Fix: Display original data:image URLs in rich display mode to maintain transparency

* initial terminal output chunking

* changeset

* Update src/core/task/index.ts

Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com>

* cleanup

---------

Co-authored-by: ellipsis-dev[bot] <65095814+ellipsis-dev[bot]@users.noreply.github.com>

.changeset/fast-readers-care.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": minor
+---
+
+Added chunking to terminal outputs

.changeset/khaki-cooks-walk.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-update ordering of messages during task restore

.changeset/late-mice-double.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+gRPC over vscode message bus to make messaging better

.changeset/nasty-gifts-look.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Allow user to modify a previous message to continue chat

.changeset/nasty-months-wave.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Add command to focus chat input

.changeset/pretty-eels-breathe.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+fix vertexai token count

.changeset/rare-starfishes-design.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Allow user to send context with an option selection

.changeset/rich-pens-buy.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Safer check for xAI reasoning content

.changeset/six-birds-move.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": minor
+---
+
+Add an edit button to the Cline Rules toggle modal to open rule files

.changeset/slimy-jeans-dance.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+MCP ImageContent as tool result

.changeset/ten-bananas-begin.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Readme update

.changeset/twelve-eyes-teach.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Add OpenAI o3 & 4o-mini

.changeset/twelve-meals-applaud.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": minor
+---
+
+Add modal UI for toggling Cline Rules

.codespellrc

@@ -0,0 +1,6 @@
+[codespell]
+# Ref: https://github.com/codespell-project/codespell#using-a-config-file
+skip = .git*,*.svg,package-lock.json,*.css,.codespellrc,locales
+check-hidden = true
+ignore-regex = (\b(optIn|isTaller)\b|https://\S+)
+# ignore-words-list =

.github/pull_request_template.md

@@ -13,7 +13,10 @@
 -   [ ] 🐛 Bug fix (non-breaking change which fixes an issue)
 -   [ ] ✨ New feature (non-breaking change which adds functionality)
 -   [ ] 💥 Breaking change (fix or feature that would cause existing functionality to not work as expected)
+-   [ ] ♻️ Refactor Changes
+-   [ ] 💅 Cosmetic Changes
 -   [ ] 📚 Documentation update
+-   [ ] 🏃 Workflow Changes
 
 ### Pre-flight Checklist
 

.github/workflows/changeset-converter.yml

@@ -2,6 +2,7 @@ name: Changeset Converter
 run-name: Changeset Conversion
 
 on:
+    workflow_dispatch:
     pull_request:
         types: [closed]
 
@@ -13,16 +14,36 @@ env:
 jobs:
     # Job 1: Create version bump PR when changesets are merged to main
     changeset-pr-version-bump:
-        if: >
-            github.event_name == 'pull_request' &&
-            github.event.pull_request.merged == true &&
-            github.event.pull_request.base.ref == 'main' &&
-            github.actor != 'github-actions'
+        if: |
+            github.event_name == 'workflow_dispatch' ||
+            (
+                github.event_name == 'pull_request' &&
+                github.event.pull_request.merged == true &&
+                github.event.pull_request.base.ref == 'main' &&
+                github.actor != 'github-actions'
+            )
         runs-on: ubuntu-latest
         permissions:
             contents: write
             pull-requests: write
         steps:
+            - name: Check user for team affiliation
+              id: team_check
+              if: github.event_name == 'workflow_dispatch'
+              uses: morfien101/actions-authorized-user@4a3cfbf0bcb3cafe4a71710a278920c5d94bb38b
+              with:
+                  username: ${{ github.actor }}
+                  team: "deployer"
+                  github_token: ${{ secrets.GITHUB_TOKEN }}
+
+            - name: Check if user is authorized
+              if: github.event_name == 'workflow_dispatch'
+              run: |
+                  if [ "${{ steps.team_check.outputs.authorized }}" != "true" ]; then
+                      echo "User is not authorized to run this workflow."
+                      exit 1
+                  fi
+
             - name: Git Checkout
               uses: actions/checkout@v4
               with:

.github/workflows/codespell.yml

@@ -0,0 +1,25 @@
+# Codespell configuration is within .codespellrc
+---
+name: Codespell
+
+on:
+    push:
+        branches: [main]
+    pull_request:
+        branches: [main]
+
+permissions:
+    contents: read
+
+jobs:
+    codespell:
+        name: Check for spelling errors
+        runs-on: ubuntu-latest
+
+        steps:
+            - name: Checkout
+              uses: actions/checkout@v4
+            - name: Annotate locations with typos
+              uses: codespell-project/codespell-problem-matcher@v1
+            - name: Codespell
+              uses: codespell-project/actions-codespell@v2

.github/workflows/test.yml

@@ -73,8 +73,9 @@ jobs:
             - name: Build Extension
               run: npm run compile
 
-            - name: Unit Tests
-              run: npm run test:unit
+            # Disabling due to compatability with test framework and ESM modules
+            # - name: Unit Tests
+            #   run: npm run test:unit
 
             # Run extension tests with coverage
             - name: Extension Tests with Coverage

.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

@@ -1,11 +1,64 @@
 # 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
 -   Add ability to detect when user edits files manually so Cline knows to re-read, leading to reduced diff edit errors
 -   Add improvements to file mention searching for faster searching
--   Add scoring logic to file mentions to sort and exlcude results based on relevance
+-   Add scoring logic to file mentions to sort and exclude results based on relevance
 -   Add Support for Bytedance Doubao (Thanks Tunixer!)
 -   Fix to prevent duplicate BOM (Thanks bamps53!)
 
@@ -483,7 +536,7 @@
 -   Adds "Always allow read-only operations" setting to let Claude read files and view directories without needing approval (off by default)
 -   Implement sliding window context management to keep tasks going past 200k tokens
 -   Adds Google Cloud Vertex AI support and updates Claude 3.5 Sonnet max output to 8192 tokens for all providers.
--   Improves system prompt to gaurd against lazy edits (less "//rest of code here")
+-   Improves system prompt to guard against lazy edits (less "//rest of code here")
 
 ## [1.3.0]
 

docs/tools/cline-tools-guide.md

@@ -55,7 +55,7 @@ Cline is your AI assistant that can:
 
 ## Available Tools
 
-For the most up-to-date implementation details, you can view the full source code in the [Cline repository](https://github.com/cline/cline/blob/main/src/core/Cline.ts).
+For the most up-to-date implementation details, you can view the full source code in the [Cline repository](https://github.com/cline/cline/blob/main/src/core/task/index.ts).
 
