messages", () => {
+		const anthropicMessages: Anthropic.Messages.MessageParam[] = [
+			{ role: "user", content: "Hello" },
+			{ role: "assistant", content: "Hi there" },
+		]
+
+		const result = convertToVsCodeLmMessages(anthropicMessages)
+
+		result.should.have.length(2)
+		result[0].role.should.equal(vscode.LanguageModelChatMessageRole.User)
+		result[0].content[0].should.be.instanceof(vscode.LanguageModelTextPart)
+		const textPart0 = result[0].content[0] as vscode.LanguageModelTextPart
+		textPart0.should.have.property("value", "Hello")
+
+		result[1].role.should.equal(vscode.LanguageModelChatMessageRole.Assistant)
+		result[1].content[0].should.be.instanceof(vscode.LanguageModelTextPart)
+		const textPart1 = result[1].content[0] as vscode.LanguageModelTextPart
+		textPart1.should.have.property("value", "Hi there")
+	})
+
+	it("should convert complex user messages with tool results", () => {
+		const anthropicMessages: Anthropic.Messages.MessageParam[] = [
+			{
+				role: "user",
+				content: [
+					{ type: "text", text: "User text" },
+					{
+						type: "tool_result",
+						tool_use_id: "tool-123",
+						content: [{ type: "text", text: "Tool result" }],
+					},
+				],
+			},
+		]
+
+		const result = convertToVsCodeLmMessages(anthropicMessages)
+
+		result.should.have.length(1)
+		result[0].role.should.equal(vscode.LanguageModelChatMessageRole.User)
+		result[0].content.should.have.length(2)
+
+		// Check that the first content part is a ToolResultPart
+		result[0].content[0].should.be.instanceof(vscode.LanguageModelToolResultPart)
+		const toolResultPart = result[0].content[0] as vscode.LanguageModelToolResultPart
+		toolResultPart.should.have.property("callId", "tool-123")
+
+		// Skip detailed testing of internal structure as it may vary
+		// Just verify it's the right type with the right ID
+
+		// Check the second content part is a TextPart
+		result[0].content[1].should.be.instanceof(vscode.LanguageModelTextPart)
+		const textPart = result[0].content[1] as vscode.LanguageModelTextPart
+		textPart.should.have.property("value", "User text")
+	})
+
+	it("should convert complex assistant messages with tool calls", () => {
+		const anthropicMessages: Anthropic.Messages.MessageParam[] = [
+			{
+				role: "assistant",
+				content: [
+					{ type: "text", text: "Assistant text" },
+					{
+						type: "tool_use",
+						id: "tool-123",
+						name: "testTool",
+						input: { param: "value" },
+					},
+				],
+			},
+		]
+
+		const result = convertToVsCodeLmMessages(anthropicMessages)
+
+		result.should.have.length(1)
+		result[0].role.should.equal(vscode.LanguageModelChatMessageRole.Assistant)
+		result[0].content.should.have.length(2)
+
+		result[0].content[0].should.be.instanceof(vscode.LanguageModelToolCallPart)
+		const toolCallPart = result[0].content[0] as vscode.LanguageModelToolCallPart
+		toolCallPart.should.have.property("callId", "tool-123")
+		toolCallPart.should.have.property("name", "testTool")
+		toolCallPart.should.have.property("input")
+		toolCallPart.input.should.deepEqual({ param: "value" })
+
+		result[0].content[1].should.be.instanceof(vscode.LanguageModelTextPart)
+		const textPart = result[0].content[1] as vscode.LanguageModelTextPart
+		textPart.should.have.property("value", "Assistant text")
+	})
+
+	it("should handle image blocks with appropriate placeholders", () => {
+		const anthropicMessages: Anthropic.Messages.MessageParam[] = [
+			{
+				role: "user",
+				content: [
+					{
+						type: "image",
+						source: {
+							type: "base64",
+							media_type: "image/jpeg",
+							data: "base64data",
+						},
+					},
+				],
+			},
+		]
+
+		const result = convertToVsCodeLmMessages(anthropicMessages)
+
+		result.should.have.length(1)
+		result[0].content[0].should.be.instanceof(vscode.LanguageModelTextPart)
+		const textPart = result[0].content[0] as vscode.LanguageModelTextPart
+		textPart.should.have.property("value")
+		textPart.value.should.match(/Image \(base64\): image\/jpeg not supported by VSCode LM API/)
+	})
+})
+
+describe("convertToAnthropicMessage", () => {
+	it("should convert VSCode assistant messages to Anthropic format", () => {
+		const vsCodeMsg = vscode.LanguageModelChatMessage.Assistant([
+			new vscode.LanguageModelTextPart("Test message"),
+			new vscode.LanguageModelToolCallPart("tool-id", "testTool", { param: "value" }),
+		])
+
+		const result = convertToAnthropicMessage(vsCodeMsg)
+
+		result.should.have.property("role", "assistant")
+		result.should.have.property("content").which.is.an.Array()
+		result.content.should.have.length(2)
+
+		// Check properties carefully to avoid null reference errors
+		if (result.content && result.content.length >= 1) {
+			const textContent = result.content[0]
+			if (textContent) {
+				textContent.should.have.property("type", "text")
+				if (textContent.type === "text") {
+					textContent.should.have.property("text", "Test message")
+				}
+			}
+		}
+
+		if (result.content && result.content.length >= 2) {
+			const toolContent = result.content[1]
+			if (toolContent) {
+				toolContent.should.have.property("type", "tool_use")
+				if (toolContent.type === "tool_use") {
+					toolContent.should.have.property("id", "tool-id")
+					toolContent.should.have.property("name", "testTool")
+					toolContent.should.have.property("input").which.deepEqual({ param: "value" })
+				}
+			}
+		}
+	})
+
+	it("should throw an error for non-assistant messages", () => {
+		const vsCodeMsg = vscode.LanguageModelChatMessage.User("User message")
+
+		try {
+			convertToAnthropicMessage(vsCodeMsg)
+			throw new Error("Should have thrown an error")
+		} catch (error: any) {
+			error.message.should.match(/Only assistant messages are supported/)
+		}
+	})
+})

src/api/transform/vscode-lm-format.ts

@@ -4,7 +4,7 @@ import * as vscode from "vscode"
 /**
  * Safely converts a value into a plain object.
  */
-function asObjectSafe(value: any): object {
+export function asObjectSafe(value: any): object {
 	// Handle null/undefined
 	if (!value) {
 		return {}
@@ -145,7 +145,9 @@ export function convertToVsCodeLmMessages(
 	return vsCodeLmMessages
 }
 
-export function convertToAnthropicRole(vsCodeLmMessageRole: vscode.LanguageModelChatMessageRole): string | null {
+export function convertToAnthropicRole(
+	vsCodeLmMessageRole: vscode.LanguageModelChatMessageRole,
+): Anthropic.Messages.MessageParam["role"] | null {
 	switch (vsCodeLmMessageRole) {
 		case vscode.LanguageModelChatMessageRole.Assistant:
 			return "assistant"
@@ -156,10 +158,8 @@ export function convertToAnthropicRole(vsCodeLmMessageRole: vscode.LanguageModel
 	}
 }
 
-export async function convertToAnthropicMessage(
-	vsCodeLmMessage: vscode.LanguageModelChatMessage,
-): Promise<Anthropic.Messages.Message> {
-	const anthropicRole: string | null = convertToAnthropicRole(vsCodeLmMessage.role)
+export function convertToAnthropicMessage(vsCodeLmMessage: vscode.LanguageModelChatMessage): Anthropic.Messages.Message {
+	const anthropicRole = convertToAnthropicRole(vsCodeLmMessage.role)
 	if (anthropicRole !== "assistant") {
 		throw new Error("Cline <Language Model API>: Only assistant messages are supported.")
 	}

src/core/assistant-message/diff.ts

@@ -284,7 +284,7 @@ export async function constructNewFileContent(diffContent: string, originalConte
 							;[searchMatchIndex, searchEndIndex] = blockMatch
 						} else {
 							throw new Error(
-								`The SEARCH block:\n${currentSearchContent.trimEnd()}\n...does not match anything in the file.`,
+								`The SEARCH block:\n${currentSearchContent.trimEnd()}\n...does not match anything in the file or was searched out of order in the provided blocks.`,
 							)
 						}
 					}

src/core/assistant-message/index.ts

@@ -23,6 +23,7 @@ export const toolUseNames = [
 	"plan_mode_respond",
 	"load_mcp_documentation",
 	"attempt_completion",
+	"new_task",
 ] as const
 
 // Converts array of tool call names into a union type ("execute_command" | "read_file" | ...)
@@ -49,6 +50,7 @@ export const toolParamNames = [
 	"options",
 	"response",
 	"result",
+	"context",
 ] as const
 
 export type ToolParamName = (typeof toolParamNames)[number]

src/core/context-management/context-error-handling.ts

@@ -1,10 +0,0 @@
-export function checkIsOpenRouterContextWindowError(error: any): boolean {
-	return error.code === 400 && error.message?.includes("context length")
-}
-
-export function checkIsAnthropicContextWindowError(response: any): boolean {
-	return (
-		response?.error?.error?.type === "invalid_request_error" &&
-		response?.error?.error?.message?.includes("prompt is too long")
-	)
-}

src/core/context/context-management/ContextManager-legacy.ts

@@ -0,0 +1,102 @@
+import { Anthropic } from "@anthropic-ai/sdk"
+import { ClineApiReqInfo, ClineMessage } from "../../../shared/ExtensionMessage"
+import { ApiHandler } from "../../../api"
+import { getContextWindowInfo } from "./context-window-utils"
+
+class ContextManager {
+	getNewContextMessagesAndMetadata(
+		apiConversationHistory: Anthropic.Messages.MessageParam[],
+		clineMessages: ClineMessage[],
+		api: ApiHandler,
+		conversationHistoryDeletedRange: [number, number] | undefined,
+		previousApiReqIndex: number,
+	) {
+		let updatedConversationHistoryDeletedRange = false
+
+		// If the previous API request's total token usage is close to the context window, truncate the conversation history to free up space for the new request
+		if (previousApiReqIndex >= 0) {
+			const previousRequest = clineMessages[previousApiReqIndex]
+			if (previousRequest && previousRequest.text) {
+				const { tokensIn, tokensOut, cacheWrites, cacheReads }: ClineApiReqInfo = JSON.parse(previousRequest.text)
+				const totalTokens = (tokensIn || 0) + (tokensOut || 0) + (cacheWrites || 0) + (cacheReads || 0)
+				const { maxAllowedSize } = getContextWindowInfo(api)
+
+				// This is the most reliable way to know when we're close to hitting the context window.
+				if (totalTokens >= maxAllowedSize) {
+					// Since the user may switch between models with different context windows, truncating half may not be enough (ie if switching from claude 200k to deepseek 64k, half truncation will only remove 100k tokens, but we need to remove much more)
+					// So if totalTokens/2 is greater than maxAllowedSize, we truncate 3/4 instead of 1/2
+					// FIXME: truncating the conversation in a way that is optimal for prompt caching AND takes into account multi-context window complexity is something we need to improve
+					const keep = totalTokens / 2 > maxAllowedSize ? "quarter" : "half"
+
+					// NOTE: it's okay that we overwriteConversationHistory in resume task since we're only ever removing the last user message and not anything in the middle which would affect this range
+					conversationHistoryDeletedRange = this.getNextTruncationRange(
+						apiConversationHistory,
+						conversationHistoryDeletedRange,
+						keep,
+					)
+
+					updatedConversationHistoryDeletedRange = true
+				}
+			}
+		}
+
+		// conversationHistoryDeletedRange is updated only when we're close to hitting the context window, so we don't continuously break the prompt cache
+		const truncatedConversationHistory = this.getTruncatedMessages(apiConversationHistory, conversationHistoryDeletedRange)
+
+		return {
+			conversationHistoryDeletedRange: conversationHistoryDeletedRange,
+			updatedConversationHistoryDeletedRange: updatedConversationHistoryDeletedRange,
+			truncatedConversationHistory: truncatedConversationHistory,
+		}
+	}
+
+	public getNextTruncationRange(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		currentDeletedRange: [number, number] | undefined,
+		keep: "half" | "quarter",
+	): [number, number] {
+		// Since we always keep the first message, currentDeletedRange[0] will always be 1 (for now until we have a smarter truncation algorithm)
+		const rangeStartIndex = 1
+		const startOfRest = currentDeletedRange ? currentDeletedRange[1] + 1 : 1
+
+		let messagesToRemove: number
+		if (keep === "half") {
+			// Remove half of remaining user-assistant pairs
+			// We first calculate half of the messages then divide by 2 to get the number of pairs.
+			// After flooring, we multiply by 2 to get the number of messages.
+			// Note that this will also always be an even number.
+			messagesToRemove = Math.floor((apiMessages.length - startOfRest) / 4) * 2 // Keep even number
+		} else {
+			// Remove 3/4 of remaining user-assistant pairs
+			// We calculate 3/4ths of the messages then divide by 2 to get the number of pairs.
+			// After flooring, we multiply by 2 to get the number of messages.
+			// Note that this will also always be an even number.
+			messagesToRemove = Math.floor(((apiMessages.length - startOfRest) * 3) / 4 / 2) * 2
+		}
+
+		let rangeEndIndex = startOfRest + messagesToRemove - 1
+
+		// Make sure the last message being removed is a user message, so that the next message after the initial task message is an assistant message. This preservers the user-assistant-user-assistant structure.
+		// NOTE: anthropic format messages are always user-assistant-user-assistant, while openai format messages can have multiple user messages in a row (we use anthropic format throughout cline)
+		if (apiMessages[rangeEndIndex].role !== "user") {
+			rangeEndIndex -= 1
+		}
+
+		// this is an inclusive range that will be removed from the conversation history
+		return [rangeStartIndex, rangeEndIndex]
+	}
+
+	public getTruncatedMessages(
+		messages: Anthropic.Messages.MessageParam[],
+		deletedRange: [number, number] | undefined,
+	): Anthropic.Messages.MessageParam[] {
+		if (!deletedRange) {
+			return messages
+		}
+
+		const [start, end] = deletedRange
+		// the range is inclusive - both start and end indices and everything in between will be removed from the final result.
+		// NOTE: if you try to console log these, don't forget that logging a reference to an array may not provide the same result as logging a slice() snapshot of that array at that exact moment. The following DOES in fact include the latest assistant message.
+		return [...messages.slice(0, start), ...messages.slice(end + 1)]
+	}
+}

src/core/context/context-management/ContextManager.ts

@@ -1,13 +1,13 @@
-import { Anthropic } from "@anthropic-ai/sdk"
-import { ClineApiReqInfo, ClineMessage } from "../../shared/ExtensionMessage"
-import { ApiHandler } from "../../api"
-import { OpenAiHandler } from "../../api/providers/openai"
-import { formatResponse } from "../prompts/responses"
-import { GlobalFileNames } from "../storage/disk"
-import { fileExistsAtPath } from "../../utils/fs"
+import { getContextWindowInfo } from "./context-window-utils"
+import { formatResponse } from "../../prompts/responses"
+import { GlobalFileNames } from "../../storage/disk"
+import { fileExistsAtPath } from "../../../utils/fs"
 import * as path from "path"
 import fs from "fs/promises"
 import cloneDeep from "clone-deep"
+import { ClineApiReqInfo, ClineMessage } from "../../../shared/ExtensionMessage"
+import { ApiHandler } from "../../../api"
+import { Anthropic } from "@anthropic-ai/sdk"
 
 enum EditType {
 	UNDEFINED = 0,
@@ -125,25 +125,7 @@ export class ContextManager {
 				const timestamp = previousRequest.ts
 				const { tokensIn, tokensOut, cacheWrites, cacheReads }: ClineApiReqInfo = JSON.parse(previousRequest.text)
 				const totalTokens = (tokensIn || 0) + (tokensOut || 0) + (cacheWrites || 0) + (cacheReads || 0)
-				let contextWindow = api.getModel().info.contextWindow || 128_000
-				// FIXME: hack to get anyone using openai compatible with deepseek to have the proper context window instead of the default 128k. We need a way for the user to specify the context window for models they input through openai compatible
-				if (api instanceof OpenAiHandler && api.getModel().id.toLowerCase().includes("deepseek")) {
-					contextWindow = 64_000
-				}
-				let maxAllowedSize: number
-				switch (contextWindow) {
-					case 64_000: // deepseek models
-						maxAllowedSize = contextWindow - 27_000
-						break
-					case 128_000: // most models
-						maxAllowedSize = contextWindow - 30_000
-						break
-					case 200_000: // claude models
-						maxAllowedSize = contextWindow - 40_000
-						break
-					default:
-						maxAllowedSize = Math.max(contextWindow - 40_000, contextWindow * 0.8) // for deepseek, 80% of 64k meant only ~10k buffer which was too small and resulted in users getting context window errors.
-				}
+				const { maxAllowedSize } = getContextWindowInfo(api)
 
 				// This is the most reliable way to know when we're close to hitting the context window.
 				if (totalTokens >= maxAllowedSize) {

src/core/context/context-management/context-error-handling.ts

@@ -0,0 +1,18 @@
+export function checkIsOpenRouterContextWindowError(error: any): boolean {
+	try {
+		return error.code === 400 && error.message?.includes("context length")
+	} catch (e: unknown) {
+		return false
+	}
+}
+
+export function checkIsAnthropicContextWindowError(response: any): boolean {
+	try {
+		return (
+			response?.error?.error?.type === "invalid_request_error" &&
+			response?.error?.error?.message?.includes("prompt is too long")
+		)
+	} catch (e: unknown) {
+		return false
+	}
+}

src/core/context/context-management/context-window-utils.ts

@@ -0,0 +1,35 @@
+import { ApiHandler } from "../../../api"
+import { OpenAiHandler } from "../../../api/providers/openai"
+
+/**
+ * Gets context window information for the given API handler
+ *
+ * @param api The API handler to get context window information for
+ * @returns An object containing the raw context window size and the effective max allowed size
+ */
+export function getContextWindowInfo(api: ApiHandler) {
+	let contextWindow = api.getModel().info.contextWindow || 128_000
+	// FIXME: hack to get anyone using openai compatible with deepseek to have the proper context window instead of the default 128k. We need a way for the user to specify the context window for models they input through openai compatible
+
+	// Handle special cases like DeepSeek
+	if (api instanceof OpenAiHandler && api.getModel().id.toLowerCase().includes("deepseek")) {
+		contextWindow = 64_000
+	}
+
+	let maxAllowedSize: number
+	switch (contextWindow) {
+		case 64_000: // deepseek models
+			maxAllowedSize = contextWindow - 27_000
+			break
+		case 128_000: // most models
+			maxAllowedSize = contextWindow - 30_000
+			break
+		case 200_000: // claude models
+			maxAllowedSize = contextWindow - 40_000
+			break
+		default:
+			maxAllowedSize = Math.max(contextWindow - 40_000, contextWindow * 0.8) // for deepseek, 80% of 64k meant only ~10k buffer which was too small and resulted in users getting context window errors.
+	}
+
+	return { contextWindow, maxAllowedSize }
+}

src/core/context/context-tracking/ContextTrackerTypes.ts

@@ -10,11 +10,14 @@ export interface FileMetadataEntry {
 	user_edit_date?: number | null
 }
 
-export interface TaskMetadata {
-	files_in_context: FileMetadataEntry[]
+export interface ModelMetadataEntry {
+	ts: number
+	model_id: string
+	model_provider_id: string
+	mode: string
 }
 
-// Interface for the controller to avoid direct dependency
-export interface ControllerLike {
-	context: vscode.ExtensionContext
+export interface TaskMetadata {
+	files_in_context: FileMetadataEntry[]
+	model_usage: ModelMetadataEntry[]
 }

src/core/context/context-tracking/FileContextTracker.test.ts

@@ -4,12 +4,11 @@ import * as sinon from "sinon"
 import * as vscode from "vscode"
 import * as path from "path"
 import { FileContextTracker } from "./FileContextTracker"
-import * as diskModule from "../storage/disk"
-import type { TaskMetadata, ControllerLike, FileMetadataEntry } from "./FileContextTrackerTypes"
+import * as diskModule from "../../storage/disk"
+import type { TaskMetadata, FileMetadataEntry } from "./ContextTrackerTypes"
 
 describe("FileContextTracker", () => {
 	let sandbox: sinon.SinonSandbox
-	let mockController: ControllerLike
 	let mockContext: vscode.ExtensionContext
 	let mockWorkspace: sinon.SinonStub
 	let mockFileSystemWatcher: any
@@ -41,25 +40,21 @@ describe("FileContextTracker", () => {
 		const originalCreateFileSystemWatcher = vscode.workspace.createFileSystemWatcher
 		vscode.workspace.createFileSystemWatcher = function () {
 			return mockFileSystemWatcher
-		} as any
+		}
 
 		// Mock controller and context
 		mockContext = {
 			globalStorageUri: { fsPath: "/mock/storage" },
 		} as unknown as vscode.ExtensionContext
 
-		mockController = {
-			context: mockContext,
-		}
-
 		// Mock disk module functions
-		mockTaskMetadata = { files_in_context: [] }
+		mockTaskMetadata = { files_in_context: [], model_usage: [] }
 		getTaskMetadataStub = sandbox.stub(diskModule, "getTaskMetadata").resolves(mockTaskMetadata)
 		saveTaskMetadataStub = sandbox.stub(diskModule, "saveTaskMetadata").resolves()
 
 		// Create tracker instance
 		taskId = "test-task-id"
-		tracker = new FileContextTracker(mockController, taskId)
+		tracker = new FileContextTracker(mockContext, taskId)
 	})
 
 	afterEach(() => {

src/core/context/context-tracking/FileContextTracker.ts

@@ -1,7 +1,7 @@
 import * as path from "path"
 import * as vscode from "vscode"
-import { getTaskMetadata, saveTaskMetadata } from "../storage/disk"
-import type { FileMetadataEntry, ControllerLike } from "./FileContextTrackerTypes"
+import { getTaskMetadata, saveTaskMetadata } from "../../storage/disk"
+import type { FileMetadataEntry } from "./ContextTrackerTypes"
 
 // This class is responsible for tracking file operations that may result in stale context.
 // If a user modifies a file outside of Cline, the context may become stale and need to be updated.
@@ -16,29 +16,19 @@ import type { FileMetadataEntry, ControllerLike } from "./FileContextTrackerType
 // If the full contents of a file are pass to Cline via a tool, mention, or edit, the file is marked as active.
 // If a file is modified outside of Cline, we detect and track this change to prevent stale context.
 export class FileContextTracker {
+	private context: vscode.ExtensionContext
 	readonly taskId: string
-	private controllerRef: WeakRef<ControllerLike>
 
 	// File tracking and watching
 	private fileWatchers = new Map<string, vscode.FileSystemWatcher>()
 	private recentlyModifiedFiles = new Set<string>()
 	private recentlyEditedByCline = new Set<string>()
 
-	constructor(controller: ControllerLike, taskId: string) {
-		this.controllerRef = new WeakRef(controller)
+	constructor(context: vscode.ExtensionContext, taskId: string) {
+		this.context = context
 		this.taskId = taskId
 	}
 
-	// While a task is ref'd by a controller, it will always have access to the extension context
-	// This error is thrown if the controller derefs the task after e.g., aborting the task
-	private context(): vscode.ExtensionContext {
-		const context = this.controllerRef.deref()?.context
-		if (!context) {
-			throw new Error("Unable to access extension context")
-		}
-		return context
-	}
-
 	// Gets the current working directory or returns undefined if it cannot be determined
 	private getCwd(): string | undefined {
 		const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0)
@@ -89,9 +79,8 @@ export class FileContextTracker {
 				return
 			}
 
-			const context = this.context()
 			// Add file to metadata
-			await this.addFileToFileContextTracker(context, this.taskId, filePath, operation)
+			await this.addFileToFileContextTracker(this.context, this.taskId, filePath, operation)
 
 			// Set up file watcher for this file
 			await this.setupFileWatcher(filePath)

src/core/context/context-tracking/ModelContextTracker.test.ts

@@ -0,0 +1,205 @@
+import { describe, it, beforeEach, afterEach } from "mocha"
+import { expect } from "chai"
+import * as sinon from "sinon"
+import * as vscode from "vscode"
+import { ModelContextTracker } from "./ModelContextTracker"
+import * as diskModule from "../../storage/disk"
+import type { TaskMetadata } from "./ContextTrackerTypes"
+
+describe("ModelContextTracker", () => {
+	let sandbox: sinon.SinonSandbox
+	let mockContext: vscode.ExtensionContext
+	let tracker: ModelContextTracker
+	let taskId: string
+	let mockTaskMetadata: TaskMetadata
+	let getTaskMetadataStub: sinon.SinonStub
+	let saveTaskMetadataStub: sinon.SinonStub
+
+	beforeEach(() => {
+		sandbox = sinon.createSandbox()
+
+		// Mock controller and context
+		mockContext = {
+			globalStorageUri: { fsPath: "/mock/storage" },
+		} as unknown as vscode.ExtensionContext
+
+		// Mock disk module functions
+		mockTaskMetadata = { files_in_context: [], model_usage: [] }
+		getTaskMetadataStub = sandbox.stub(diskModule, "getTaskMetadata").resolves(mockTaskMetadata)
+		saveTaskMetadataStub = sandbox.stub(diskModule, "saveTaskMetadata").resolves()
+
+		// Create tracker instance
+		taskId = "test-task-id"
+		tracker = new ModelContextTracker(mockContext, taskId)
+	})
+
+	afterEach(() => {
+		sandbox.restore()
+	})
+
+	it("should record model usage with correct data", async () => {
+		// Test data
+		const apiProviderId = "anthropic"
+		const modelId = "claude-3-opus"
+		const mode = "act"
+
+		// Use a fake timer to have a predictable timestamp
+		const fakeNow = 1617293940000 // Some fixed timestamp
+		const clock = sandbox.useFakeTimers(fakeNow)
+
+		try {
+			// Call the method being tested
+			await tracker.recordModelUsage(apiProviderId, modelId, mode)
+
+			// Verify getTaskMetadata was called with correct parameters
+			expect(getTaskMetadataStub.calledOnce).to.be.true
+			expect(getTaskMetadataStub.firstCall.args[1]).to.equal(taskId)
+
+			// Verify saveTaskMetadata was called with the correct data
+			expect(saveTaskMetadataStub.calledOnce).to.be.true
+
+			// Extract the saved metadata from the call arguments
+			const savedMetadata = saveTaskMetadataStub.firstCall.args[2]
+
+			// Verify model_usage array has one entry
+			expect(savedMetadata.model_usage.length).to.equal(1)
+
+			// Verify the entry has the correct properties
+			const modelUsageEntry = savedMetadata.model_usage[0]
+			expect(modelUsageEntry.ts).to.equal(fakeNow)
+			expect(modelUsageEntry.model_id).to.equal(modelId)
+			expect(modelUsageEntry.model_provider_id).to.equal(apiProviderId)
+			expect(modelUsageEntry.mode).to.equal(mode)
+		} finally {
+			// Restore the clock
+			clock.restore()
+		}
+	})
+
+	it("should throw an error when controller is dereferenced", async () => {
+		// Create a new tracker with a controller that will be garbage collected
+		const weakTracker = new ModelContextTracker(mockContext, taskId)
+
+		// Force the WeakRef to return null by overriding the deref method
+		const weakRef = { deref: sandbox.stub().returns(null) }
+		sandbox.stub(WeakRef.prototype, "deref").callsFake(() => weakRef.deref())
+
+		try {
+			// Try to call the method - this should throw
+			await weakTracker.recordModelUsage("any-provider", "any-model", "any-mode")
+
+			// If we get here, the test should fail
+			expect.fail("Expected an error to be thrown")
+		} catch (error) {
+			// Verify the error message
+			expect(error.message).to.equal("Unable to access extension context")
+		}
+	})
+
+	it("should append model usage to existing entries", async () => {
+		// Add an existing model usage entry
+		const existingTimestamp = 1617200000000
+		mockTaskMetadata.model_usage = [
+			{
+				ts: existingTimestamp,
+				model_id: "existing-model",
+				model_provider_id: "existing-provider",
+				mode: "plan",
+			},
+		]
+
+		// Test data for new entry
+		const apiProviderId = "anthropic"
+		const modelId = "claude-3-sonnet"
+		const mode = "act"
+
+		// Use a fake timer
+		const newTimestamp = 1617300000000
+		const clock = sandbox.useFakeTimers(newTimestamp)
+
+		try {
+			// Call the method being tested
+			await tracker.recordModelUsage(apiProviderId, modelId, mode)
+
+			// Verify saveTaskMetadata was called
+			expect(saveTaskMetadataStub.calledOnce).to.be.true
+
+			// Extract the saved metadata
+			const savedMetadata = saveTaskMetadataStub.firstCall.args[2]
+
+			// Verify model_usage array now has two entries
+			expect(savedMetadata.model_usage.length).to.equal(2)
+
+			// Verify the existing entry is preserved
+			expect(savedMetadata.model_usage[0]).to.deep.equal({
+				ts: existingTimestamp,
+				model_id: "existing-model",
+				model_provider_id: "existing-provider",
+				mode: "plan",
+			})
+
+			// Verify the new entry has correct data
+			expect(savedMetadata.model_usage[1]).to.deep.equal({
+				ts: newTimestamp,
+				model_id: modelId,
+				model_provider_id: apiProviderId,
+				mode: mode,
+			})
+		} finally {
+			clock.restore()
+		}
+	})
+
+	it("should handle multiple model usages in sequence", async () => {
+		// Test data for sequential calls
+		const usages = [
+			{ provider: "anthropic", model: "claude-3-opus", mode: "plan" },
+			{ provider: "openai", model: "gpt-4", mode: "act" },
+			{ provider: "anthropic", model: "claude-3-haiku", mode: "plan" },
+		]
+
+		// Use a fake timer that advances with each call
+		const startTime = 1617300000000
+		const clock = sandbox.useFakeTimers(startTime)
+
+		try {
+			// Record multiple model usages
+			for (let i = 0; i < usages.length; i++) {
+				const { provider, model, mode } = usages[i]
+
+				// Advance time by 1 second for each call
+				clock.tick(1000)
+				const expectedTime = startTime + (i + 1) * 1000
+
+				// Reset history between calls to check individual call behavior
+				getTaskMetadataStub.resetHistory()
+				saveTaskMetadataStub.resetHistory()
+
+				// Reset mock metadata for each iteration to avoid accumulation
+				mockTaskMetadata.model_usage = []
+
+				// Call the method
+				await tracker.recordModelUsage(provider, model, mode)
+
+				// Verify interaction with disk module
+				expect(getTaskMetadataStub.calledOnce).to.be.true
+				expect(saveTaskMetadataStub.calledOnce).to.be.true
+
+				// Get the saved metadata
+				const savedMetadata = saveTaskMetadataStub.firstCall.args[2]
+
+				// Since we reset the array for each call, we should always have 1 entry
+				expect(savedMetadata.model_usage.length).to.equal(1)
+
+				// Check the entry
+				const entry = savedMetadata.model_usage[0]
+				expect(entry.ts).to.equal(expectedTime)
+				expect(entry.model_id).to.equal(model)
+				expect(entry.model_provider_id).to.equal(provider)
+				expect(entry.mode).to.equal(mode)
+			}
+		} finally {
+			clock.restore()
+		}
+	})
+})

src/core/context/context-tracking/ModelContextTracker.ts

@@ -0,0 +1,40 @@
+import * as vscode from "vscode"
+import { getTaskMetadata, saveTaskMetadata } from "../../storage/disk"
+
+export class ModelContextTracker {
+	readonly taskId: string
+	private context: vscode.ExtensionContext
+
+	constructor(context: vscode.ExtensionContext, taskId: string) {
+		this.context = context
+		this.taskId = taskId
+	}
+
+	async recordModelUsage(apiProviderId: string, modelId: string, mode: string) {
+		const metadata = await getTaskMetadata(this.context, this.taskId)
+
+		if (!metadata.model_usage) {
+			metadata.model_usage = []
+		}
+
+		// check to see if the last entry is the same as the new one
+		const lastEntry = metadata.model_usage[metadata.model_usage.length - 1]
+		if (
+			lastEntry &&
+			lastEntry.model_id === modelId &&
+			lastEntry.model_provider_id === apiProviderId &&
+			lastEntry.mode === mode
+		) {
+			return
+		}
+
+		metadata.model_usage.push({
+			ts: Date.now(),
+			model_id: modelId,
+			model_provider_id: apiProviderId,
+			mode: mode,
+		})
+
+		await saveTaskMetadata(this.context, this.taskId, metadata)
+	}
+}