 Cline has access to the following tools for various tasks:
 
@@ -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
+# Cline
 
 <p align="center">
     <img src="https://media.githubusercontent.com/media/cline/cline/main/assets/docs/demo.gif" width="100%" />
@@ -20,7 +20,7 @@
 <a href="https://github.com/cline/cline/discussions/categories/feature-requests?discussions_q=is%3Aopen+category%3A%22Feature+Requests%22+sort%3Atop" target="_blank"><strong>功能请求</strong></a>
 </td>
 <td align="center">
-<a href="https://cline.bot/join-us" target="_blank"><strong>我们正在招聘！</strong></a>
+<a href="https://docs.cline.bot/getting-started/for-new-coders" target="_blank"><strong>新手上路</strong></a>
 </td>
 </tbody>
 </table>
@@ -33,12 +33,12 @@
 1. 输入你的任务并添加图像，将模型转换为功能应用程序或通过截图修复错误。
 2. Cline 首先分析你的文件结构和源代码 AST，运行正则表达式搜索，并阅读相关文件以了解现有项目。通过仔细管理添加到上下文中的信息，Cline 即使在大型复杂项目中也能提供有价值的帮助，而不会使上下文窗口过载。
 3. 一旦 Cline 获得所需信息，他可以：
-        - 创建和编辑文件 + 监控 linter/编译器错误，从而主动修复诸如缺少导入和语法错误等问题。
-        - 直接在你的终端中执行命令并监控其输出，从而在编辑文件后对开发服务器问题做出反应。
-        - 对于 Web 开发任务，Cline 可以在无头浏览器中启动网站，点击、输入、滚动并捕获截图和控制台日志，从而修复运行时错误和视觉错误。
+   - 创建和编辑文件 + 监控 linter/编译器错误，从而主动修复诸如缺少导入和语法错误等问题。
+   - 直接在你的终端中执行命令并监控其输出，从而在编辑文件后对开发服务器问题做出反应。
+   - 对于 Web 开发任务，Cline 可以在无头浏览器中启动网站，点击、输入、滚动并捕获截图和控制台日志，从而修复运行时错误和视觉错误。
 4. 当任务完成时，Cline 将通过终端命令如 `open -a "Google Chrome" index.html` 向你展示结果，你可以通过点击按钮运行该命令。
 
-> [!提示]
+> [!TIP]
 > 使用 `CMD/CTRL + Shift + P` 快捷键打开命令面板并输入 "Cline: Open In New Tab" 将扩展作为标签在编辑器中打开。这让你可以与文件资源管理器并排使用 Cline，更清楚地看到他如何改变你的工作空间。
 