src/core/context/instructions/user-instructions/cline-rules.ts

@@ -0,0 +1,65 @@
+import path from "path"
+import { GlobalFileNames } from "../../../storage/disk"
+import { fileExistsAtPath, isDirectory, readDirectory } from "../../../../utils/fs"
+import { formatResponse } from "../../../prompts/responses"
+import fs from "fs/promises"
+
+export const getGlobalClineRules = async (globalClineRulesFilePath: string) => {
+	if (await fileExistsAtPath(globalClineRulesFilePath)) {
+		if (await isDirectory(globalClineRulesFilePath)) {
+			try {
+				const rulesFilePaths = await readDirectory(globalClineRulesFilePath)
+				const rulesFilesTotalContent = await getClineRulesFilesTotalContent(rulesFilePaths, globalClineRulesFilePath)
+				const clineRulesFileInstructions = formatResponse.clineRulesGlobalDirectoryInstructions(rulesFilesTotalContent)
+				return clineRulesFileInstructions
+			} catch {
+				console.error(`Failed to read .clinerules directory at ${globalClineRulesFilePath}`)
+			}
+		} else {
+			console.error(`${globalClineRulesFilePath} is not a directory`)
+			return undefined
+		}
+	}
+
+	return undefined
+}
+
+export const getLocalClineRules = async (cwd: string) => {
+	const clineRulesFilePath = path.resolve(cwd, GlobalFileNames.clineRules)
+
+	let clineRulesFileInstructions: string | undefined
+
+	if (await fileExistsAtPath(clineRulesFilePath)) {
+		if (await isDirectory(clineRulesFilePath)) {
+			try {
+				const rulesFilePaths = await readDirectory(path.join(cwd, GlobalFileNames.clineRules))
+				const rulesFilesTotalContent = await getClineRulesFilesTotalContent(rulesFilePaths, cwd)
+				clineRulesFileInstructions = formatResponse.clineRulesLocalDirectoryInstructions(cwd, rulesFilesTotalContent)
+			} catch {
+				console.error(`Failed to read .clinerules directory at ${clineRulesFilePath}`)
+			}
+		} else {
+			try {
+				const ruleFileContent = (await fs.readFile(clineRulesFilePath, "utf8")).trim()
+				if (ruleFileContent) {
+					clineRulesFileInstructions = formatResponse.clineRulesLocalFileInstructions(cwd, ruleFileContent)
+				}
+			} catch {
+				console.error(`Failed to read .clinerules file at ${clineRulesFilePath}`)
+			}
+		}
+	}
+
+	return clineRulesFileInstructions
+}
+
+const getClineRulesFilesTotalContent = async (rulesFilePaths: string[], basePath: string) => {
+	const ruleFilesTotalContent = await Promise.all(
+		rulesFilePaths.map(async (filePath) => {
+			const ruleFilePath = path.resolve(basePath, filePath)
+			const ruleFilePathRelative = path.relative(basePath, ruleFilePath)
+			return `${ruleFilePathRelative}\n` + (await fs.readFile(ruleFilePath, "utf8")).trim()
+		}),
+	).then((contents) => contents.join("\n\n"))
+	return ruleFilesTotalContent
+}

src/core/controller/index.ts

@@ -1,10 +1,9 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import axios from "axios"
+import type { AxiosRequestConfig } from "axios"
 import crypto from "crypto"
-import { execa } from "execa"
 import fs from "fs/promises"
 import { setTimeout as setTimeoutPromise } from "node:timers/promises"
-import os from "os"
 import pWaitFor from "p-wait-for"
 import * as path from "path"
 import * as vscode from "vscode"
@@ -17,10 +16,12 @@ import { selectImages } from "../../integrations/misc/process-images"
 import { getTheme } from "../../integrations/theme/getTheme"
 import WorkspaceTracker from "../../integrations/workspace/WorkspaceTracker"
 import { ClineAccountService } from "../../services/account/ClineAccountService"
+import { discoverChromeInstances } from "../../services/browser/BrowserDiscovery"
+import { BrowserSession } from "../../services/browser/BrowserSession"
 import { McpHub } from "../../services/mcp/McpHub"
+import { searchWorkspaceFiles } from "../../services/search/file-search"
 import { telemetryService } from "../../services/telemetry/TelemetryService"
 import { ApiProvider, ModelInfo } from "../../shared/api"
-import { findLast } from "../../shared/array"
 import { ChatContent } from "../../shared/ChatContent"
 import { ChatSettings } from "../../shared/ChatSettings"
 import { ExtensionMessage, ExtensionState, Invoke, Platform } from "../../shared/ExtensionMessage"
@@ -30,9 +31,10 @@ import { TelemetrySetting } from "../../shared/TelemetrySetting"
 import { ClineCheckpointRestore, WebviewMessage } from "../../shared/WebviewMessage"
 import { fileExistsAtPath } from "../../utils/fs"
 import { searchCommits } from "../../utils/git"
+import { getWorkspacePath } from "../../utils/path"
 import { getTotalTasksSize } from "../../utils/storage"
-import { Task } from "../task"
 import { openMention } from "../mentions"
+import { ensureMcpServersDirectoryExists, ensureSettingsDirectoryExists, GlobalFileNames } from "../storage/disk"
 import {
 	getAllExtensionState,
 	getGlobalState,
@@ -42,12 +44,7 @@ import {
 	updateApiConfiguration,
 	updateGlobalState,
 } from "../storage/state"
-import { WebviewProvider } from "../webview"
-import { BrowserSession } from "../../services/browser/BrowserSession"
-import { GlobalFileNames } from "../storage/disk"
-import { discoverChromeInstances } from "../../services/browser/BrowserDiscovery"
-import { searchWorkspaceFiles } from "../../services/search/file-search"
-import { getWorkspacePath } from "../../utils/path"
+import { Task } from "../task"
 
 /*
 https://github.com/microsoft/vscode-webview-ui-toolkit-samples/blob/main/default/weather-webview/src/providers/WeatherViewProvider.ts
@@ -56,25 +53,37 @@ https://github.com/KumarVariable/vscode-extension-sidebar-html/blob/master/src/c
 */
 
 export class Controller {
+	private postMessage: (message: ExtensionMessage) => Thenable<boolean> | undefined
+
 	private disposables: vscode.Disposable[] = []
 	private task?: Task
-	workspaceTracker?: WorkspaceTracker
-	mcpHub?: McpHub
-	accountService?: ClineAccountService
-	private latestAnnouncementId = "march-22-2025" // update to some unique identifier when we add a new announcement
-	private webviewProviderRef: WeakRef<WebviewProvider>
+	workspaceTracker: WorkspaceTracker
+	mcpHub: McpHub
+	accountService: ClineAccountService
+	private latestAnnouncementId = "april-11-2025" // update to some unique identifier when we add a new announcement
 
 	constructor(
 		readonly context: vscode.ExtensionContext,
 		private readonly outputChannel: vscode.OutputChannel,
-		webviewProvider: WebviewProvider,
+		postMessage: (message: ExtensionMessage) => Thenable<boolean> | undefined,
 	) {
 		this.outputChannel.appendLine("ClineProvider instantiated")
-		this.webviewProviderRef = new WeakRef(webviewProvider)
+		this.postMessage = postMessage
 
-		this.workspaceTracker = new WorkspaceTracker(this)
-		this.mcpHub = new McpHub(this)
-		this.accountService = new ClineAccountService(this)
+		this.workspaceTracker = new WorkspaceTracker((msg) => this.postMessageToWebview(msg))
+		this.mcpHub = new McpHub(
+			() => ensureMcpServersDirectoryExists(),
+			() => ensureSettingsDirectoryExists(this.context),
+			(msg) => this.postMessageToWebview(msg),
+			this.context.extension?.packageJSON?.version ?? "1.0.0",
+		)
+		this.accountService = new ClineAccountService(
+			(msg) => this.postMessageToWebview(msg),
+			async () => {
+				const { apiConfiguration } = await this.getStateToPostToWebview()
+				return apiConfiguration?.clineApiKey
+			},
+		)
 
 		// Clean up legacy checkpoints
 		cleanupLegacyCheckpoints(this.context.globalStorageUri.fsPath, this.outputChannel).catch((error) => {
@@ -97,11 +106,8 @@ export class Controller {
 				x.dispose()
 			}
 		}
-		this.workspaceTracker?.dispose()
-		this.workspaceTracker = undefined
-		this.mcpHub?.dispose()
-		this.mcpHub = undefined
-		this.accountService = undefined
+		this.workspaceTracker.dispose()
+		this.mcpHub.dispose()
 		this.outputChannel.appendLine("Disposed all disposables")
 
 		console.error("Controller disposed")
@@ -124,12 +130,19 @@ export class Controller {
 		await updateGlobalState(this.context, "userInfo", info)
 	}
 
-	async initClineWithTask(task?: string, images?: string[]) {
+	async initTask(task?: string, images?: string[], historyItem?: HistoryItem) {
 		await this.clearTask() // ensures that an existing task doesn't exist before starting a new one, although this shouldn't be possible since user must clear task before starting a new one
 		const { apiConfiguration, customInstructions, autoApprovalSettings, browserSettings, chatSettings } =
 			await getAllExtensionState(this.context)
 		this.task = new Task(
-			this,
+			this.context,
+			this.mcpHub,
+			this.workspaceTracker,
+			(historyItem) => this.updateTaskHistory(historyItem),
+			() => this.postStateToWebview(),
+			(message) => this.postMessageToWebview(message),
+			(taskId) => this.reinitExistingTaskFromId(taskId),
+			() => this.cancelTask(),
 			apiConfiguration,
 			autoApprovalSettings,
 			browserSettings,
@@ -137,29 +150,20 @@ export class Controller {
 			customInstructions,
 			task,
 			images,
-		)
-	}
-
-	async initClineWithHistoryItem(historyItem: HistoryItem) {
-		await this.clearTask()
-		const { apiConfiguration, customInstructions, autoApprovalSettings, browserSettings, chatSettings } =
-			await getAllExtensionState(this.context)
-		this.task = new Task(
-			this,
-			apiConfiguration,
-			autoApprovalSettings,
-			browserSettings,
-			chatSettings,
-			customInstructions,
-			undefined,
-			undefined,
 			historyItem,
 		)
 	}
 
+	async reinitExistingTaskFromId(taskId: string) {
+		const history = await this.getTaskWithId(taskId)
+		if (history) {
+			await this.initTask(undefined, undefined, history.historyItem)
+		}
+	}
+
 	// Send any JSON serializable data to the react app
 	async postMessageToWebview(message: ExtensionMessage) {
-		await this.webviewProviderRef.deref()?.view?.webview.postMessage(message)
+		await this.postMessage(message)
 	}
 
 	/**
@@ -259,7 +263,7 @@ export class Controller {
 				// Could also do this in extension .ts
 				//this.postMessageToWebview({ type: "text", text: `Extension: ${Date.now()}` })
 				// initializing new instance of Cline will make sure that any agentically running promises in old instance don't affect our new task. this essentially creates a fresh slate for the new task
-				await this.initClineWithTask(message.text, message.images)
+				await this.initTask(message.text, message.images)
 				break
 			case "apiConfiguration":
 				if (message.apiConfiguration) {
@@ -487,6 +491,9 @@ export class Controller {
 			case "refreshOpenRouterModels":
 				await this.refreshOpenRouterModels()
 				break
+			case "refreshRequestyModels":
+				await this.refreshRequestyModels()
+				break
 			case "refreshOpenAiModels":
 				const { apiConfiguration } = await getAllExtensionState(this.context)
 				const openAiModels = await this.getOpenAiModels(apiConfiguration.openAiBaseUrl, apiConfiguration.openAiApiKey)
@@ -575,7 +582,7 @@ export class Controller {
 				break
 			}
 			case "showMcpView": {
-				await this.postMessageToWebview({ type: "action", action: "mcpButtonClicked" })
+				await this.postMessageToWebview({ type: "action", action: "mcpButtonClicked", tab: message.tab || undefined })
 				break
 			}
 			case "openMcpSettings": {
@@ -882,6 +889,27 @@ export class Controller {
 				}
 				break
 			}
+			case "toggleFavoriteModel": {
+				if (message.modelId) {
+					const { apiConfiguration } = await getAllExtensionState(this.context)
+					const favoritedModelIds = apiConfiguration.favoritedModelIds || []
+
+					// Toggle favorite status
+					const updatedFavorites = favoritedModelIds.includes(message.modelId)
+						? favoritedModelIds.filter((id) => id !== message.modelId)
+						: [...favoritedModelIds, message.modelId]
+
+					await updateGlobalState(this.context, "favoritedModelIds", updatedFavorites)
+
+					// Capture telemetry for model favorite toggle
+					const isFavorited = !favoritedModelIds.includes(message.modelId)
+					telemetryService.captureModelFavoritesUsage(message.modelId, isFavorited)
+
+					// Post state to webview without changing any other configuration
+					await this.postStateToWebview()
+				}
+				break
+			}
 			// Add more switch case statements here as more webview message commands
 			// are created within the webview context (i.e. inside media/main.js)
 		}
@@ -907,6 +935,7 @@ export class Controller {
 			previousModeModelInfo: newModelInfo,
 			previousModeVsCodeLmModelSelector: newVsCodeLmModelSelector,
 			previousModeThinkingBudgetTokens: newThinkingBudgetTokens,
+			previousModeReasoningEffort: newReasoningEffort,
 			planActSeparateModelsSetting,
 		} = await getAllExtensionState(this.context)
 
@@ -916,6 +945,7 @@ export class Controller {
 			// Save the last model used in this mode
 			await updateGlobalState(this.context, "previousModeApiProvider", apiConfiguration.apiProvider)
 			await updateGlobalState(this.context, "previousModeThinkingBudgetTokens", apiConfiguration.thinkingBudgetTokens)
+			await updateGlobalState(this.context, "previousModeReasoningEffort", apiConfiguration.reasoningEffort)
 			switch (apiConfiguration.apiProvider) {
 				case "anthropic":
 				case "bedrock":
@@ -925,6 +955,7 @@ export class Controller {
 				case "openai-native":
 				case "qwen":
 				case "deepseek":
+				case "xai":
 					await updateGlobalState(this.context, "previousModeModelId", apiConfiguration.apiModelId)
 					break
 				case "openrouter":
@@ -955,13 +986,21 @@ export class Controller {
 					break
 				case "requesty":
 					await updateGlobalState(this.context, "previousModeModelId", apiConfiguration.requestyModelId)
+					await updateGlobalState(this.context, "previousModeModelInfo", apiConfiguration.requestyModelInfo)
 					break
 			}
 
 			// Restore the model used in previous mode
-			if (newApiProvider || newModelId || newThinkingBudgetTokens !== undefined || newVsCodeLmModelSelector) {
+			if (
+				newApiProvider ||
+				newModelId ||
+				newThinkingBudgetTokens !== undefined ||
+				newReasoningEffort ||
+				newVsCodeLmModelSelector
+			) {
 				await updateGlobalState(this.context, "apiProvider", newApiProvider)
 				await updateGlobalState(this.context, "thinkingBudgetTokens", newThinkingBudgetTokens)
+				await updateGlobalState(this.context, "reasoningEffort", newReasoningEffort)
 				switch (newApiProvider) {
 					case "anthropic":
 					case "bedrock":
@@ -971,6 +1010,7 @@ export class Controller {
 					case "openai-native":
 					case "qwen":
 					case "deepseek":
+					case "xai":
 						await updateGlobalState(this.context, "apiModelId", newModelId)
 						break
 					case "openrouter":
@@ -996,6 +1036,7 @@ export class Controller {
 						break
 					case "requesty":
 						await updateGlobalState(this.context, "requestyModelId", newModelId)
+						await updateGlobalState(this.context, "requestyModelInfo", newModelInfo)
 						break
 				}
 
@@ -1050,7 +1091,7 @@ export class Controller {
 				// 'abandoned' will prevent this cline instance from affecting future cline instance gui. this may happen if its hanging on a streaming request
 				this.task.abandoned = true
 			}
-			await this.initClineWithHistoryItem(historyItem) // clears task again, so we need to abortTask manually above
+			await this.initTask(undefined, undefined, historyItem) // clears task again, so we need to abortTask manually above
 			// await this.postStateToWebview() // new Cline instance will post state when it's ready. having this here sent an empty messages array to webview leading to virtuoso having to reload the entire list
 		}
 	}
@@ -1063,61 +1104,6 @@ export class Controller {
 		}
 	}
 
-	// MCP
-
-	async getDocumentsPath(): Promise<string> {
-		if (process.platform === "win32") {
-			try {
-				const { stdout: docsPath } = await execa("powershell", [
-					"-NoProfile", // Ignore user's PowerShell profile(s)
-					"-Command",
-					"[System.Environment]::GetFolderPath([System.Environment+SpecialFolder]::MyDocuments)",
-				])
-				const trimmedPath = docsPath.trim()
-				if (trimmedPath) {
-					return trimmedPath
-				}
-			} catch (err) {
-				console.error("Failed to retrieve Windows Documents path. Falling back to homedir/Documents.")
-			}
-		} else if (process.platform === "linux") {
-			try {
-				// First check if xdg-user-dir exists
-				await execa("which", ["xdg-user-dir"])
-
-				// If it exists, try to get XDG documents path
-				const { stdout } = await execa("xdg-user-dir", ["DOCUMENTS"])
-				const trimmedPath = stdout.trim()
-				if (trimmedPath) {
-					return trimmedPath
-				}
-			} catch {
-				// Log error but continue to fallback
-				console.error("Failed to retrieve XDG Documents path. Falling back to homedir/Documents.")
-			}
-		}
-
-		// Default fallback for all platforms
-		return path.join(os.homedir(), "Documents")
-	}
-
-	async ensureMcpServersDirectoryExists(): Promise<string> {
-		const userDocumentsPath = await this.getDocumentsPath()
-		const mcpServersDir = path.join(userDocumentsPath, "Cline", "MCP")
-		try {
-			await fs.mkdir(mcpServersDir, { recursive: true })
-		} catch (error) {
-			return "~/Documents/Cline/MCP" // in case creating a directory in documents fails for whatever reason (e.g. permissions) - this is fine since this path is only ever used in the system prompt
-		}
-		return mcpServersDir
-	}
-
-	async ensureSettingsDirectoryExists(): Promise<string> {
-		const settingsDir = path.join(this.context.globalStorageUri.fsPath, "settings")
-		await fs.mkdir(settingsDir, { recursive: true })
-		return settingsDir
-	}
-
 	// VSCode LM API
 
 	private async getVsCodeLmModels() {
@@ -1362,13 +1348,14 @@ export class Controller {
 - Start by loading the MCP documentation.
 - Use "${mcpDetails.mcpId}" as the server name in cline_mcp_settings.json.
 - Create the directory for the new MCP server before starting installation.
+- Make sure you read the user's existing cline_mcp_settings.json file before editing it with this new mcp, to not overwrite any existing servers.
 - Use commands aligned with the user's shell and operating system best practices.
 - The following README may contain instructions that conflict with the user's OS, in which case proceed thoughtfully.
 - Once installed, demonstrate the server's capabilities by using one of its tools.
 Here is the project's README to help you get started:\n\n${mcpDetails.readmeContent}\n${mcpDetails.llmsInstallationContent}`
 