 ---

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

@@ -1,20 +1,22 @@
 {
 	"name": "claude-dev",
-	"version": "3.9.1",
+	"version": "3.12.3",
 	"lockfileVersion": 3,
 	"requires": true,
 	"packages": {
 		"": {
 			"name": "claude-dev",
-			"version": "3.9.1",
+			"version": "3.12.3",
 			"license": "Apache-2.0",
 			"dependencies": {
 				"@anthropic-ai/bedrock-sdk": "^0.12.4",
 				"@anthropic-ai/sdk": "^0.37.0",
 				"@anthropic-ai/vertex-sdk": "^0.6.4",
 				"@aws-sdk/client-bedrock-runtime": "^3.758.0",
+				"@bufbuild/protobuf": "^2.2.5",
 				"@google-cloud/vertexai": "^1.9.3",
 				"@google/generative-ai": "^0.18.0",
+				"@grpc/grpc-js": "^1.9.15",
 				"@mistralai/mistralai": "^1.5.0",
 				"@modelcontextprotocol/sdk": "^1.7.0",
 				"@opentelemetry/api": "^1.4.1",
@@ -23,10 +25,12 @@
 				"@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",
 				"chokidar": "^4.0.1",
+				"chrome-launcher": "^1.1.2",
 				"clone-deep": "^4.0.1",
 				"default-shell": "^2.2.0",
 				"diff": "^5.2.0",
@@ -77,15 +81,18 @@
 				"@vscode/test-cli": "^0.0.9",
 				"@vscode/test-electron": "^2.4.0",
 				"chai": "^4.3.10",
+				"chalk": "^5.3.0",
 				"esbuild": "^0.25.0",
 				"eslint": "^8.57.0",
 				"husky": "^9.1.7",
 				"npm-run-all": "^4.1.5",
 				"prettier": "^3.3.3",
+				"protoc-gen-ts": "^0.8.7",
 				"proxyquire": "^2.1.3",
 				"should": "^13.2.3",
 				"sinon": "^19.0.2",
 				"ts-node": "^10.9.2",
+				"ts-proto": "^2.6.1",
 				"typescript": "^5.4.5"
 			},
 			"engines": {
@@ -3955,6 +3962,12 @@
 			"dev": true,
 			"license": "MIT"
 		},
+		"node_modules/@bufbuild/protobuf": {
+			"version": "2.2.5",
+			"resolved": "https://registry.npmjs.org/@bufbuild/protobuf/-/protobuf-2.2.5.tgz",
+			"integrity": "sha512-/g5EzJifw5GF8aren8wZ/G5oMuPoGeS6MQD3ca8ddcvdXR5UELUfdTZITCGNhNXynY/AYl3Z4plmxdj/tRl/hQ==",
+			"license": "(Apache-2.0 AND BSD-3-Clause)"
+		},
 		"node_modules/@changesets/apply-release-plan": {
 			"version": "7.0.8",
 			"resolved": "https://registry.npmjs.org/@changesets/apply-release-plan/-/apply-release-plan-7.0.8.tgz",
@@ -5564,6 +5577,7 @@
 			"version": "1.9.15",
 			"resolved": "https://registry.npmjs.org/@grpc/grpc-js/-/grpc-js-1.9.15.tgz",
 			"integrity": "sha512-nqE7Hc0AzI+euzUwDAy0aY5hCp10r734gMGRdU+qOPX0XSceI2ULrcXB5U2xSc5VkWwalCj4M7GzCAygZl2KoQ==",
+			"license": "Apache-2.0",
 			"dependencies": {
 				"@grpc/proto-loader": "^0.7.8",
 				"@types/node": ">=12.12.47"
@@ -6855,6 +6869,81 @@
 			"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",
@@ -9842,6 +9931,19 @@
 				"url": "https://github.com/sponsors/sindresorhus"
 			}
 		},
+		"node_modules/case-anything": {
+			"version": "2.1.13",
+			"resolved": "https://registry.npmjs.org/case-anything/-/case-anything-2.1.13.tgz",
+			"integrity": "sha512-zlOQ80VrQ2Ue+ymH5OuM/DlDq64mEm+B9UTdHULv5osUMD6HalNTblf2b1u/m6QecjsnOkBpqVZ+XPwIVsy7Ng==",
+			"dev": true,
+			"license": "MIT",
+			"engines": {
+				"node": ">=12.13"
+			},
+			"funding": {
+				"url": "https://github.com/sponsors/mesqueeb"
+			}
+		},
 		"node_modules/chai": {
 			"version": "4.3.10",
 			"resolved": "https://registry.npmjs.org/chai/-/chai-4.3.10.tgz",
@@ -9861,35 +9963,18 @@
 			}
 		},
 		"node_modules/chalk": {
-			"version": "4.1.2",
-			"resolved": "https://registry.npmjs.org/chalk/-/chalk-4.1.2.tgz",
-			"integrity": "sha512-oKnbhFyRIXpUuez8iBMmyEa4nbj4IOQyuhc/wy9kY7/WVPcwIO9VA668Pu8RkO7+0G76SLROeyw9CpQ061i4mA==",
+			"version": "5.4.1",
+			"resolved": "https://registry.npmjs.org/chalk/-/chalk-5.4.1.tgz",
+			"integrity": "sha512-zgVZuo2WcZgfUEmsn6eO3kINexW8RAE4maiQ8QNs8CtpPCSyMiYsULR3HQYkm3w8FIA3SberyMJMSldGsW+U3w==",
 			"dev": true,
 			"license": "MIT",
-			"dependencies": {
-				"ansi-styles": "^4.1.0",
-				"supports-color": "^7.1.0"
-			},
 			"engines": {
-				"node": ">=10"
+				"node": "^12.17.0 || ^14.13 || >=16.0.0"
 			},
 			"funding": {
 				"url": "https://github.com/chalk/chalk?sponsor=1"
 			}
 		},
-		"node_modules/chalk/node_modules/supports-color": {
-			"version": "7.2.0",
-			"resolved": "https://registry.npmjs.org/supports-color/-/supports-color-7.2.0.tgz",
-			"integrity": "sha512-qpCAvRl9stuOHveKsn7HncJRvv501qIacKzQlO/+Lwxc9+0q2wLyv4Dfvt80/DPn2pqOBsJdDiogXGR9+OvwRw==",
-			"dev": true,
-			"license": "MIT",
-			"dependencies": {
-				"has-flag": "^4.0.0"
-			},
-			"engines": {
-				"node": ">=8"
-			}
-		},
 		"node_modules/chardet": {
 			"version": "0.7.0",
 			"resolved": "https://registry.npmjs.org/chardet/-/chardet-0.7.0.tgz",
@@ -9966,6 +10051,24 @@