 			// Initialize task and show chat view
-			await this.initClineWithTask(task)
+			await this.initTask(task)
 			await this.postMessageToWebview({
 				type: "action",
 				action: "chatButtonClicked",
@@ -1412,7 +1399,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 				return []
 			}
 
-			const config: Record<string, any> = {}
+			const config: AxiosRequestConfig = {}
 			if (apiKey) {
 				config["headers"] = { Authorization: `Bearer ${apiKey}` }
 			}
@@ -1593,6 +1580,52 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 		return models
 	}
 
+	async refreshRequestyModels() {
+		const parsePrice = (price: any) => {
+			if (price) {
+				return parseFloat(price) * 1_000_000
+			}
+			return undefined
+		}
+
+		let models: Record<string, ModelInfo> = {}
+		try {
+			const apiKey = await getSecret(this.context, "requestyApiKey")
+			const headers = {
+				Authorization: `Bearer ${apiKey}`,
+			}
+			const response = await axios.get("https://router.requesty.ai/v1/models", { headers })
+			if (response.data?.data) {
+				for (const model of response.data.data) {
+					const modelInfo: ModelInfo = {
+						maxTokens: model.max_output_tokens || undefined,
+						contextWindow: model.context_window,
+						supportsImages: model.supports_vision || undefined,
+						supportsComputerUse: model.supports_computer_use || undefined,
+						supportsPromptCache: model.supports_caching || undefined,
+						inputPrice: parsePrice(model.input_price),
+						outputPrice: parsePrice(model.output_price),
+						cacheWritesPrice: parsePrice(model.caching_price),
+						cacheReadsPrice: parsePrice(model.cached_price),
+						description: model.description,
+					}
+					models[model.id] = modelInfo
+				}
+				console.log("Requesty models fetched", models)
+			} else {
+				console.error("Invalid response from Requesty API")
+			}
+		} catch (error) {
+			console.error("Error fetching Requesty models:", error)
+		}
+
+		await this.postMessageToWebview({
+			type: "requestyModels",
+			requestyModels: models,
+		})
+		return models
+	}
+
 	// Context menus and code actions
 
 	getFileMentionFromPath(filePath: string) {
@@ -1656,9 +1689,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 
 		const fileMention = this.getFileMentionFromPath(filePath)
 		const problemsString = this.convertDiagnosticsToProblemsString(diagnostics)
-		await this.initClineWithTask(
-			`Fix the following code in ${fileMention}\n\`\`\`\n${code}\n\`\`\`\n\nProblems:\n${problemsString}`,
-		)
+		await this.initTask(`Fix the following code in ${fileMention}\n\`\`\`\n${code}\n\`\`\`\n\nProblems:\n${problemsString}`)
 
 		console.log("fixWithCline", code, filePath, languageId, diagnostics, problemsString)
 	}
@@ -1698,6 +1729,8 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 		taskDirPath: string
 		apiConversationHistoryFilePath: string
 		uiMessagesFilePath: string
+		contextHistoryFilePath: string
+		taskMetadataFilePath: string
 		apiConversationHistory: Anthropic.MessageParam[]
 	}> {
 		const history = ((await getGlobalState(this.context, "taskHistory")) as HistoryItem[] | undefined) || []
@@ -1706,6 +1739,8 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			const taskDirPath = path.join(this.context.globalStorageUri.fsPath, "tasks", id)
 			const apiConversationHistoryFilePath = path.join(taskDirPath, GlobalFileNames.apiConversationHistory)
 			const uiMessagesFilePath = path.join(taskDirPath, GlobalFileNames.uiMessages)
+			const contextHistoryFilePath = path.join(taskDirPath, GlobalFileNames.contextHistory)
+			const taskMetadataFilePath = path.join(taskDirPath, GlobalFileNames.taskMetadata)
 			const fileExists = await fileExistsAtPath(apiConversationHistoryFilePath)
 			if (fileExists) {
 				const apiConversationHistory = JSON.parse(await fs.readFile(apiConversationHistoryFilePath, "utf8"))
@@ -1714,6 +1749,8 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 					taskDirPath,
 					apiConversationHistoryFilePath,
 					uiMessagesFilePath,
+					contextHistoryFilePath,
+					taskMetadataFilePath,
 					apiConversationHistory,
 				}
 			}
@@ -1728,7 +1765,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 		if (id !== this.task?.taskId) {
 			// non-current task
 			const { historyItem } = await this.getTaskWithId(id)
-			await this.initClineWithHistoryItem(historyItem) // clears existing task
+			await this.initTask(undefined, undefined, historyItem) // clears existing task
 		}
 		await this.postMessageToWebview({
 			type: "action",
@@ -1785,22 +1822,28 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 				console.debug("cleared task")
 			}
 
-			const { taskDirPath, apiConversationHistoryFilePath, uiMessagesFilePath } = await this.getTaskWithId(id)
-
+			const {
+				taskDirPath,
+				apiConversationHistoryFilePath,
+				uiMessagesFilePath,
+				contextHistoryFilePath,
+				taskMetadataFilePath,
+			} = await this.getTaskWithId(id)
+			const legacyMessagesFilePath = path.join(taskDirPath, "claude_messages.json")
 			const updatedTaskHistory = await this.deleteTaskFromState(id)
 
 			// Delete the task files
-			const apiConversationHistoryFileExists = await fileExistsAtPath(apiConversationHistoryFilePath)
-			if (apiConversationHistoryFileExists) {
-				await fs.unlink(apiConversationHistoryFilePath)
-			}
-			const uiMessagesFileExists = await fileExistsAtPath(uiMessagesFilePath)
-			if (uiMessagesFileExists) {
-				await fs.unlink(uiMessagesFilePath)
-			}
-			const legacyMessagesFilePath = path.join(taskDirPath, "claude_messages.json")
-			if (await fileExistsAtPath(legacyMessagesFilePath)) {
-				await fs.unlink(legacyMessagesFilePath)
+			for (const filePath of [
+				apiConversationHistoryFilePath,
+				uiMessagesFilePath,
+				contextHistoryFilePath,
+				taskMetadataFilePath,
+				legacyMessagesFilePath,
+			]) {
+				const fileExists = await fileExistsAtPath(filePath)
+				if (fileExists) {
+					await fs.unlink(filePath)
+				}
 			}
 
 			await fs.rmdir(taskDirPath) // succeeds if the dir is empty

src/core/mentions/index.ts

@@ -10,7 +10,7 @@ import { diagnosticsToProblemsString } from "../../integrations/diagnostics"
 import { getLatestTerminalOutput } from "../../integrations/terminal/get-latest-output"
 import { getCommitInfo } from "../../utils/git"
 import { getWorkingState } from "../../utils/git"
-import { FileContextTracker } from "../context-tracking/FileContextTracker"
+import { FileContextTracker } from "../context/context-tracking/FileContextTracker"
 
 export function openMention(mention?: string): void {
 	if (!mention) {

src/core/prompts/responses.ts

@@ -130,8 +130,8 @@ Otherwise, if you have not completed the task and do not need additional informa
 		cwd: string,
 		wasRecent: boolean | 0 | undefined,
 		responseText?: string,
-	) => {
-		return `[TASK RESUMPTION] ${
+	): [string, string] => {
+		const taskResumptionMessage = `[TASK RESUMPTION] ${
 			mode === "plan"
 				? `This task was interrupted ${agoText}. The conversation may have been incomplete. Be aware that the project state may have changed since then. The current working directory is now '${cwd.toPosix()}'.\n\nNote: If you previously attempted a tool use that the user did not provide a result for, you should assume the tool use was not successful. However you are in PLAN MODE, so rather than continuing the task, you must respond to the user's message.`
 				: `This task was interrupted ${agoText}. It may or may not be complete, so please reassess the task context. Be aware that the project state may have changed since then. The current working directory is now '${cwd.toPosix()}'. If the task has not been completed, retry the last step before interruption and proceed with completing the task.\n\nNote: If you previously attempted a tool use that the user did not provide a result for, you should assume the tool use was not successful and assess whether you should retry. If the last tool was a browser_action, the browser has been closed and you must launch a new browser if needed.`
@@ -139,13 +139,17 @@ Otherwise, if you have not completed the task and do not need additional informa
 			wasRecent
 				? "\n\nIMPORTANT: If the last tool use was a replace_in_file or write_to_file that was interrupted, the file was reverted back to its original state before the interrupted edit, and you do NOT need to re-read the file as you already have its up-to-date contents."
 				: ""
-		}${
+		}`
+
+		const userResponseMessage = `${
 			responseText
-				? `\n\n${mode === "plan" ? "New message to respond to with plan_mode_respond tool (be sure to provide your response in the <response> parameter)" : "New instructions for task continuation"}:\n<user_message>\n${responseText}\n</user_message>`
+				? `${mode === "plan" ? "New message to respond to with plan_mode_respond tool (be sure to provide your response in the <response> parameter)" : "New instructions for task continuation"}:\n<user_message>\n${responseText}\n</user_message>`
 				: mode === "plan"
-					? "(The user did not provide a new message. Consider asking them how they'd like you to proceed, or to switch to Act mode to continue with the task.)"
+					? "(The user did not provide a new message. Consider asking them how they'd like you to proceed, or suggest to them to switch to Act mode to continue with the task.)"
 					: ""
 		}`
+
+		return [taskResumptionMessage, userResponseMessage]
 	},
 
 	planModeInstructions: () => {
@@ -192,7 +196,7 @@ Otherwise, if you have not completed the task and do not need additional informa
 		`This is likely because the SEARCH block content doesn't match exactly with what's in the file, or if you used multiple SEARCH/REPLACE blocks they may not have been in the order they appear in the file.\n\n` +
 		`The file was reverted to its original state:\n\n` +
 		`<file_content path="${relPath.toPosix()}">\n${originalContent}\n</file_content>\n\n` +
-		`Now that you have the latest state of the file, try the operation again with fewer/more precise SEARCH blocks.\n(If you run into this error 3 times in a row, you may use the write_to_file tool as a fallback. Keep in mind, the write_to_file fallback is far from ideal, as this means you'll be re-writing the entire contents of the file just to make a few edits, which takes time and money. So let's bias towards using replace_in_file as effectively as possible)`,
+		`Now that you have the latest state of the file, try the operation again with fewer, more precise SEARCH blocks. For large files especially, it may be prudent to try to limit yourself to <5 SEARCH/REPLACE blocks at a time, then wait for the user to respond with the result of the operation before following up with another replace_in_file call to make additional edits.\n(If you run into this error 3 times in a row, you may use the write_to_file tool as a fallback.)`,
 
 	toolAlreadyUsed: (toolName: string) =>
 		`Tool [${toolName}] was not executed because a tool has already been used in this message. Only one tool may be used per message. You must assess the first tool's result before proceeding to use the next tool.`,
@@ -200,10 +204,13 @@ Otherwise, if you have not completed the task and do not need additional informa
 	clineIgnoreInstructions: (content: string) =>
 		`# .clineignore\n\n(The following is provided by a root-level .clineignore file where the user has specified files and directories that should not be accessed. When using list_files, you'll notice a ${LOCK_TEXT_SYMBOL} next to files that are blocked. Attempting to access the file's contents e.g. through read_file will result in an error.)\n\n${content}\n.clineignore`,
 
-	clineRulesDirectoryInstructions: (cwd: string, content: string) =>
+	clineRulesGlobalDirectoryInstructions: (content: string) =>
+		`# .clinerules/\n\nThe following is provided by a global .clinerules/ directory where the user has specified instructions:\n\n${content}`,
+
+	clineRulesLocalDirectoryInstructions: (cwd: string, content: string) =>
 		`# .clinerules/\n\nThe following is provided by a root-level .clinerules/ directory where the user has specified instructions for this working directory (${cwd.toPosix()})\n\n${content}`,
 
-	clineRulesFileInstructions: (cwd: string, content: string) =>
+	clineRulesLocalFileInstructions: (cwd: string, content: string) =>
 		`# .clinerules\n\nThe following is provided by a root-level .clinerules file where the user has specified instructions for this working directory (${cwd.toPosix()})\n\n${content}`,
 }
 

src/core/prompts/system.ts

@@ -233,17 +233,27 @@ Your final result description here
 <command>Command to demonstrate result (optional)</command>
 </attempt_completion>
 
+## new_task
+Description: Request to create a new task with preloaded context. The user will be presented with a preview of the context and can choose to create a new task or keep chatting in the current conversation. The user may choose to start a new task at any point.
+Parameters:
+- context: (required) The context to preload the new task with. This should include:
+  * Comprehensively explain what has been accomplished in the current task - mention specific file names that are relevant
+  * The specific next steps or focus for the new task - mention specific file names that are relevant
+  * Any critical information needed to continue the work
+  * Clear indication of how this new task relates to the overall workflow
+  * This should be akin to a long handoff file, enough for a totally new developer to be able to pick up where you left off and know exactly what to do next and which files to look at.
+Usage:
+<new_task>
+<context>context to preload new task with</context>
+</new_task>
+
 ## plan_mode_respond
 Description: Respond to the user's inquiry in an effort to plan a solution to the user's task. This tool should be used when you need to provide a response to a question or statement from the user about how you plan to accomplish the task. This tool is only available in PLAN MODE. The environment_details will specify the current mode, if it is not PLAN MODE then you should not use this tool. Depending on the user's message, you may ask questions to get clarification about the user's request, architect a solution to the task, and to brainstorm ideas with the user. For example, if the user's task is to create a website, you may start by asking some clarifying questions, then present a detailed plan for how you will accomplish the task given the context, and perhaps engage in a back and forth to finalize the details before the user switches you to ACT MODE to implement the solution.
 Parameters:
 - response: (required) The response to provide to the user. Do not try to use tools in this parameter, this is simply a chat response. (You MUST use the response parameter, do not simply place the response text directly within <plan_mode_respond> tags.)
-- options: (optional) An array of 2-5 options for the user to choose from. Each option should be a string describing a possible choice or path forward in the planning process. This can help guide the discussion and make it easier for the user to provide input on key decisions. You may not always need to provide options, but it may be helpful in many cases where it can save the user from having to type out a response manually. Do NOT present an option to toggle to Act mode, as this will be something you need to direct the user to do manually themselves.
 Usage:
 <plan_mode_respond>
 <response>Your response here</response>
-<options>
-Array of options here (optional), e.g. ["Option 1", "Option 2", "Option 3"]
-</options>
 </plan_mode_respond>
 
 ## load_mcp_documentation
@@ -284,7 +294,25 @@ Usage:
 </content>
 </write_to_file>
 
-## Example 3: Requesting to make targeted edits to a file
+## Example 3: Creating a new task
+
+<new_task>
+<context>
+Authentication System Implementation:
+- We've implemented the basic user model with email/password
+- Password hashing is working with bcrypt
+- Login endpoint is functional with proper validation
+- JWT token generation is implemented
+
+Next Steps:
+- Implement refresh token functionality
+- Add token validation middleware
+- Create password reset flow
+- Implement role-based access control
+</context>
+</new_task>
+
+## Example 4: Requesting to make targeted edits to a file
 
 <replace_in_file>
 <path>src/components/App.tsx</path>
@@ -319,7 +347,7 @@ return (
 </diff>
 </replace_in_file>
 
-## Example 4: Requesting to use an MCP tool
+## Example 5: Requesting to use an MCP tool
 
 <use_mcp_tool>
 <server_name>weather-server</server_name>
@@ -332,7 +360,7 @@ return (
 </arguments>
 </use_mcp_tool>
 
-## Example 5: Another example of using an MCP tool (where the server name is a unique identifier such as a URL)
+## Example 6: Another example of using an MCP tool (where the server name is a unique identifier such as a URL)
 