 				"url": "https://paulmillr.com/funding/"
 			}
 		},
+		"node_modules/chrome-launcher": {
+			"version": "1.1.2",
+			"resolved": "https://registry.npmjs.org/chrome-launcher/-/chrome-launcher-1.1.2.tgz",
+			"integrity": "sha512-YclTJey34KUm5jB1aEJCq807bSievi7Nb/TU4Gu504fUYi3jw3KCIaH6L7nFWQhdEgH3V+wCh+kKD1P5cXnfxw==",
+			"license": "Apache-2.0",
+			"dependencies": {
+				"@types/node": "*",
+				"escape-string-regexp": "^4.0.0",
+				"is-wsl": "^2.2.0",
+				"lighthouse-logger": "^2.0.1"
+			},
+			"bin": {
+				"print-chrome-path": "bin/print-chrome-path.js"
+			},
+			"engines": {
+				"node": ">=12.13.0"
+			}
+		},
 		"node_modules/chromium-bidi": {
 			"version": "0.6.5",
 			"resolved": "https://registry.npmjs.org/chromium-bidi/-/chromium-bidi-0.6.5.tgz",
@@ -10517,6 +10620,19 @@
 				"node": ">=8"
 			}
 		},
+		"node_modules/detect-libc": {
+			"version": "1.0.3",
+			"resolved": "https://registry.npmjs.org/detect-libc/-/detect-libc-1.0.3.tgz",
+			"integrity": "sha512-pGjwhsmsp4kL2RTz08wcOlGN83otlqHeD/Z5T8GXZB+/YcpQ/dgo+lbU8ZsGxV0HIvqqxo9l7mqYwyYMD9bKDg==",
+			"dev": true,
+			"license": "Apache-2.0",
+			"bin": {
+				"detect-libc": "bin/detect-libc.js"
+			},
+			"engines": {
+				"node": ">=0.10"
+			}
+		},
 		"node_modules/devtools-protocol": {
 			"version": "0.0.1342118",
 			"resolved": "https://registry.npmjs.org/devtools-protocol/-/devtools-protocol-0.0.1342118.tgz",
@@ -10627,6 +10743,16 @@
 				"url": "https://github.com/fb55/domutils?sponsor=1"
 			}
 		},
+		"node_modules/dprint-node": {
+			"version": "1.0.8",
+			"resolved": "https://registry.npmjs.org/dprint-node/-/dprint-node-1.0.8.tgz",
+			"integrity": "sha512-iVKnUtYfGrYcW1ZAlfR/F59cUVL8QIhWoBJoSjkkdua/dkWIgjZfiLMeTjiB06X0ZLkQ0M2C1VbUj/CxkIf1zg==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"detect-libc": "^1.0.3"
+			}
+		},
 		"node_modules/duck": {
 			"version": "0.1.12",
 			"resolved": "https://registry.npmjs.org/duck/-/duck-0.1.12.tgz",
@@ -10971,7 +11097,6 @@
 			"version": "4.0.0",
 			"resolved": "https://registry.npmjs.org/escape-string-regexp/-/escape-string-regexp-4.0.0.tgz",
 			"integrity": "sha512-TtpcNJ3XAzx3Gq8sWRzJaVajRs0uVxA2YAkdb1jm2YkPz4G6egUFAyA3n5vtEIZefPk5Wa4UXbKuS5fKkJWdgA==",
-			"dev": true,
 			"license": "MIT",
 			"engines": {
 				"node": ">=10"
@@ -11098,6 +11223,23 @@
 				"concat-map": "0.0.1"
 			}
 		},
+		"node_modules/eslint/node_modules/chalk": {
+			"version": "4.1.2",
+			"resolved": "https://registry.npmjs.org/chalk/-/chalk-4.1.2.tgz",
+			"integrity": "sha512-oKnbhFyRIXpUuez8iBMmyEa4nbj4IOQyuhc/wy9kY7/WVPcwIO9VA668Pu8RkO7+0G76SLROeyw9CpQ061i4mA==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"ansi-styles": "^4.1.0",
+				"supports-color": "^7.1.0"
+			},
+			"engines": {
+				"node": ">=10"
+			},
+			"funding": {
+				"url": "https://github.com/chalk/chalk?sponsor=1"
+			}
+		},
 		"node_modules/eslint/node_modules/glob-parent": {
 			"version": "6.0.2",
 			"resolved": "https://registry.npmjs.org/glob-parent/-/glob-parent-6.0.2.tgz",
@@ -11146,6 +11288,19 @@
 				"node": ">=8"
 			}
 		},
+		"node_modules/eslint/node_modules/supports-color": {
+			"version": "7.2.0",
+			"resolved": "https://registry.npmjs.org/supports-color/-/supports-color-7.2.0.tgz",
+			"integrity": "sha512-qpCAvRl9stuOHveKsn7HncJRvv501qIacKzQlO/+Lwxc9+0q2wLyv4Dfvt80/DPn2pqOBsJdDiogXGR9+OvwRw==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"has-flag": "^4.0.0"
+			},
+			"engines": {
+				"node": ">=8"
+			}
+		},
 		"node_modules/espree": {
 			"version": "9.6.1",
 			"resolved": "https://registry.npmjs.org/espree/-/espree-9.6.1.tgz",
@@ -12897,6 +13052,21 @@
 				"url": "https://github.com/sponsors/ljharb"
 			}
 		},
+		"node_modules/is-docker": {
+			"version": "2.2.1",
+			"resolved": "https://registry.npmjs.org/is-docker/-/is-docker-2.2.1.tgz",
+			"integrity": "sha512-F+i2BKsFrH66iaUFc0woD8sLy8getkwTwtOBjvs56Cx4CgJDeKQeqfz8wAYiSb8JOprWhHH5p77PbmYCvvUuXQ==",
+			"license": "MIT",
+			"bin": {
+				"is-docker": "cli.js"
+			},
+			"engines": {
+				"node": ">=8"
+			},
+			"funding": {
+				"url": "https://github.com/sponsors/sindresorhus"
+			}
+		},
 		"node_modules/is-extglob": {
 			"version": "2.1.1",
 			"resolved": "https://registry.npmjs.org/is-extglob/-/is-extglob-2.1.1.tgz",
@@ -13167,6 +13337,18 @@
 				"node": ">=0.10.0"
 			}
 		},
+		"node_modules/is-wsl": {
+			"version": "2.2.0",
+			"resolved": "https://registry.npmjs.org/is-wsl/-/is-wsl-2.2.0.tgz",
+			"integrity": "sha512-fKzAra0rGJUUBwGBgNkHZuToZcn+TtXHpeCgmkMJMMYx1sQDYaCSyjJBSCa2nH1DGm7s3n1oBnohoVTBaN7Lww==",
+			"license": "MIT",
+			"dependencies": {
+				"is-docker": "^2.0.0"
+			},
+			"engines": {
+				"node": ">=8"
+			}
+		},
 		"node_modules/isarray": {
 			"version": "1.0.0",
 			"resolved": "https://registry.npmjs.org/isarray/-/isarray-1.0.0.tgz",
@@ -13438,6 +13620,31 @@
 				"immediate": "~3.0.5"
 			}
 		},
+		"node_modules/lighthouse-logger": {
+			"version": "2.0.1",
+			"resolved": "https://registry.npmjs.org/lighthouse-logger/-/lighthouse-logger-2.0.1.tgz",
+			"integrity": "sha512-ioBrW3s2i97noEmnXxmUq7cjIcVRjT5HBpAYy8zE11CxU9HqlWHHeRxfeN1tn8F7OEMVPIC9x1f8t3Z7US9ehQ==",
+			"license": "Apache-2.0",
+			"dependencies": {
+				"debug": "^2.6.9",
+				"marky": "^1.2.2"
+			}
+		},
+		"node_modules/lighthouse-logger/node_modules/debug": {
+			"version": "2.6.9",
+			"resolved": "https://registry.npmjs.org/debug/-/debug-2.6.9.tgz",
+			"integrity": "sha512-bC7ElrdJaJnPbAP+1EotYvqZsb3ecl5wi6Bfi6BJTUcNowp6cvspg0jXznRTKDjm/E7AdgFBVeAPVMNcKGsHMA==",
+			"license": "MIT",
+			"dependencies": {
+				"ms": "2.0.0"
+			}
+		},
+		"node_modules/lighthouse-logger/node_modules/ms": {
+			"version": "2.0.0",
+			"resolved": "https://registry.npmjs.org/ms/-/ms-2.0.0.tgz",
+			"integrity": "sha512-Tpp60P6IUJDTuOq/5Z8cdskzJujfwqfOTkrwIwj7IRISpnkJnT6SyJ4PCPnGMoFjC9ddhal5KVIYtAt97ix05A==",
+			"license": "MIT"
+		},
 		"node_modules/load-json-file": {
 			"version": "4.0.0",
 			"resolved": "https://registry.npmjs.org/load-json-file/-/load-json-file-4.0.0.tgz",
@@ -13513,6 +13720,36 @@
 				"url": "https://github.com/sponsors/sindresorhus"
 			}
 		},
+		"node_modules/log-symbols/node_modules/chalk": {
+			"version": "4.1.2",
+			"resolved": "https://registry.npmjs.org/chalk/-/chalk-4.1.2.tgz",
+			"integrity": "sha512-oKnbhFyRIXpUuez8iBMmyEa4nbj4IOQyuhc/wy9kY7/WVPcwIO9VA668Pu8RkO7+0G76SLROeyw9CpQ061i4mA==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"ansi-styles": "^4.1.0",
+				"supports-color": "^7.1.0"
+			},
+			"engines": {
+				"node": ">=10"
+			},
+			"funding": {
+				"url": "https://github.com/chalk/chalk?sponsor=1"
+			}
+		},
+		"node_modules/log-symbols/node_modules/supports-color": {
+			"version": "7.2.0",
+			"resolved": "https://registry.npmjs.org/supports-color/-/supports-color-7.2.0.tgz",
+			"integrity": "sha512-qpCAvRl9stuOHveKsn7HncJRvv501qIacKzQlO/+Lwxc9+0q2wLyv4Dfvt80/DPn2pqOBsJdDiogXGR9+OvwRw==",
+			"dev": true,
+			"license": "MIT",
+			"dependencies": {
+				"has-flag": "^4.0.0"
+			},
+			"engines": {
+				"node": ">=8"
+			}
+		},
 		"node_modules/long": {
 			"version": "5.2.4",
 			"resolved": "https://registry.npmjs.org/long/-/long-5.2.4.tgz",
@@ -13616,6 +13853,12 @@
 				"sprintf-js": "~1.0.2"
 			}
 		},
+		"node_modules/marky": {
+			"version": "1.2.5",
+			"resolved": "https://registry.npmjs.org/marky/-/marky-1.2.5.tgz",
+			"integrity": "sha512-q9JtQJKjpsVxCRVgQ+WapguSbKC3SQ5HEzFGPAJMStgh3QjCawp00UKv3MTTAArTmGmmPUvllHZoNbZ3gs0I+Q==",
+			"license": "Apache-2.0"
+		},
 		"node_modules/math-intrinsics": {
 			"version": "1.1.0",
 			"resolved": "https://registry.npmjs.org/math-intrinsics/-/math-intrinsics-1.1.0.tgz",
@@ -14616,19 +14859,6 @@
 				"url": "https://github.com/sponsors/sindresorhus"
 			}
 		},
-		"node_modules/ora/node_modules/chalk": {
-			"version": "5.3.0",
-			"resolved": "https://registry.npmjs.org/chalk/-/chalk-5.3.0.tgz",
-			"integrity": "sha512-dLitG79d+GV1Nb/VYcCDFivJeK1hiukt9QjRNVOsUtTy1rR1YJsmpGGTZ3qJos+uw7WmWF4wUwBd9jxjocFC2w==",
-			"dev": true,
-			"license": "MIT",
-			"engines": {
-				"node": "^12.17.0 || ^14.13 || >=16.0.0"
-			},
-			"funding": {
-				"url": "https://github.com/chalk/chalk?sponsor=1"
-			}
-		},
 		"node_modules/ora/node_modules/emoji-regex": {
 			"version": "10.3.0",
 			"resolved": "https://registry.npmjs.org/emoji-regex/-/emoji-regex-10.3.0.tgz",
@@ -15214,6 +15444,20 @@
 				"node": ">=12.0.0"
 			}
 		},
+		"node_modules/protoc-gen-ts": {
+			"version": "0.8.7",
+			"resolved": "https://registry.npmjs.org/protoc-gen-ts/-/protoc-gen-ts-0.8.7.tgz",
+			"integrity": "sha512-jr4VJey2J9LVYCV7EVyVe53g1VMw28cCmYJhBe5e3YX5wiyiDwgxWxeDf9oTqAe4P1bN/YGAkW2jhlH8LohwiQ==",
+			"dev": true,
+			"license": "MIT",
+			"bin": {
+				"protoc-gen-ts": "protoc-gen-ts.js"
+			},
+			"funding": {
+				"type": "individual",
+				"url": "https://www.buymeacoffee.com/thesayyn"
+			}
+		},
 		"node_modules/proxy-addr": {
 			"version": "2.0.7",
 			"resolved": "https://registry.npmjs.org/proxy-addr/-/proxy-addr-2.0.7.tgz",