 <use_mcp_tool>
 <server_name>github.com/modelcontextprotocol/servers/tree/main/src/github</server_name>
@@ -591,7 +619,8 @@ You accomplish a given task iteratively, breaking it down into clear steps and w
 
 export function addUserInstructions(
 	settingsCustomInstructions?: string,
-	clineRulesFileInstructions?: string,
+	globalClineRulesFileInstructions?: string,
+	localClineRulesFileInstructions?: string,
 	clineIgnoreInstructions?: string,
 	preferredLanguageInstructions?: string,
 ) {
@@ -602,8 +631,11 @@ export function addUserInstructions(
 	if (settingsCustomInstructions) {
 		customInstructions += settingsCustomInstructions + "\n\n"
 	}
-	if (clineRulesFileInstructions) {
-		customInstructions += clineRulesFileInstructions + "\n\n"
+	if (globalClineRulesFileInstructions) {
+		customInstructions += globalClineRulesFileInstructions + "\n\n"
+	}
+	if (localClineRulesFileInstructions) {
+		customInstructions += localClineRulesFileInstructions + "\n\n"
 	}
 	if (clineIgnoreInstructions) {
 		customInstructions += clineIgnoreInstructions

src/core/storage/disk.ts

@@ -4,19 +4,9 @@ import fs from "fs/promises"
 import { Anthropic } from "@anthropic-ai/sdk"
 import { fileExistsAtPath } from "../../utils/fs"
 import { ClineMessage } from "../../shared/ExtensionMessage"
-
-export interface FileMetadataEntry {
-	path: string
-	record_state: "active" | "stale"
-	record_source: "read_tool" | "user_edited" | "cline_edited" | "file_mentioned"
-	cline_read_date: number | null
-	cline_edit_date: number | null
-	user_edit_date?: number | null
-}
-
-export interface TaskMetadata {
-	files_in_context: FileMetadataEntry[]
-}
+import { TaskMetadata } from "../context/context-tracking/ContextTrackerTypes"
+import os from "os"
+import { execa } from "execa"
 
 export const GlobalFileNames = {
 	apiConversationHistory: "api_conversation_history.json",
@@ -28,6 +18,42 @@ export const GlobalFileNames = {
 	taskMetadata: "task_metadata.json",
 }
 
+export async function getDocumentsPath(): Promise<string> {
+	if (process.platform === "win32") {
+		try {
+			const { stdout: docsPath } = await execa("powershell", [
+				"-NoProfile", // Ignore user's PowerShell profile(s)
+				"-Command",
+				"[System.Environment]::GetFolderPath([System.Environment+SpecialFolder]::MyDocuments)",
+			])
+			const trimmedPath = docsPath.trim()
+			if (trimmedPath) {
+				return trimmedPath
+			}
+		} catch (err) {
+			console.error("Failed to retrieve Windows Documents path. Falling back to homedir/Documents.")
+		}
+	} else if (process.platform === "linux") {
+		try {
+			// First check if xdg-user-dir exists
+			await execa("which", ["xdg-user-dir"])
+
+			// If it exists, try to get XDG documents path
+			const { stdout } = await execa("xdg-user-dir", ["DOCUMENTS"])
+			const trimmedPath = stdout.trim()
+			if (trimmedPath) {
+				return trimmedPath
+			}
+		} catch {
+			// Log error but continue to fallback
+			console.error("Failed to retrieve XDG Documents path. Falling back to homedir/Documents.")
+		}
+	}
+
+	// Default fallback for all platforms
+	return path.join(os.homedir(), "Documents")
+}
+
 export async function ensureTaskDirectoryExists(context: vscode.ExtensionContext, taskId: string): Promise<string> {
 	const globalStoragePath = context.globalStorageUri.fsPath
 	const taskDir = path.join(globalStoragePath, "tasks", taskId)
@@ -35,6 +61,34 @@ export async function ensureTaskDirectoryExists(context: vscode.ExtensionContext
 	return taskDir
 }
 
+export async function ensureRulesDirectoryExists(): Promise<string> {
+	const userDocumentsPath = await getDocumentsPath()
+	const clineRulesDir = path.join(userDocumentsPath, "Cline", "Rules")
+	try {
+		await fs.mkdir(clineRulesDir, { recursive: true })
+	} catch (error) {
+		return path.join(os.homedir(), "Documents", "Cline", "Rules") // in case creating a directory in documents fails for whatever reason (e.g. permissions) - this is fine because we will fail gracefully with a path that does not exist
+	}
+	return clineRulesDir
+}
+
+export async function ensureMcpServersDirectoryExists(): Promise<string> {
+	const userDocumentsPath = await getDocumentsPath()
+	const mcpServersDir = path.join(userDocumentsPath, "Cline", "MCP")
+	try {
+		await fs.mkdir(mcpServersDir, { recursive: true })
+	} catch (error) {
+		return "~/Documents/Cline/MCP" // in case creating a directory in documents fails for whatever reason (e.g. permissions) - this is fine since this path is only ever used in the system prompt
+	}
+	return mcpServersDir
+}
+
+export async function ensureSettingsDirectoryExists(context: vscode.ExtensionContext): Promise<string> {
+	const settingsDir = path.join(context.globalStorageUri.fsPath, "settings")
+	await fs.mkdir(settingsDir, { recursive: true })
+	return settingsDir
+}
+
 export async function getSavedApiConversationHistory(
 	context: vscode.ExtensionContext,
 	taskId: string,
@@ -96,7 +150,7 @@ export async function getTaskMetadata(context: vscode.ExtensionContext, taskId:
 	} catch (error) {
 		console.error("Failed to read task metadata:", error)
 	}
-	return { files_in_context: [] }
+	return { files_in_context: [], model_usage: [] }
 }
 
 export async function saveTaskMetadata(context: vscode.ExtensionContext, taskId: string, metadata: TaskMetadata) {

src/core/storage/state-keys.ts

@@ -54,6 +54,7 @@ export type GlobalStateKey =
 	| "previousModeApiProvider"
 	| "previousModeModelId"
 	| "previousModeThinkingBudgetTokens"
+	| "previousModeReasoningEffort"
 	| "previousModeVsCodeLmModelSelector"
 	| "previousModeModelInfo"
 	| "liteLlmBaseUrl"
@@ -61,9 +62,12 @@ export type GlobalStateKey =
 	| "liteLlmUsePromptCache"
 	| "qwenApiLine"
 	| "requestyModelId"
+	| "requestyModelInfo"
 	| "togetherModelId"
 	| "mcpMarketplaceCatalog"
 	| "telemetrySetting"
 	| "asksageApiUrl"
 	| "thinkingBudgetTokens"
+	| "reasoningEffort"
 	| "planActSeparateModelsSetting"
+	| "favoritedModelIds"

src/core/storage/state.ts

@@ -83,6 +83,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		deepSeekApiKey,
 		requestyApiKey,
 		requestyModelId,
+		requestyModelInfo,
 		togetherApiKey,
 		togetherModelId,
 		qwenApiKey,
@@ -108,6 +109,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		previousModeModelInfo,
 		previousModeVsCodeLmModelSelector,
 		previousModeThinkingBudgetTokens,
+		previousModeReasoningEffort,
 		qwenApiLine,
 		liteLlmApiKey,
 		telemetrySetting,
@@ -115,8 +117,10 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		asksageApiUrl,
 		xaiApiKey,
 		thinkingBudgetTokens,
+		reasoningEffort,
 		sambanovaApiKey,
 		planActSeparateModelsSettingRaw,
+		favoritedModelIds,
 	] = await Promise.all([
 		getGlobalState(context, "apiProvider") as Promise<ApiProvider | undefined>,
 		getGlobalState(context, "apiModelId") as Promise<string | undefined>,
@@ -149,6 +153,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		getSecret(context, "deepSeekApiKey") as Promise<string | undefined>,
 		getSecret(context, "requestyApiKey") as Promise<string | undefined>,
 		getGlobalState(context, "requestyModelId") as Promise<string | undefined>,
+		getGlobalState(context, "requestyModelInfo") as Promise<ModelInfo | undefined>,
 		getSecret(context, "togetherApiKey") as Promise<string | undefined>,
 		getGlobalState(context, "togetherModelId") as Promise<string | undefined>,
 		getSecret(context, "qwenApiKey") as Promise<string | undefined>,
@@ -174,6 +179,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		getGlobalState(context, "previousModeModelInfo") as Promise<ModelInfo | undefined>,
 		getGlobalState(context, "previousModeVsCodeLmModelSelector") as Promise<vscode.LanguageModelChatSelector | undefined>,
 		getGlobalState(context, "previousModeThinkingBudgetTokens") as Promise<number | undefined>,
+		getGlobalState(context, "previousModeReasoningEffort") as Promise<string | undefined>,
 		getGlobalState(context, "qwenApiLine") as Promise<string | undefined>,
 		getSecret(context, "liteLlmApiKey") as Promise<string | undefined>,
 		getGlobalState(context, "telemetrySetting") as Promise<TelemetrySetting | undefined>,
@@ -181,8 +187,10 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		getGlobalState(context, "asksageApiUrl") as Promise<string | undefined>,
 		getSecret(context, "xaiApiKey") as Promise<string | undefined>,
 		getGlobalState(context, "thinkingBudgetTokens") as Promise<number | undefined>,
+		getGlobalState(context, "reasoningEffort") as Promise<string | undefined>,
 		getSecret(context, "sambanovaApiKey") as Promise<string | undefined>,
 		getGlobalState(context, "planActSeparateModelsSetting") as Promise<boolean | undefined>,
+		getGlobalState(context, "favoritedModelIds") as Promise<string[] | undefined>,
 	])
 
 	let apiProvider: ApiProvider
@@ -254,6 +262,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 			deepSeekApiKey,
 			requestyApiKey,
 			requestyModelId,
+			requestyModelInfo,
 			togetherApiKey,
 			togetherModelId,
 			qwenApiKey,
@@ -267,6 +276,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 			vsCodeLmModelSelector,
 			o3MiniReasoningEffort,
 			thinkingBudgetTokens,
+			reasoningEffort,
 			liteLlmBaseUrl,
 			liteLlmModelId,
 			liteLlmApiKey,
@@ -275,6 +285,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 			asksageApiUrl,
 			xaiApiKey,
 			sambanovaApiKey,
+			favoritedModelIds,
 		},
 		lastShownAnnouncementId,
 		customInstructions,
@@ -288,6 +299,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		previousModeModelInfo,
 		previousModeVsCodeLmModelSelector,
 		previousModeThinkingBudgetTokens,
+		previousModeReasoningEffort,
 		mcpMarketplaceEnabled,
 		telemetrySetting: telemetrySetting || "unset",
 		planActSeparateModelsSetting,
@@ -326,6 +338,7 @@ export async function updateApiConfiguration(context: vscode.ExtensionContext, a
 		deepSeekApiKey,
 		requestyApiKey,
 		requestyModelId,
+		requestyModelInfo,
 		togetherApiKey,
 		togetherModelId,
 		qwenApiKey,
@@ -345,8 +358,10 @@ export async function updateApiConfiguration(context: vscode.ExtensionContext, a
 		asksageApiUrl,
 		xaiApiKey,
 		thinkingBudgetTokens,
+		reasoningEffort,
 		clineApiKey,
 		sambanovaApiKey,
+		favoritedModelIds,
 	} = apiConfiguration
 	await updateGlobalState(context, "apiProvider", apiProvider)
 	await updateGlobalState(context, "apiModelId", apiModelId)
@@ -393,12 +408,15 @@ export async function updateApiConfiguration(context: vscode.ExtensionContext, a
 	await updateGlobalState(context, "liteLlmUsePromptCache", liteLlmUsePromptCache)
 	await updateGlobalState(context, "qwenApiLine", qwenApiLine)
 	await updateGlobalState(context, "requestyModelId", requestyModelId)
+	await updateGlobalState(context, "requestyModelInfo", requestyModelInfo)
 	await updateGlobalState(context, "togetherModelId", togetherModelId)
 	await storeSecret(context, "asksageApiKey", asksageApiKey)
 	await updateGlobalState(context, "asksageApiUrl", asksageApiUrl)
 	await updateGlobalState(context, "thinkingBudgetTokens", thinkingBudgetTokens)
+	await updateGlobalState(context, "reasoningEffort", reasoningEffort)
 	await storeSecret(context, "clineApiKey", clineApiKey)
 	await storeSecret(context, "sambanovaApiKey", sambanovaApiKey)
+	await updateGlobalState(context, "favoritedModelIds", favoritedModelIds)
 }
 
 export async function resetExtensionState(context: vscode.ExtensionContext) {

src/core/task/index.ts

@@ -1,6 +1,5 @@
 import { Anthropic } from "@anthropic-ai/sdk"
 import cloneDeep from "clone-deep"
-import fs from "fs/promises"
 import getFolderSize from "get-folder-size"
 import { setTimeout as setTimeoutPromise } from "node:timers/promises"
 import os from "os"
@@ -48,37 +47,43 @@ import {
 	ClineSayBrowserAction,
 	ClineSayTool,
 	COMPLETION_RESULT_CHANGES_FLAG,
+	ExtensionMessage,
 } from "../../shared/ExtensionMessage"
 import { getApiMetrics } from "../../shared/getApiMetrics"
 import { HistoryItem } from "../../shared/HistoryItem"
 import { DEFAULT_LANGUAGE_SETTINGS, getLanguageKey, LanguageDisplay } from "../../shared/Languages"
 import { ClineAskResponse, ClineCheckpointRestore } from "../../shared/WebviewMessage"
 import { calculateApiCostAnthropic } from "../../utils/cost"
-import { fileExistsAtPath, isDirectory } from "../../utils/fs"
-import { arePathsEqual, getReadablePath } from "../../utils/path"
+import { fileExistsAtPath } from "../../utils/fs"
+import { arePathsEqual, getReadablePath, isLocatedInWorkspace } from "../../utils/path"
 import { fixModelHtmlEscaping, removeInvalidChars } from "../../utils/string"
 import { AssistantMessageContent, parseAssistantMessage, ToolParamName, ToolUseName } from ".././assistant-message"
 import { constructNewFileContent } from ".././assistant-message/diff"
-import { ContextManager } from ".././context-management/ContextManager"
 import { ClineIgnoreController } from ".././ignore/ClineIgnoreController"
 import { parseMentions } from ".././mentions"
 import { formatResponse } from ".././prompts/responses"
 import { addUserInstructions, SYSTEM_PROMPT } from ".././prompts/system"
-import { FileContextTracker } from "../context-tracking/FileContextTracker"
+import { getContextWindowInfo } from "../context/context-management/context-window-utils"
+import { FileContextTracker } from "../context/context-tracking/FileContextTracker"
+import { ModelContextTracker } from "../context/context-tracking/ModelContextTracker"
 import {
 	checkIsAnthropicContextWindowError,
 	checkIsOpenRouterContextWindowError,
-} from "../context-management/context-error-handling"
-import { Controller } from "../controller"
+} from "../context/context-management/context-error-handling"
+import WorkspaceTracker from "../../integrations/workspace/WorkspaceTracker"
+import { McpHub } from "../../services/mcp/McpHub"
+import { ContextManager } from "../context/context-management/ContextManager"
+import { loadMcpDocumentation } from "../prompts/loadMcpDocumentation"
 import {
+	ensureRulesDirectoryExists,
 	ensureTaskDirectoryExists,
 	getSavedApiConversationHistory,
 	getSavedClineMessages,
 	saveApiConversationHistory,
 	saveClineMessages,
-	GlobalFileNames,
 } from "../storage/disk"
-import { loadMcpDocumentation } from "../prompts/loadMcpDocumentation"
+import { getGlobalClineRules, getLocalClineRules } from "../context/instructions/user-instructions/cline-rules"
+import { getGlobalState } from "../storage/state"
 
 const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) ?? path.join(os.homedir(), "Desktop") // may or may not exist but fs checking existence would immediately ask for permission which would be bad UX, need to come up with a better solution
 
@@ -86,8 +91,17 @@ type ToolResponse = string | Array<Anthropic.TextBlockParam | Anthropic.ImageBlo
 type UserContent = Array<Anthropic.ContentBlockParam>
 
 export class Task {
+	// dependencies
+	private context: vscode.ExtensionContext
+	private mcpHub: McpHub
+	private workspaceTracker: WorkspaceTracker
+	private updateTaskHistory: (historyItem: HistoryItem) => Promise<HistoryItem[]>
+	private postStateToWebview: () => Promise<void>
+	private postMessageToWebview: (message: ExtensionMessage) => Promise<void>
+	private reinitExistingTaskFromId: (taskId: string) => Promise<void>
+	private cancelTask: () => Promise<void>
+
 	readonly taskId: string
-	readonly apiProvider?: string
 	api: ApiHandler
 	private terminalManager: TerminalManager
 	private urlContentFetcher: UrlContentFetcher
@@ -107,7 +121,6 @@ export class Task {
 	private lastMessageTs?: number
 	private consecutiveAutoApprovedRequestsCount: number = 0
 	private consecutiveMistakeCount: number = 0
-	private controllerRef: WeakRef<Controller>
 	private abort: boolean = false
 	didFinishAbortingStream = false
 	abandoned = false
@@ -119,8 +132,9 @@ export class Task {
 	isAwaitingPlanResponse = false
 	didRespondToPlanAskBySwitchingMode = false
 