@@ -16818,6 +17062,42 @@
 				"node": ">=0.3.1"
 			}
 		},
+		"node_modules/ts-poet": {
+			"version": "6.11.0",
+			"resolved": "https://registry.npmjs.org/ts-poet/-/ts-poet-6.11.0.tgz",
+			"integrity": "sha512-r5AGF8vvb+GjBsnqiTqbLhN1/U2FJt6BI+k0dfCrkKzWvUhNlwMmq9nDHuucHs45LomgHjZPvYj96dD3JawjJA==",
+			"dev": true,
+			"license": "Apache-2.0",
+			"dependencies": {
+				"dprint-node": "^1.0.8"
+			}
+		},
+		"node_modules/ts-proto": {
+			"version": "2.7.0",
+			"resolved": "https://registry.npmjs.org/ts-proto/-/ts-proto-2.7.0.tgz",
+			"integrity": "sha512-BGHjse2wTOeswOqnnPKinpxmbaRd882so/e1En6ww59YMG7AO9Kg4vPpJcbVfrpBixPRDqHafXD/RDyd2T99GA==",
+			"dev": true,
+			"license": "ISC",
+			"dependencies": {
+				"@bufbuild/protobuf": "^2.0.0",
+				"case-anything": "^2.1.13",
+				"ts-poet": "^6.7.0",
+				"ts-proto-descriptors": "2.0.0"
+			},
+			"bin": {
+				"protoc-gen-ts_proto": "protoc-gen-ts_proto"
+			}
+		},
+		"node_modules/ts-proto-descriptors": {
+			"version": "2.0.0",
+			"resolved": "https://registry.npmjs.org/ts-proto-descriptors/-/ts-proto-descriptors-2.0.0.tgz",
+			"integrity": "sha512-wHcTH3xIv11jxgkX5OyCSFfw27agpInAd6yh89hKG6zqIXnjW9SYqSER2CVQxdPj4czeOhGagNvZBEbJPy7qkw==",
+			"dev": true,
+			"license": "ISC",
+			"dependencies": {
+				"@bufbuild/protobuf": "^2.0.0"
+			}
+		},
 		"node_modules/tslib": {
 			"version": "1.14.1",
 			"resolved": "https://registry.npmjs.org/tslib/-/tslib-1.14.1.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"
@@ -116,6 +116,21 @@
 				"command": "cline.fixWithCline",
 				"title": "Fix with Cline",
 				"category": "Cline"
+			},
+			{
+				"command": "cline.focusChatInput",
+				"title": "Jump to Chat Input",
+				"category": "Cline"
+			}
+		],
+		"keybindings": [
+			{
+				"command": "cline.addToChat",
+				"key": "cmd+'",
+				"mac": "cmd+'",
+				"win": "ctrl+'",
+				"linux": "ctrl+'",
+				"when": "editorHasSelection"
 			}
 		],
 		"menus": {
@@ -214,21 +229,6 @@
 					},
 					"description": "Settings for VSCode Language Model API"
 				},
-				"cline.mcp.mode": {
-					"type": "string",
-					"enum": [
-						"full",
-						"server-use-only",
-						"off"
-					],
-					"enumDescriptions": [
-						"Enable all MCP functionality (server use and build instructions)",
-						"Enable MCP server use only (excludes instructions about building MCP servers)",
-						"Disable all MCP functionality"
-					],
-					"default": "full",
-					"description": "Controls MCP inclusion in prompts, reduces token usage if you only need access to certain functionality."
-				},
 				"cline.enableCheckpoints": {
 					"type": "boolean",
 					"default": true,
@@ -292,8 +292,13 @@
 		"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",
+		"protos": "node proto/build-proto.js && prettier src/shared/proto --write && prettier src/core/controller --write",
+		"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",
@@ -301,8 +306,9 @@
 		"lint": "eslint src --ext ts && eslint webview-ui/src --ext ts",
 		"format": "prettier . --check",
 		"format:fix": "prettier . --write",
-		"test": "vscode-test",
+		"test": "npm-run-all test:unit test:integration",
 		"test:ci": "node scripts/test-ci.js",
+		"test:integration": "vscode-test",
 		"test:unit": "TS_NODE_PROJECT='./tsconfig.unit-test.json' mocha",
 		"test:coverage": "vscode-test --coverage",
 		"install:all": "npm install && cd webview-ui && npm install",
@@ -334,15 +340,18 @@
 		"@vscode/test-cli": "^0.0.9",
 		"@vscode/test-electron": "^2.4.0",
 		"chai": "^4.3.10",
+		"chalk": "^5.3.0",
 		"esbuild": "^0.25.0",
 		"eslint": "^8.57.0",
 		"husky": "^9.1.7",
 		"npm-run-all": "^4.1.5",
 		"prettier": "^3.3.3",
+		"protoc-gen-ts": "^0.8.7",
 		"proxyquire": "^2.1.3",
 		"should": "^13.2.3",
 		"sinon": "^19.0.2",
 		"ts-node": "^10.9.2",
+		"ts-proto": "^2.6.1",
 		"typescript": "^5.4.5"
 	},
 	"dependencies": {
@@ -350,8 +359,10 @@
 		"@anthropic-ai/sdk": "^0.37.0",
 		"@anthropic-ai/vertex-sdk": "^0.6.4",
 		"@aws-sdk/client-bedrock-runtime": "^3.758.0",
+		"@bufbuild/protobuf": "^2.2.5",
 		"@google-cloud/vertexai": "^1.9.3",
 		"@google/generative-ai": "^0.18.0",
+		"@grpc/grpc-js": "^1.9.15",
 		"@mistralai/mistralai": "^1.5.0",
 		"@modelcontextprotocol/sdk": "^1.7.0",
 		"@opentelemetry/api": "^1.4.1",
@@ -360,10 +371,12 @@
 		"@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",
 		"chokidar": "^4.0.1",
+		"chrome-launcher": "^1.1.2",
 		"clone-deep": "^4.0.1",
 		"default-shell": "^2.2.0",
 		"diff": "^5.2.0",

proto/browser.proto

@@ -0,0 +1,22 @@
+syntax = "proto3";
+
+package cline;
+
+import "common.proto";
+
+service BrowserService {
+  rpc getBrowserConnectionInfo(EmptyRequest) returns (BrowserConnectionInfo);
+  rpc testBrowserConnection(StringRequest) returns (BrowserConnection);
+}
+
+message BrowserConnectionInfo {
+  bool is_connected = 1;
+  bool is_remote = 2;
+  optional string host = 3;
+}
+
+message BrowserConnection {
+  bool success = 1;
+  string message = 2;
+  optional string endpoint = 3;
+}

proto/build-proto.js

@@ -0,0 +1,157 @@
+#!/usr/bin/env node
+
+import * as fs from "fs/promises"
+import * as path from "path"
+import { execSync } from "child_process"
+import { globby } from "globby"
+import chalk from "chalk"
+
+// Get script directory and root directory
+const SCRIPT_DIR = path.dirname(new URL(import.meta.url).pathname)
+const ROOT_DIR = path.resolve(SCRIPT_DIR, "..")
+
+async function main() {
+	console.log(chalk.bold.blue("Starting Protocol Buffer code generation..."))
+
+	// Check if protoc is installed
+	try {
+		const options = { stdio: "ignore" }
+		execSync("protoc --version", options)
+	} catch (error) {
+		console.warn(chalk.yellow("Warning: protoc is not installed. Skipping proto generation."))
+		console.warn(chalk.yellow("To install Protocol Buffers compiler, visit: https://grpc.io/docs/protoc-installation/"))
+		process.exit(0) // Exit with success as requested
+	}
+
+	// Check if ts-proto plugin is available
+	const TS_PROTO_PLUGIN = path.join(ROOT_DIR, "node_modules", ".bin", "protoc-gen-ts_proto")
+	try {
+		await fs.access(TS_PROTO_PLUGIN)
+	} catch (error) {
+		console.error(chalk.red("Error: ts-proto plugin not found at"), TS_PROTO_PLUGIN)
+		console.error(chalk.red('Please run "npm install" to install the required dependencies.'))
+		process.exit(1)
+	}
+
+	// Define output directories
+	const TS_OUT_DIR = path.join(ROOT_DIR, "src", "shared", "proto")
+
+	// Create output directory if it doesn't exist
+	await fs.mkdir(TS_OUT_DIR, { recursive: true })
+
+	// Clean up existing generated files
+	console.log(chalk.cyan("Cleaning up existing generated TypeScript files..."))
+	const existingFiles = await globby("**/*.ts", { cwd: TS_OUT_DIR })
+	for (const file of existingFiles) {
+		await fs.unlink(path.join(TS_OUT_DIR, file))
+	}
+
+	// Process all proto files
+	console.log(chalk.cyan("Processing proto files from"), SCRIPT_DIR)
+	const protoFiles = await globby("**/*.proto", { cwd: SCRIPT_DIR })
+
+	for (const protoFile of protoFiles) {
+		console.log(chalk.cyan(`Generating TypeScript code for ${protoFile}...`))
+
+		// Build the protoc command with proper path handling for cross-platform
+		const protocCommand = [
+			"protoc",
+			`--plugin=protoc-gen-ts_proto="${TS_PROTO_PLUGIN}"`,
+			`--ts_proto_out="${TS_OUT_DIR}"`,
+			"--ts_proto_opt=outputServices=generic-definitions,env=node,esModuleInterop=true,useDate=false,useOptionals=messages",
+			`--proto_path="${SCRIPT_DIR}"`,
+			`"${path.join(SCRIPT_DIR, protoFile)}"`,
+		].join(" ")
+
+		try {
+			const execOptions = {
+				stdio: "inherit",
+			}
+			execSync(protocCommand, execOptions)
+		} catch (error) {
+			console.error(chalk.red(`Error generating TypeScript for ${protoFile}:`), error)
+			process.exit(1)
+		}
+	}
+
+	console.log(chalk.green("Protocol Buffer code generation completed successfully."))
+	console.log(chalk.green(`TypeScript files generated in: ${TS_OUT_DIR}`))
+
+	// Generate method registration files
+	await generateMethodRegistrations()
+
+	// Make the script executable
+	try {
+		await fs.chmod(path.join(SCRIPT_DIR, "build-proto.js"), 0o755)
+	} catch (error) {
+		console.warn(chalk.yellow("Warning: Could not make script executable:"), error)
+	}
+}
+
+async function generateMethodRegistrations() {
+	console.log(chalk.cyan("Generating method registration files..."))
+
+	const serviceDirs = [
+		path.join(ROOT_DIR, "src", "core", "controller", "browser"),
+		// Add more service directories here as needed
+	]
+
+	for (const serviceDir of serviceDirs) {
+		try {
+			await fs.access(serviceDir)
+		} catch (error) {
+			console.log(chalk.gray(`Skipping ${serviceDir} - directory does not exist`))
+			continue
+		}
+
+		const serviceName = path.basename(serviceDir)
+		const registryFile = path.join(serviceDir, "methods.ts")
+
+		console.log(chalk.cyan(`Generating method registrations for ${serviceName}...`))
+
+		// Get all TypeScript files in the service directory
+		const files = await globby("*.ts", { cwd: serviceDir })
+
+		// Filter out index.ts and methods.ts
+		const implementationFiles = files.filter((file) => file !== "index.ts" && file !== "methods.ts")
+
+		// Create the output file with header
+		let content = `// AUTO-GENERATED FILE - DO NOT MODIFY DIRECTLY
+// Generated by proto/build-proto.js
+
+// Import all method implementations
+import { registerMethod } from "./index"\n`
+
+		// Add imports for all implementation files
+		for (const file of implementationFiles) {
+			const baseName = path.basename(file, ".ts")
+			content += `import { ${baseName} } from "./${baseName}"\n`
+		}
+
+		// Add registration function
+		content += `\n// Register all ${serviceName} service methods
+export function registerAllMethods(): void {
+\t// Register each method with the registry\n`
+
+		// Add registration statements
+		for (const file of implementationFiles) {
+			const baseName = path.basename(file, ".ts")
+			content += `\tregisterMethod("${baseName}", ${baseName})\n`
+		}
+
+		// Close the function
+		content += `}`
+
+		// Write the file
+		await fs.writeFile(registryFile, content)
+		console.log(chalk.green(`Generated ${registryFile}`))
+	}
+
+	console.log(chalk.green("Method registration files generated successfully."))
+}
+
+// Run the main function
+main().catch((error) => {
+	console.error(chalk.red("Error:"), error)
+	process.exit(1)
+})

proto/common.proto

@@ -0,0 +1,40 @@
+syntax = "proto3";
+
+package cline;
+
+message Metadata {
+}
+
+message EmptyRequest {
+  Metadata metadata = 1;
+}
+
+message Empty {
+}
+
+message StringRequest {
+  Metadata metadata = 1;
+  string value = 2;
+}
+
+message String {
+  string value = 1;
+}
+
+message Int64Request {
+  Metadata metadata = 1;
+  int64 value = 2;
+}
+
+message Int64 {
+  int64 value = 1;
+}
+
+message BytesRequest {
+  Metadata metadata = 1;
+  bytes value = 2;
+}
+
+message Bytes {
+  bytes value = 1;
+}

proto/package.json

@@ -0,0 +1,3 @@
+{
+	"type": "module"
+}

scripts/test-ci.js

@@ -9,10 +9,10 @@ try {
 		execSync("which xvfb-run", { stdio: "ignore" })
 
 		console.log("xvfb-run is installed. Running tests with xvfb-run...")
-		execSync("xvfb-run -a npm run test", { stdio: "inherit" })
+		execSync("xvfb-run -a npm run test:integration", { stdio: "inherit" })
 	} else {
 		console.log("Non-Linux environment detected. Running tests normally.")
-		execSync("npm run test", { stdio: "inherit" })
+		execSync("npm run test:integration", { stdio: "inherit" })
 	}
 } catch (error) {
 	if (process.platform === "linux") {