-	// File tracking
+	// Metadata tracking
 	private fileContextTracker: FileContextTracker
+	private modelContextTracker: ModelContextTracker
 
 	// streaming
 	isWaitingForFirstChunk = false
@@ -137,7 +151,14 @@ export class Task {
 	private didAutomaticallyRetryFailedApiRequest = false
 
 	constructor(
-		controller: Controller,
+		context: vscode.ExtensionContext,
+		mcpHub: McpHub,
+		workspaceTracker: WorkspaceTracker,
+		updateTaskHistory: (historyItem: HistoryItem) => Promise<HistoryItem[]>,
+		postStateToWebview: () => Promise<void>,
+		postMessageToWebview: (message: ExtensionMessage) => Promise<void>,
+		reinitExistingTaskFromId: (taskId: string) => Promise<void>,
+		cancelTask: () => Promise<void>,
 		apiConfiguration: ApiConfiguration,
 		autoApprovalSettings: AutoApprovalSettings,
 		browserSettings: BrowserSettings,
@@ -147,15 +168,21 @@ export class Task {
 		images?: string[],
 		historyItem?: HistoryItem,
 	) {
+		this.context = context
+		this.mcpHub = mcpHub
+		this.workspaceTracker = workspaceTracker
+		this.updateTaskHistory = updateTaskHistory
+		this.postStateToWebview = postStateToWebview
+		this.postMessageToWebview = postMessageToWebview
+		this.reinitExistingTaskFromId = reinitExistingTaskFromId
+		this.cancelTask = cancelTask
 		this.clineIgnoreController = new ClineIgnoreController(cwd)
 		this.clineIgnoreController.initialize().catch((error) => {
 			console.error("Failed to initialize ClineIgnoreController:", error)
 		})
-		this.controllerRef = new WeakRef(controller)
-		this.apiProvider = apiConfiguration.apiProvider
 		this.terminalManager = new TerminalManager()
-		this.urlContentFetcher = new UrlContentFetcher(controller.context)
-		this.browserSession = new BrowserSession(controller.context, browserSettings)
+		this.urlContentFetcher = new UrlContentFetcher(context)
+		this.browserSession = new BrowserSession(context, browserSettings)
 		this.contextManager = new ContextManager()
 		this.diffViewProvider = new DiffViewProvider(cwd)
 		this.customInstructions = customInstructions
@@ -174,8 +201,8 @@ export class Task {
 		}
 
 		// Initialize file context tracker
-		this.fileContextTracker = new FileContextTracker(controller, this.taskId)
-
+		this.fileContextTracker = new FileContextTracker(context, this.taskId)
+		this.modelContextTracker = new ModelContextTracker(context, this.taskId)
 		// Now that taskId is initialized, we can build the API handler
 		this.api = buildApiHandler({
 			...apiConfiguration,
@@ -192,19 +219,20 @@ export class Task {
 			this.startTask(task, images)
 		}
 
+		// initialize telemetry
 		if (historyItem) {
 			// Open task from history
-			telemetryService.captureTaskRestarted(this.taskId, this.apiProvider)
+			telemetryService.captureTaskRestarted(this.taskId, apiConfiguration.apiProvider)
 		} else {
 			// New task started
-			telemetryService.captureTaskCreated(this.taskId, this.apiProvider)
+			telemetryService.captureTaskCreated(this.taskId, apiConfiguration.apiProvider)
 		}
 	}
 
 	// While a task is ref'd by a controller, it will always have access to the extension context
 	// This error is thrown if the controller derefs the task after e.g., aborting the task
 	private getContext(): vscode.ExtensionContext {
-		const context = this.controllerRef.deref()?.context
+		const context = this.context
 		if (!context) {
 			throw new Error("Unable to access extension context")
 		}
@@ -256,7 +284,7 @@ export class Task {
 			} catch (error) {
 				console.error("Failed to get task directory size:", taskDir, error)
 			}
-			await this.controllerRef.deref()?.updateTaskHistory({
+			await this.updateTaskHistory({
 				id: this.taskId,
 				ts: lastRelevantMessage.ts,
 				task: taskMessage.text ?? "",
@@ -291,15 +319,12 @@ export class Task {
 			case "workspace":
 				if (!this.checkpointTracker && !this.checkpointTrackerErrorMessage) {
 					try {
-						this.checkpointTracker = await CheckpointTracker.create(
-							this.taskId,
-							this.controllerRef.deref()?.context.globalStorageUri.fsPath,
-						)
+						this.checkpointTracker = await CheckpointTracker.create(this.taskId, this.context.globalStorageUri.fsPath)
 					} catch (error) {
 						const errorMessage = error instanceof Error ? error.message : "Unknown error"
 						console.error("Failed to initialize checkpoint tracker:", errorMessage)
 						this.checkpointTrackerErrorMessage = errorMessage
-						await this.controllerRef.deref()?.postStateToWebview()
+						await this.postStateToWebview()
 						vscode.window.showErrorMessage(errorMessage)
 						didWorkspaceRestoreFail = true
 					}
@@ -381,17 +406,17 @@ export class Task {
 
 			await this.saveClineMessagesAndUpdateHistory()
 
-			await this.controllerRef.deref()?.postMessageToWebview({ type: "relinquishControl" })
+			await this.postMessageToWebview({ type: "relinquishControl" })
 
-			this.controllerRef.deref()?.cancelTask() // the task is already cancelled by the provider beforehand, but we need to re-init to get the updated messages
+			this.cancelTask() // the task is already cancelled by the provider beforehand, but we need to re-init to get the updated messages
 		} else {
-			await this.controllerRef.deref()?.postMessageToWebview({ type: "relinquishControl" })
+			await this.postMessageToWebview({ type: "relinquishControl" })
 		}
 	}
 
 	async presentMultifileDiff(messageTs: number, seeNewChangesSinceLastTaskCompletion: boolean) {
 		const relinquishButton = () => {
-			this.controllerRef.deref()?.postMessageToWebview({ type: "relinquishControl" })
+			this.postMessageToWebview({ type: "relinquishControl" })
 		}
 
 		console.log("presentMultifileDiff", messageTs)
@@ -412,15 +437,12 @@ export class Task {
 		// TODO: handle if this is called from outside original workspace, in which case we need to show user error message we cant show diff outside of workspace?
 		if (!this.checkpointTracker && !this.checkpointTrackerErrorMessage) {
 			try {
-				this.checkpointTracker = await CheckpointTracker.create(
-					this.taskId,
-					this.controllerRef.deref()?.context.globalStorageUri.fsPath,
-				)
+				this.checkpointTracker = await CheckpointTracker.create(this.taskId, this.context.globalStorageUri.fsPath)
 			} catch (error) {
 				const errorMessage = error instanceof Error ? error.message : "Unknown error"
 				console.error("Failed to initialize checkpoint tracker:", errorMessage)
 				this.checkpointTrackerErrorMessage = errorMessage
-				await this.controllerRef.deref()?.postStateToWebview()
+				await this.postStateToWebview()
 				vscode.window.showErrorMessage(errorMessage)
 				relinquishButton()
 				return
@@ -527,10 +549,7 @@ export class Task {
 
 		if (!this.checkpointTracker && !this.checkpointTrackerErrorMessage) {
 			try {
-				this.checkpointTracker = await CheckpointTracker.create(
-					this.taskId,
-					this.controllerRef.deref()?.context.globalStorageUri.fsPath,
-				)
+				this.checkpointTracker = await CheckpointTracker.create(this.taskId, this.context.globalStorageUri.fsPath)
 			} catch (error) {
 				const errorMessage = error instanceof Error ? error.message : "Unknown error"
 				console.error("Failed to initialize checkpoint tracker:", errorMessage)
@@ -601,8 +620,8 @@ export class Task {
 					lastMessage.partial = partial
 					// todo be more efficient about saving and posting only new data or one whole message at a time so ignore partial for saves, and only post parts of partial message instead of whole array in new listener
 					// await this.saveClineMessagesAndUpdateHistory()
-					// await this.controllerRef.deref()?.postStateToWebview()
-					await this.controllerRef.deref()?.postMessageToWebview({
+					// await this.postStateToWebview()
+					await this.postMessageToWebview({
 						type: "partialMessage",
 						partialMessage: lastMessage,
 					})
@@ -621,7 +640,7 @@ export class Task {
 						text,
 						partial,
 					})
-					await this.controllerRef.deref()?.postStateToWebview()
+					await this.postStateToWebview()
 					throw new Error("Current ask promise was ignored 2")
 				}
 			} else {
@@ -644,8 +663,8 @@ export class Task {
 					lastMessage.text = text
 					lastMessage.partial = false
 					await this.saveClineMessagesAndUpdateHistory()
-					// await this.controllerRef.deref()?.postStateToWebview()
-					await this.controllerRef.deref()?.postMessageToWebview({
+					// await this.postStateToWebview()
+					await this.postMessageToWebview({
 						type: "partialMessage",
 						partialMessage: lastMessage,
 					})
@@ -662,7 +681,7 @@ export class Task {
 						ask: type,
 						text,
 					})
-					await this.controllerRef.deref()?.postStateToWebview()
+					await this.postStateToWebview()
 				}
 			}
 		} else {
@@ -679,7 +698,7 @@ export class Task {
 				ask: type,
 				text,
 			})
-			await this.controllerRef.deref()?.postStateToWebview()
+			await this.postStateToWebview()
 		}
 
 		await pWaitFor(() => this.askResponse !== undefined || this.lastMessageTs !== askTs, { interval: 100 })
@@ -718,7 +737,7 @@ export class Task {
 					lastMessage.text = text
 					lastMessage.images = images
 					lastMessage.partial = partial
-					await this.controllerRef.deref()?.postMessageToWebview({
+					await this.postMessageToWebview({
 						type: "partialMessage",
 						partialMessage: lastMessage,
 					})
@@ -734,7 +753,7 @@ export class Task {
 						images,
 						partial,
 					})
-					await this.controllerRef.deref()?.postStateToWebview()
+					await this.postStateToWebview()
 				}
 			} else {
 				// partial=false means its a complete version of a previously partial message
@@ -748,8 +767,8 @@ export class Task {
 
 					// instead of streaming partialMessage events, we do a save and post like normal to persist to disk
 					await this.saveClineMessagesAndUpdateHistory()
-					// await this.controllerRef.deref()?.postStateToWebview()
-					await this.controllerRef.deref()?.postMessageToWebview({
+					// await this.postStateToWebview()
+					await this.postMessageToWebview({
 						type: "partialMessage",
 						partialMessage: lastMessage,
 					}) // more performant than an entire postStateToWebview
@@ -764,7 +783,7 @@ export class Task {
 						text,
 						images,
 					})
-					await this.controllerRef.deref()?.postStateToWebview()
+					await this.postStateToWebview()
 				}
 			}
 		} else {
@@ -778,7 +797,7 @@ export class Task {
 				text,
 				images,
 			})
-			await this.controllerRef.deref()?.postStateToWebview()
+			await this.postStateToWebview()
 		}
 	}
 
@@ -797,7 +816,7 @@ export class Task {
 		if (lastMessage?.partial && lastMessage.type === type && (lastMessage.ask === askOrSay || lastMessage.say === askOrSay)) {
 			this.clineMessages.pop()
 			await this.saveClineMessagesAndUpdateHistory()
-			await this.controllerRef.deref()?.postStateToWebview()
+			await this.postStateToWebview()
 		}
 	}
 
@@ -809,7 +828,7 @@ export class Task {
 		this.clineMessages = []
 		this.apiConversationHistory = []
 
-		await this.controllerRef.deref()?.postStateToWebview()
+		await this.postStateToWebview()
 
 		await this.say("text", task, images)
 
@@ -942,16 +961,27 @@ export class Task {
 
 		const wasRecent = lastClineMessage?.ts && Date.now() - lastClineMessage.ts < 30_000
 
-		newUserContent.push({
-			type: "text",
-			text: formatResponse.taskResumption(
-				this.chatSettings?.mode === "plan" ? "plan" : "act",
-				agoText,
-				cwd,
-				wasRecent,
-				responseText,
-			),
-		})
+		const [taskResumptionMessage, userResponseMessage] = formatResponse.taskResumption(
+			this.chatSettings?.mode === "plan" ? "plan" : "act",
+			agoText,
+			cwd,
+			wasRecent,
+			responseText,
+		)
+
+		if (taskResumptionMessage !== "") {
+			newUserContent.push({
+				type: "text",
+				text: taskResumptionMessage,
+			})
+		}
+
+		if (userResponseMessage !== "") {
+			newUserContent.push({
+				type: "text",
+				text: userResponseMessage,
+			})
+		}
 
 		if (responseImages && responseImages.length > 0) {
 			newUserContent.push(...formatResponse.imageBlocks(responseImages))
@@ -1013,6 +1043,12 @@ export class Task {
 		})
 
 		if (!isAttemptCompletionMessage) {
+			// ensure we aren't creating a duplicate checkpoint
+			const lastMessage = this.clineMessages.at(-1)
+			if (lastMessage?.say === "checkpoint_created") {
+				return
+			}
+
 			// For non-attempt completion we just say checkpoints
 			await this.say("checkpoint_created")
 			this.checkpointTracker?.commit().then(async (commitHash) => {
@@ -1155,7 +1191,7 @@ export class Task {
 	}
 
 	// Check if the tool should be auto-approved based on the settings
-	// Returns bool for most tools, tuple for execute_command (and future nested auto appoved settings)
+	// Returns bool for most tools, and tuple for tools with nested settings
 	shouldAutoApproveTool(toolName: ToolUseName): boolean | [boolean, boolean] {
 		if (this.autoApprovalSettings.enabled) {
 			switch (toolName) {
@@ -1163,14 +1199,20 @@ export class Task {
 				case "list_files":
 				case "list_code_definition_names":
 				case "search_files":
-					return this.autoApprovalSettings.actions.readFiles
+					return [
+						this.autoApprovalSettings.actions.readFiles,
+						this.autoApprovalSettings.actions.readFilesExternally ?? false,
+					]
 				case "write_to_file":
 				case "replace_in_file":
-					return this.autoApprovalSettings.actions.editFiles
+					return [
+						this.autoApprovalSettings.actions.editFiles,
+						this.autoApprovalSettings.actions.editFilesExternally ?? false,
+					]
 				case "execute_command":
 					return [
-						this.autoApprovalSettings.actions.executeSafeCommands,
-						this.autoApprovalSettings.actions.executeAllCommands,
+						this.autoApprovalSettings.actions.executeSafeCommands ?? false,
+						this.autoApprovalSettings.actions.executeAllCommands ?? false,
 					]
 				case "browser_action":
 					return this.autoApprovalSettings.actions.useBrowser
@@ -1182,6 +1224,32 @@ export class Task {
 		return false
 	}
 
+	// Check if the tool should be auto-approved based on the settings
+	// and the path of the action. Returns true if the tool should be auto-approved
+	// based on the user's settings and the path of the action.
+	shouldAutoApproveToolWithPath(blockname: ToolUseName, autoApproveActionpath: string | undefined): boolean {
+		let isLocalRead: boolean = false
+		if (autoApproveActionpath) {
+			const absolutePath = path.resolve(cwd, autoApproveActionpath)
+			isLocalRead = absolutePath.startsWith(cwd)
+		} else {
+			// If we do not get a path for some reason, default to a (safer) false return
+			isLocalRead = false
+		}
+
+		// Get auto-approve settings for local and external edits
+		const autoApproveResult = this.shouldAutoApproveTool(blockname)
+		const [autoApproveLocal, autoApproveExternal] = Array.isArray(autoApproveResult)
+			? autoApproveResult
+			: [autoApproveResult, false]
+
+		if ((isLocalRead && autoApproveLocal) || (!isLocalRead && autoApproveLocal && autoApproveExternal)) {
+			return true
+		} else {
+			return false
+		}
+	}
+
 	private formatErrorWithStatusCode(error: any): string {
 		const statusCode = error.status || error.statusCode || (error.response && error.response.status)
 		const message = error.message ?? JSON.stringify(serializeError(error), null, 2)
@@ -1192,21 +1260,16 @@ export class Task {
 
 	async *attemptApiRequest(previousApiReqIndex: number): ApiStream {
 		// Wait for MCP servers to be connected before generating system prompt
-		await pWaitFor(() => this.controllerRef.deref()?.mcpHub?.isConnecting !== true, { timeout: 10_000 }).catch(() => {
+		await pWaitFor(() => this.mcpHub.isConnecting !== true, { timeout: 10_000 }).catch(() => {
 			console.error("MCP servers failed to connect in time")
 		})
 
-		const mcpHub = this.controllerRef.deref()?.mcpHub
-		if (!mcpHub) {
-			throw new Error("MCP hub not available")
-		}
-
 		const disableBrowserTool = vscode.workspace.getConfiguration("cline").get<boolean>("disableBrowserTool") ?? false
 		const modelSupportsComputerUse = this.api.getModel().info.supportsComputerUse ?? false
 
 		const supportsComputerUse = modelSupportsComputerUse && !disableBrowserTool // only enable computer use if the model supports it and the user hasn't disabled it
 
-		let systemPrompt = await SYSTEM_PROMPT(cwd, supportsComputerUse, mcpHub, this.browserSettings)
+		let systemPrompt = await SYSTEM_PROMPT(cwd, supportsComputerUse, this.mcpHub, this.browserSettings)
 
 		let settingsCustomInstructions = this.customInstructions?.trim()
 		const preferredLanguage = getLanguageKey(
@@ -1216,38 +1279,11 @@ export class Task {
 			preferredLanguage && preferredLanguage !== DEFAULT_LANGUAGE_SETTINGS
 				? `# Preferred Language\n\nSpeak in ${preferredLanguage}.`
 				: ""
-		const clineRulesFilePath = path.resolve(cwd, GlobalFileNames.clineRules)
-		let clineRulesFileInstructions: string | undefined
-		if (await fileExistsAtPath(clineRulesFilePath)) {
-			if (await isDirectory(clineRulesFilePath)) {
-				try {
-					// Read all files in the .clinerules/ directory.
-					const ruleFiles = await fs
-						.readdir(clineRulesFilePath, { withFileTypes: true, recursive: true })
-						.then((files) => files.filter((file) => file.isFile()))
-						.then((files) => files.map((file) => path.resolve(file.parentPath, file.name)))
-					const ruleFilesTotalContent = await Promise.all(
-						ruleFiles.map(async (file) => {
-							const ruleFilePath = path.resolve(clineRulesFilePath, file)
-							const ruleFilePathRelative = path.relative(cwd, ruleFilePath)
-							return `${ruleFilePathRelative}\n` + (await fs.readFile(ruleFilePath, "utf8")).trim()
-						}),
-					).then((contents) => contents.join("\n\n"))
-					clineRulesFileInstructions = formatResponse.clineRulesDirectoryInstructions(cwd, ruleFilesTotalContent)
-				} catch {
-					console.error(`Failed to read .clinerules directory at ${clineRulesFilePath}`)
-				}
-			} else {
-				try {
-					const ruleFileContent = (await fs.readFile(clineRulesFilePath, "utf8")).trim()
-					if (ruleFileContent) {
-						clineRulesFileInstructions = formatResponse.clineRulesFileInstructions(cwd, ruleFileContent)
-					}
-				} catch {
-					console.error(`Failed to read .clinerules file at ${clineRulesFilePath}`)
-				}
-			}
-		}
+
+		const localClineRulesFileInstructions = await getLocalClineRules(cwd)
+
+		const globalClineRulesFilePath = await ensureRulesDirectoryExists()
+		const globalClineRulesFileInstructions = await getGlobalClineRules(globalClineRulesFilePath)
 
 		const clineIgnoreContent = this.clineIgnoreController.clineIgnoreContent
 		let clineIgnoreInstructions: string | undefined
@@ -1257,14 +1293,16 @@ export class Task {
 
 		if (
 			settingsCustomInstructions ||
-			clineRulesFileInstructions ||
+			globalClineRulesFileInstructions ||
+			localClineRulesFileInstructions ||
 			clineIgnoreInstructions ||
 			preferredLanguageInstructions
 		) {
 			// altering the system prompt mid-task will break the prompt cache, but in the grand scheme this will not change often so it's better to not pollute user messages with it the way we have to with <potentially relevant details>
 			systemPrompt += addUserInstructions(
 				settingsCustomInstructions,
-				clineRulesFileInstructions,
+				globalClineRulesFileInstructions,
+				localClineRulesFileInstructions,
 				clineIgnoreInstructions,
 				preferredLanguageInstructions,
 			)
@@ -1475,6 +1513,8 @@ export class Task {
 							return `[${block.name}]`
 						case "attempt_completion":
 							return `[${block.name}]`
+						case "new_task":
+							return `[${block.name} for creating a new task]`
 					}
 				}
 
@@ -1715,12 +1755,14 @@ export class Task {
 								tool: fileExists ? "editedExistingFile" : "newFileCreated",
 								path: getReadablePath(cwd, removeClosingTag("path", relPath)),
 								content: diff || content,
+								operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
 							}
 
 							if (block.partial) {
 								// update gui message
 								const partialMessage = JSON.stringify(sharedMessageProps)
-								if (this.shouldAutoApproveTool(block.name)) {
+
+								if (this.shouldAutoApproveToolWithPath(block.name, relPath)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool") // in case the user changes auto-approval settings mid stream
 									await this.say("tool", partialMessage, undefined, block.partial)
 								} else {
@@ -1777,6 +1819,7 @@ export class Task {
 								const completeMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: diff || content,
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
 									// ? formatResponse.createPrettyPatch(
 									// 		relPath,
 									// 		this.diffViewProvider.originalContent,
@@ -1784,8 +1827,7 @@ export class Task {
 									// 	)
 									// : undefined,
 								} satisfies ClineSayTool)
-
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, relPath)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", completeMessage, undefined, false)
 									this.consecutiveAutoApprovedRequestsCount++
@@ -1875,7 +1917,7 @@ export class Task {
 								}
 
 								if (!fileExists) {
-									this.controllerRef.deref()?.workspaceTracker?.populateFilePaths()
+									this.workspaceTracker.populateFilePaths()
 								}
 
 								await this.diffViewProvider.reset()
@@ -1903,8 +1945,9 @@ export class Task {
 								const partialMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: undefined,
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", partialMessage, undefined, block.partial)
 								} else {
@@ -1933,8 +1976,9 @@ export class Task {
 								const completeMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: absolutePath,
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", completeMessage, undefined, false) // need to be sending partialValue bool, since undefined has its own purpose in that the message is treated neither as a partial or completion of a partial, but as a single complete message
 									this.consecutiveAutoApprovedRequestsCount++
@@ -1980,8 +2024,9 @@ export class Task {
 								const partialMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: "",
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", partialMessage, undefined, block.partial)
 								} else {
@@ -2011,8 +2056,9 @@ export class Task {
 								const completeMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: result,
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", completeMessage, undefined, false)
 									this.consecutiveAutoApprovedRequestsCount++
@@ -2050,8 +2096,9 @@ export class Task {
 								const partialMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: "",
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", partialMessage, undefined, block.partial)
 								} else {
@@ -2078,8 +2125,9 @@ export class Task {
 								const completeMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: result,
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", completeMessage, undefined, false)
 									this.consecutiveAutoApprovedRequestsCount++
@@ -2121,8 +2169,9 @@ export class Task {
 								const partialMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: "",
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", partialMessage, undefined, block.partial)
 								} else {
@@ -2157,8 +2206,9 @@ export class Task {
 								const completeMessage = JSON.stringify({
 									...sharedMessageProps,
 									content: results,
+									operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
 								} satisfies ClineSayTool)
-								if (this.shouldAutoApproveTool(block.name)) {
+								if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 									this.removeLastPartialMessageIfExistsWithType("ask", "tool")
 									await this.say("tool", completeMessage, undefined, false)
 									this.consecutiveAutoApprovedRequestsCount++
@@ -2265,10 +2315,9 @@ export class Task {
 									await this.say("browser_action_result", "") // starts loading spinner
 
 									// Re-make browserSession to make sure latest settings apply
-									const localContext = this.controllerRef.deref()?.context
-									if (localContext) {
+									if (this.context) {
 										await this.browserSession.dispose()
-										this.browserSession = new BrowserSession(localContext, this.browserSettings)
+										this.browserSession = new BrowserSession(this.context, this.browserSettings)
 									} else {
 										console.warn("no controller context available for browserSession")
 									}
@@ -2466,7 +2515,7 @@ export class Task {
 								}
 
 								// Re-populate file paths in case the command modified the workspace (vscode listeners do not trigger unless the user manually creates/deletes files)
-								this.controllerRef.deref()?.workspaceTracker?.populateFilePaths()
+								this.workspaceTracker.populateFilePaths()
 
 								pushToolResult(result)
 
@@ -2548,9 +2597,8 @@ export class Task {
 									arguments: mcp_arguments,
 								} satisfies ClineAskUseMcpServer)
 
-								const isToolAutoApproved = this.controllerRef
-									.deref()
-									?.mcpHub?.connections?.find((conn) => conn.server.name === server_name)
+								const isToolAutoApproved = this.mcpHub.connections
+									?.find((conn) => conn.server.name === server_name)
 									?.server.tools?.find((tool) => tool.name === tool_name)?.autoApprove
 
 								if (this.shouldAutoApproveTool(block.name) && isToolAutoApproved) {
@@ -2570,9 +2618,7 @@ export class Task {
 
 								// now execute the tool
 								await this.say("mcp_server_request_started") // same as browser_action_result
-								const toolResult = await this.controllerRef
-									.deref()
-									?.mcpHub?.callTool(server_name, tool_name, parsedArguments)
+								const toolResult = await this.mcpHub.callTool(server_name, tool_name, parsedArguments)
 
 								// TODO: add progress indicator and ability to parse images and non-text responses
 								const toolResultPretty =
@@ -2660,7 +2706,7 @@ export class Task {
 
 								// now execute the tool
 								await this.say("mcp_server_request_started")
-								const resourceResult = await this.controllerRef.deref()?.mcpHub?.readResource(server_name, uri)
+								const resourceResult = await this.mcpHub.readResource(server_name, uri)
 								const resourceResultPretty =
 									resourceResult?.contents
 										.map((item) => {
@@ -2743,6 +2789,51 @@ export class Task {
 							break
 						}
 					}
+					case "new_task": {
+						const context: string | undefined = block.params.context
+						try {
+							if (block.partial) {
+								await this.ask("new_task", removeClosingTag("context", context), block.partial).catch(() => {})
+								break
+							} else {
+								if (!context) {
+									this.consecutiveMistakeCount++
+									pushToolResult(await this.sayAndCreateMissingParamError("new_task", "context"))
+									break
+								}
+								this.consecutiveMistakeCount = 0
+
+								if (this.autoApprovalSettings.enabled && this.autoApprovalSettings.enableNotifications) {
+									showSystemNotification({
+										subtitle: "Cline wants to start a new task...",
+										message: `Cline is suggesting to start a new task with: ${context}`,
+									})
+								}
+
+								const { text, images } = await this.ask("new_task", context, false)
+
+								// If the user provided a response, treat it as feedback
+								if (text || images?.length) {
+									await this.say("user_feedback", text ?? "", images)
+									pushToolResult(
+										formatResponse.toolResult(
+											`The user provided feedback instead of creating a new task:\n<feedback>\n${text}\n</feedback>`,
+											images,
+										),
+									)
+								} else {
+									// If no response, the user clicked the "Create New Task" button
+									pushToolResult(
+										formatResponse.toolResult(`The user has created a new task with the provided context.`),
+									)
+								}
+								break
+							}
+						} catch (error) {
+							await handleError("creating new task", error)
+							break
+						}
+					}
 					case "plan_mode_respond": {
 						const response: string | undefined = block.params.response
 						const optionsRaw: string | undefined = block.params.options
@@ -2834,14 +2925,7 @@ export class Task {
 								break
 							} else {
 								await this.say("load_mcp_documentation", "", undefined, false)
-
-								const mcpHub = this.controllerRef.deref()?.mcpHub
-								if (!mcpHub) {
-									throw new Error("MCP hub not available")
-								}
-
-								pushToolResult(await loadMcpDocumentation(mcpHub))
-
+								pushToolResult(await loadMcpDocumentation(this.mcpHub))
 								break
 							}
 						} catch (error) {
@@ -3049,6 +3133,14 @@ export class Task {
 			throw new Error("Cline instance aborted")
 		}
 
+		// Used to know what models were used in the task if user wants to export metadata for error reporting purposes
+		const currentProviderId = (await getGlobalState(this.getContext(), "apiProvider")) as string
+		if (currentProviderId && this.api.getModel().id) {
+			try {
+				await this.modelContextTracker.recordModelUsage(currentProviderId, this.api.getModel().id, this.chatSettings.mode)
+			} catch {}
+		}
+
 		if (this.consecutiveMistakeCount >= 3) {
 			if (this.autoApprovalSettings.enabled && this.autoApprovalSettings.enableNotifications) {
 				showSystemNotification({
@@ -3118,7 +3210,7 @@ export class Task {
 		if (!this.checkpointTracker && !this.checkpointTrackerErrorMessage) {
 			try {
 				this.checkpointTracker = await pTimeout(
-					CheckpointTracker.create(this.taskId, this.controllerRef.deref()?.context.globalStorageUri.fsPath),
+					CheckpointTracker.create(this.taskId, this.context.globalStorageUri.fsPath),
 					{
 						milliseconds: 15_000,
 						message:
@@ -3152,7 +3244,7 @@ export class Task {
 			content: userContent,
 		})
 
-		telemetryService.captureConversationTurnEvent(this.taskId, this.apiProvider, this.api.getModel().id, "user")
+		telemetryService.captureConversationTurnEvent(this.taskId, currentProviderId, this.api.getModel().id, "user")
 
 		// since we sent off a placeholder api_req_started message to update the webview while waiting to actually start the API request (to load potential details for example), we need to update the text of that message
 		const lastApiReqIndex = findLastIndex(this.clineMessages, (m) => m.say === "api_req_started")
@@ -3160,7 +3252,7 @@ export class Task {
 			request: userContent.map((block) => formatContentBlockToMarkdown(block)).join("\n\n"),
 		} satisfies ClineApiReqInfo)
 		await this.saveClineMessagesAndUpdateHistory()
-		await this.controllerRef.deref()?.postStateToWebview()
+		await this.postStateToWebview()
 
 		try {
 			let cacheWriteTokens = 0
@@ -3229,7 +3321,7 @@ export class Task {
 				updateApiReqMsg(cancelReason, streamingFailedMessage)
 				await this.saveClineMessagesAndUpdateHistory()
 
-				telemetryService.captureConversationTurnEvent(this.taskId, this.apiProvider, this.api.getModel().id, "assistant")
+				telemetryService.captureConversationTurnEvent(this.taskId, currentProviderId, this.api.getModel().id, "assistant")
 
 				// signals to provider that it can retrieve the saved messages from disk, as abortTask can not be awaited on in nature
 				this.didFinishAbortingStream = true
@@ -3320,11 +3412,7 @@ export class Task {
 					const errorMessage = this.formatErrorWithStatusCode(error)
 
 					await abortStream("streaming_failed", errorMessage)
-					const history = await this.controllerRef.deref()?.getTaskWithId(this.taskId)
-					if (history) {
-						await this.controllerRef.deref()?.initClineWithHistoryItem(history.historyItem)
-						// await this.controllerRef.deref()?.postStateToWebview()
-					}
+					await this.reinitExistingTaskFromId(this.taskId)
 				}
 			} finally {
 				this.isStreaming = false
@@ -3343,7 +3431,7 @@ export class Task {
 					}
 					updateApiReqMsg()
 					await this.saveClineMessagesAndUpdateHistory()
-					await this.controllerRef.deref()?.postStateToWebview()
+					await this.postStateToWebview()
 				})
 			}
 
@@ -3367,13 +3455,13 @@ export class Task {
 
 			updateApiReqMsg()
 			await this.saveClineMessagesAndUpdateHistory()
-			await this.controllerRef.deref()?.postStateToWebview()
+			await this.postStateToWebview()
 
 			// now add to apiconversationhistory
 			// need to save assistant responses to file before proceeding to tool use since user can exit at any moment and we wouldn't be able to save the assistant's response
 			let didEndLoop = false
 			if (assistantMessage.length > 0) {
-				telemetryService.captureConversationTurnEvent(this.taskId, this.apiProvider, this.api.getModel().id, "assistant")
+				telemetryService.captureConversationTurnEvent(this.taskId, currentProviderId, this.api.getModel().id, "assistant")
 
 				await this.addToApiConversationHistory({
 					role: "assistant",
@@ -3628,6 +3716,36 @@ export class Task {
 			}
 		}
 
+		// Add context window usage information
+		const { contextWindow, maxAllowedSize } = getContextWindowInfo(this.api)
+
+		// Get the token count from the most recent API request to accurately reflect context management
+		const getTotalTokensFromApiReqMessage = (msg: ClineMessage) => {
+			if (!msg.text) {
+				return 0
+			}
+			try {
+				const { tokensIn, tokensOut, cacheWrites, cacheReads } = JSON.parse(msg.text)
+				return (tokensIn || 0) + (tokensOut || 0) + (cacheWrites || 0) + (cacheReads || 0)
+			} catch (e) {
+				return 0
+			}
+		}
+
+		const modifiedMessages = combineApiRequests(combineCommandSequences(this.clineMessages.slice(1)))
+		const lastApiReqMessage = findLast(modifiedMessages, (msg) => {
+			if (msg.say !== "api_req_started") {
+				return false
+			}
+			return getTotalTokensFromApiReqMessage(msg) > 0
+		})
+
+		const lastApiReqTotalTokens = lastApiReqMessage ? getTotalTokensFromApiReqMessage(lastApiReqMessage) : 0
+		const usagePercentage = Math.round((lastApiReqTotalTokens / contextWindow) * 100)
+
+		details += "\n\n# Context Window Usage"
+		details += `\n${lastApiReqTotalTokens.toLocaleString()} / ${(contextWindow / 1000).toLocaleString()}K tokens used (${usagePercentage}%)`
+
 		details += "\n\n# Current Mode"
 		if (this.chatSettings.mode === "plan") {
 			details += "\nPLAN MODE\n" + formatResponse.planModeInstructions()

src/core/webview/index.ts

@@ -23,7 +23,7 @@ export class WebviewProvider implements vscode.WebviewViewProvider {
 		private readonly outputChannel: vscode.OutputChannel,
 	) {
 		WebviewProvider.activeInstances.add(this)
-		this.controller = new Controller(context, outputChannel, this)
+		this.controller = new Controller(context, outputChannel, (message) => this.view?.webview.postMessage(message))
 	}
 
 	async dispose() {

src/extension.ts

@@ -9,6 +9,8 @@ import { DIFF_VIEW_URI_SCHEME } from "./integrations/editor/DiffViewProvider"
 import assert from "node:assert"
 import { telemetryService } from "./services/telemetry/TelemetryService"
 import { WebviewProvider } from "./core/webview"
+import { createTestServer, shutdownTestServer } from "./services/test/TestServer"
+import { ErrorService } from "./services/error/ErrorService"
 
 /*
 Built using https://github.com/microsoft/vscode-webview-ui-toolkit
@@ -27,12 +29,14 @@ export function activate(context: vscode.ExtensionContext) {
 	outputChannel = vscode.window.createOutputChannel("Cline")
 	context.subscriptions.push(outputChannel)
 
+	ErrorService.initialize()
 	Logger.initialize(outputChannel)
 	Logger.log("Cline extension activated")
 
 	const sidebarWebview = new WebviewProvider(context, outputChannel)
 
 	vscode.commands.executeCommand("setContext", "cline.isDevMode", IS_DEV && IS_DEV === "true")
+	vscode.commands.executeCommand("setContext", "cline.isTestMode", IS_TEST && IS_TEST === "true")
 
 	context.subscriptions.push(
 		vscode.window.registerWebviewViewProvider(WebviewProvider.sideBarId, sidebarWebview, {
@@ -377,7 +381,7 @@ export function activate(context: vscode.ExtensionContext) {
 
 	// Register the command handler
 	context.subscriptions.push(
-		vscode.commands.registerCommand("cline.fixWithCline", async (range: vscode.Range, diagnostics: any[]) => {
+		vscode.commands.registerCommand("cline.fixWithCline", async (range: vscode.Range, diagnostics: vscode.Diagnostic[]) => {
 			const editor = vscode.window.activeTextEditor
 			if (!editor) {
 				return
@@ -393,13 +397,12 @@ export function activate(context: vscode.ExtensionContext) {
 		}),
 	)
 
-	return createClineAPI(outputChannel, sidebarWebview.controller)
-}
+	// Set up test server if in test mode
+	if (IS_TEST === "true") {
+		createTestServer(sidebarWebview)
+	}
 
-// This method is called when your extension is deactivated
-export function deactivate() {
-	telemetryService.shutdown()
-	Logger.log("Cline extension deactivated")
+	return createClineAPI(outputChannel, sidebarWebview.controller)
 }
 
 // TODO: Find a solution for automatically removing DEV related content from production builds.
@@ -408,8 +411,18 @@ export function deactivate() {
 //
 // This is a workaround to reload the extension when the source code changes
 // since vscode doesn't support hot reload for extensions
-const { IS_DEV, DEV_WORKSPACE_FOLDER } = process.env
+const { IS_DEV, DEV_WORKSPACE_FOLDER, IS_TEST } = process.env
 
+// This method is called when your extension is deactivated
+export function deactivate() {
+	// Shutdown the test server if it exists
+	shutdownTestServer()
+
+	telemetryService.shutdown()
+	Logger.log("Cline extension deactivated")
+}
+
+// Set up development mode file watcher
 if (IS_DEV && IS_DEV !== "false") {
 	assert(DEV_WORKSPACE_FOLDER, "DEV_WORKSPACE_FOLDER must be set in development")
 	const watcher = vscode.workspace.createFileSystemWatcher(new vscode.RelativePattern(DEV_WORKSPACE_FOLDER, "src/**/*"))

src/integrations/editor/DiffViewProvider.ts

@@ -105,8 +105,11 @@ export class DiffViewProvider {
 		const beginningOfDocument = new vscode.Position(0, 0)
 		diffEditor.selection = new vscode.Selection(beginningOfDocument, beginningOfDocument)
 
-		for (let i = 0; i < diffLines.length; i++) {
-			const currentLine = this.streamedLines.length + i
+		// Instead of animating each line, we'll update in larger chunks
+		const currentLine = this.streamedLines.length + diffLines.length - 1
+		if (currentLine >= 0) {
+			// Only proceed if we have new lines
+
 			// Replace all content up to the current line with accumulated lines
 			// This is necessary (as compared to inserting one line at a time) to handle cases where html tags on previous lines are auto closed for example
 			const edit = new vscode.WorkspaceEdit()
@@ -114,12 +117,33 @@ export class DiffViewProvider {
 			const contentToReplace = accumulatedLines.slice(0, currentLine + 1).join("\n") + "\n"
 			edit.replace(document.uri, rangeToReplace, contentToReplace)
 			await vscode.workspace.applyEdit(edit)
-			// Update decorations
+
+			// Update decorations for the entire changed section
 			this.activeLineController.setActiveLine(currentLine)
 			this.fadedOverlayController.updateOverlayAfterLine(currentLine, document.lineCount)
-			// Scroll to the current line
-			this.scrollEditorToLine(currentLine)
+
+			// Scroll to the last changed line
+			if (diffLines.length <= 5) {
+				// For small changes, just jump directly to the line
+				this.scrollEditorToLine(currentLine)
+			} else {
+				// For larger changes, create a quick scrolling animation
+				const startLine = this.streamedLines.length
+				const endLine = currentLine
+				const totalLines = endLine - startLine
+				const numSteps = 10 // Adjust this number to control animation speed
+				const stepSize = Math.max(1, Math.floor(totalLines / numSteps))
+
+				// Create and await the smooth scrolling animation
+				for (let line = startLine; line <= endLine; line += stepSize) {
+					this.activeDiffEditor?.revealRange(new vscode.Range(line, 0, line, 0), vscode.TextEditorRevealType.InCenter)
+					await new Promise((resolve) => setTimeout(resolve, 16)) // ~60fps
+				}
+				// Ensure we end at the final line
+				this.scrollEditorToLine(currentLine)
+			}
 		}
+
 		// Update the streamedLines with the new accumulated content
 		this.streamedLines = accumulatedLines
 		if (isFinal) {

src/integrations/misc/open-file.ts

@@ -13,7 +13,7 @@ export async function openImage(dataUri: string) {
 	const imageBuffer = Buffer.from(base64Data, "base64")
 	const tempFilePath = path.join(os.tmpdir(), `temp_image_${Date.now()}.${format}`)
 	try {
-		await vscode.workspace.fs.writeFile(vscode.Uri.file(tempFilePath), imageBuffer)
+		await vscode.workspace.fs.writeFile(vscode.Uri.file(tempFilePath), new Uint8Array(imageBuffer))
 		await vscode.commands.executeCommand("vscode.open", vscode.Uri.file(tempFilePath))
 	} catch (error) {
 		vscode.window.showErrorMessage(`Error opening image: ${error}`)

src/integrations/workspace/WorkspaceTracker.ts

@@ -2,17 +2,17 @@ import * as vscode from "vscode"
 import * as path from "path"
 import { listFiles } from "../../services/glob/list-files"
 import { Controller } from "../../core/controller"
+import { ExtensionMessage } from "../../shared/ExtensionMessage"
 
 const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0)
 
 // Note: this is not a drop-in replacement for listFiles at the start of tasks, since that will be done for Desktops when there is no workspace selected
 class WorkspaceTracker {
-	private controllerRef: WeakRef<Controller>
 	private disposables: vscode.Disposable[] = []
 	private filePaths: Set<string> = new Set()
 
-	constructor(controller: Controller) {
-		this.controllerRef = new WeakRef(controller)
+	constructor(private readonly postMessageToWebview: (message: ExtensionMessage) => Promise<void>) {
+		this.postMessageToWebview = postMessageToWebview
 		this.registerListeners()
 	}
 
@@ -85,7 +85,7 @@ class WorkspaceTracker {
 		if (!cwd) {
 			return
 		}
-		this.controllerRef.deref()?.postMessageToWebview({
+		this.postMessageToWebview({
 			type: "workspaceUpdated",
 			filePaths: Array.from(this.filePaths).map((file) => {
 				const relativePath = path.relative(cwd, file).toPosix()

src/services/account/ClineAccountService.ts

@@ -1,26 +1,18 @@
 import axios, { AxiosRequestConfig, AxiosResponse } from "axios"
-import { Controller } from "../../core/controller"
 import type { BalanceResponse, PaymentTransaction, UsageTransaction } from "../../shared/ClineAccount"
+import { ExtensionMessage } from "../../shared/ExtensionMessage"
 
 export class ClineAccountService {
 	private readonly baseUrl = "https://api.cline.bot/v1"
-	private controllerRef: WeakRef<Controller>
+	private postMessageToWebview: (message: ExtensionMessage) => Promise<void>
+	private getClineApiKey: () => Promise<string | undefined>
 
-	constructor(controller: Controller) {
-		this.controllerRef = new WeakRef(controller)
-	}
-
-	/**
-	 * Get the user's Cline Account key from the apiConfiguration
-	 */
-	private async getClineApiKey(): Promise<string | undefined> {
-		const provider = this.controllerRef.deref()
-		if (!provider) {
-			return undefined
-		}
-
-		const { apiConfiguration } = await provider.getStateToPostToWebview()
-		return apiConfiguration?.clineApiKey
+	constructor(
+		postMessageToWebview: (message: ExtensionMessage) => Promise<void>,
+		getClineApiKey: () => Promise<string | undefined>,
+	) {
+		this.postMessageToWebview = postMessageToWebview
+		this.getClineApiKey = getClineApiKey
 	}
 
 	/**
@@ -64,7 +56,7 @@ export class ClineAccountService {
 			const data = await this.authenticatedRequest<BalanceResponse>("/user/credits/balance")
 
 			// Post to webview
-			await this.controllerRef.deref()?.postMessageToWebview({
+			await this.postMessageToWebview({
 				type: "userCreditsBalance",
 				userCreditsBalance: data,
 			})
@@ -84,7 +76,7 @@ export class ClineAccountService {
 			const data = await this.authenticatedRequest<UsageTransaction[]>("/user/credits/usage")
 
 			// Post to webview
-			await this.controllerRef.deref()?.postMessageToWebview({
+			await this.postMessageToWebview({
 				type: "userCreditsUsage",
 				userCreditsUsage: data,
 			})
@@ -104,7 +96,7 @@ export class ClineAccountService {
 			const data = await this.authenticatedRequest<PaymentTransaction[]>("/user/credits/payments")
 
 			// Post to webview
-			await this.controllerRef.deref()?.postMessageToWebview({
+			await this.postMessageToWebview({
 				type: "userCreditsPayments",
 				userCreditsPayments: data,
 			})

src/services/browser/BrowserSession.ts

@@ -2,7 +2,8 @@ import * as vscode from "vscode"
 import * as fs from "fs/promises"
 import * as path from "path"
 import { exec, spawn } from "child_process"
-import { Browser, Page, ScreenshotOptions, TimeoutError, launch, connect } from "puppeteer-core"
+import { Browser, Page, TimeoutError, launch, connect } from "puppeteer-core"
+import type { ScreenshotOptions, ConsoleMessage } from "puppeteer-core"
 // @ts-ignore
 import PCR from "puppeteer-chromium-resolver"
 import pWaitFor from "p-wait-for"
@@ -451,7 +452,7 @@ export class BrowserSession {
 		const logs: string[] = []
 		let lastLogTs = Date.now()
 
-		const consoleListener = (msg: any) => {
+		const consoleListener = (msg: ConsoleMessage) => {
 			if (msg.type() === "log") {
 				logs.push(msg.text())
 			} else {

src/services/error/ErrorService.ts

@@ -0,0 +1,96 @@
+import * as Sentry from "@sentry/browser"
+import * as vscode from "vscode"
+import { telemetryService } from "../telemetry/TelemetryService"
+import * as pkg from "../../../package.json"
+
+let telemetryLevel = vscode.workspace.getConfiguration("telemetry").get<string>("telemetryLevel", "all")
+let isTelemetryEnabled = ["all", "error"].includes(telemetryLevel)
+
+vscode.workspace.onDidChangeConfiguration(() => {
+	telemetryLevel = vscode.workspace.getConfiguration("telemetry").get<string>("telemetryLevel", "all")
+	isTelemetryEnabled = ["all", "error"].includes(telemetryLevel)
+	ErrorService.toggleEnabled(isTelemetryEnabled)
+	if (isTelemetryEnabled) {
+		ErrorService.setLevel(telemetryLevel as "error" | "all")
+	}
+})
+
+export class ErrorService {
+	private static serviceEnabled: boolean
+	private static serviceLevel: string
+
+	static initialize() {
+		// Initialize sentry
+		Sentry.init({
+			dsn: "https://7936780e3f0f0290fcf8d4a395c249b7@o4509028819664896.ingest.us.sentry.io/4509052955983872",
+			environment: process.env.NODE_ENV,
+			release: `cline@${pkg.version}`,
+			integrations: [Sentry.browserTracingIntegration(), Sentry.replayIntegration()],
+			beforeSend(event) {
+				// TelemetryService keeps track of whether the user has opted in to telemetry/error reporting
+				const isUserManuallyOptedIn = telemetryService.isTelemetryEnabled()
+				if (isUserManuallyOptedIn && ErrorService.isEnabled()) {
+					return event
+				}
+				return null
+			},
+		})
+
+		ErrorService.toggleEnabled(true)
+		ErrorService.setLevel("error")
+	}
+
+	static toggleEnabled(state: boolean) {
+		if (state === false) {
+			ErrorService.serviceEnabled = false
+			return
+		}
+		// If we are trying to enable the service, check that we are allowed to.
+		if (isTelemetryEnabled) {
+			ErrorService.serviceEnabled = true
+		}
+	}
+
+	static setLevel(level: "error" | "all") {
+		switch (telemetryLevel) {
+			case "error": {
+				if (level === "error") {
+					ErrorService.serviceLevel = level
+				}
+				break
+			}
+			default: {
+				ErrorService.serviceLevel = level
+			}
+		}
+	}
+
+	static logException(error: Error): void {
+		// Don't log if telemetry is off
+		const isUserManuallyOptedIn = telemetryService.isTelemetryEnabled()
+		if (!isUserManuallyOptedIn || !ErrorService.isEnabled()) {
+			return
+		}
+		// Log the error to Sentry
+		Sentry.captureException(error)
+	}
+
+	static logMessage(message: string, level: "error" | "warning" | "log" | "debug" | "info" = "log"): void {
+		// Don't log if telemetry is off
+		const isUserManuallyOptedIn = telemetryService.isTelemetryEnabled()
+		if (!isUserManuallyOptedIn || !ErrorService.isEnabled()) {
+			return
+		}
+		if (ErrorService.serviceLevel === "error" && level === "error") {
+			// Log the message if allowed
+			Sentry.captureMessage(message, { level })
+			return
+		}
+		// Log the message if allowed
+		Sentry.captureMessage(message, { level })
+	}
+
+	static isEnabled(): boolean {
+		return ErrorService.serviceEnabled
+	}
+}

src/services/logging/Logger.ts

@@ -1,4 +1,5 @@
 import type { OutputChannel } from "vscode"
+import { ErrorService } from "../error/ErrorService"
 
 /**
  * Simple logging utility for the extension's backend code.
@@ -12,7 +13,25 @@ export class Logger {
 		Logger.outputChannel = outputChannel
 	}
 
+	static error(message: string, exception?: Error) {
+		Logger.outputChannel.appendLine(`ERROR: ${message}`)
+		ErrorService.logMessage(message, "error")
+		exception && ErrorService.logException(exception)
+	}
+	static warn(message: string) {
+		Logger.outputChannel.appendLine(`WARN: ${message}`)
+		ErrorService.logMessage(message, "warning")
+	}
 	static log(message: string) {
-		Logger.outputChannel.appendLine(message)
+		Logger.outputChannel.appendLine(`LOG: ${message}`)
+	}
+	static debug(message: string) {
+		Logger.outputChannel.appendLine(`DEBUG: ${message}`)
+	}
+	static info(message: string) {
+		Logger.outputChannel.appendLine(`INFO: ${message}`)
+	}
+	static trace(message: string) {
+		Logger.outputChannel.appendLine(`TRACE: ${message}`)
 	}
 }