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/litellm.ts

@@ -71,9 +71,37 @@ export class LiteLlmHandler implements ApiHandler {
 			temperature = undefined // Thinking mode doesn't support temperature
 		}
 
+		// Define cache control object if prompt caching is enabled
+		const cacheControl = this.options.liteLlmUsePromptCache ? { cache_control: { type: "ephemeral" } } : undefined
+
+		// Add cache_control to system message if enabled
+		const enhancedSystemMessage = {
+			...systemMessage,
+			...(cacheControl && cacheControl),
+		}
+
+		// Find the last two user messages to apply caching
+		const userMsgIndices = formattedMessages.reduce(
+			(acc, msg, index) => (msg.role === "user" ? [...acc, index] : acc),
+			[] as number[],
+		)
+		const lastUserMsgIndex = userMsgIndices[userMsgIndices.length - 1] ?? -1
+		const secondLastUserMsgIndex = userMsgIndices[userMsgIndices.length - 2] ?? -1
+
+		// Apply cache_control to the last two user messages if enabled
+		const enhancedMessages = formattedMessages.map((message, index) => {
+			if ((index === lastUserMsgIndex || index === secondLastUserMsgIndex) && cacheControl) {
+				return {
+					...message,
+					...cacheControl,
+				}
+			}
+			return message
+		})
+
 		const stream = await this.client.chat.completions.create({
 			model: this.options.liteLlmModelId || liteLlmDefaultModelId,
-			messages: [systemMessage, ...formattedMessages],
+			messages: [enhancedSystemMessage, ...enhancedMessages],
 			temperature,
 			stream: true,
 			stream_options: { include_usage: true },
@@ -111,10 +139,27 @@ export class LiteLlmHandler implements ApiHandler {
 			if (chunk.usage) {
 				const totalCost =
 					(inputCost * chunk.usage.prompt_tokens) / 1e6 + (outputCost * chunk.usage.completion_tokens) / 1e6
+
+				// Extract cache-related information if available
+				// Need to use type assertion since these properties are not in the standard OpenAI types
+				const usage = chunk.usage as {
+					prompt_tokens: number
+					completion_tokens: number
+					cache_creation_input_tokens?: number
+					prompt_cache_miss_tokens?: number
+					cache_read_input_tokens?: number
+					prompt_cache_hit_tokens?: number
+				}
+
+				const cacheWriteTokens = usage.cache_creation_input_tokens || usage.prompt_cache_miss_tokens || 0
+				const cacheReadTokens = usage.cache_read_input_tokens || usage.prompt_cache_hit_tokens || 0
+
 				yield {
 					type: "usage",
-					inputTokens: chunk.usage.prompt_tokens || 0,
-					outputTokens: chunk.usage.completion_tokens || 0,
+					inputTokens: usage.prompt_tokens || 0,
+					outputTokens: usage.completion_tokens || 0,
+					cacheWriteTokens: cacheWriteTokens > 0 ? cacheWriteTokens : undefined,
+					cacheReadTokens: cacheReadTokens > 0 ? cacheReadTokens : undefined,
 					totalCost,
 				}
 			}

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
@@ -50,7 +50,7 @@ export class OpenAiNativeHandler implements ApiHandler {
 			case "o1":
 			case "o1-preview":
 			case "o1-mini": {
-				// o1 doesnt support streaming, non-1 temp, or system prompt
+				// o1 doesn't support streaming, non-1 temp, or system prompt
 				const response = await this.client.chat.completions.create({
 					model: model.id,
 					messages: [{ role: "user", content: systemPrompt }, ...convertToOpenAiMessages(messages)],
@@ -64,6 +64,8 @@ export class OpenAiNativeHandler implements ApiHandler {
 
 				break
 			}
+			case "o4-mini":
+			case "o3":
 			case "o3-mini": {
 				const stream = await this.client.chat.completions.create({
 					model: model.id,

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
@@ -35,7 +35,7 @@ export class OpenAiHandler implements ApiHandler {
 		const modelId = this.options.openAiModelId ?? ""
 		const isDeepseekReasoner = modelId.includes("deepseek-reasoner")
 		const isR1FormatRequired = this.options.openAiModelInfo?.isR1FormatRequired ?? false
-		const isO3Mini = modelId.includes("o3-mini")
+		const isReasoningModelFamily = modelId.includes("o3") || modelId.includes("o4")
 
 		let openAiMessages: OpenAI.Chat.ChatCompletionMessageParam[] = [
 			{ role: "system", content: systemPrompt },
@@ -55,7 +55,7 @@ export class OpenAiHandler implements ApiHandler {
 			openAiMessages = convertToR1Format([{ role: "user", content: systemPrompt }, ...messages])
 		}
 
-		if (isO3Mini) {
+		if (isReasoningModelFamily) {
 			openAiMessages = [{ role: "developer", content: systemPrompt }, ...convertToOpenAiMessages(messages)]
 			temperature = undefined // does not support temperature
 			reasoningEffort = (this.options.o3MiniReasoningEffort as ChatCompletionReasoningEffort) || "medium"

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/vertex.ts

@@ -5,6 +5,7 @@ import { ApiHandler } from "../"
 import { ApiHandlerOptions, ModelInfo, vertexDefaultModelId, VertexModelId, vertexModels } from "../../shared/api"
 import { ApiStream } from "../transform/stream"
 import { VertexAI } from "@google-cloud/vertexai"
+import { calculateApiCostOpenAI } from "../../utils/cost"
 
 // https://docs.anthropic.com/en/api/claude-on-vertex-ai
 export class VertexHandler implements ApiHandler {
@@ -266,6 +267,17 @@ export class VertexHandler implements ApiHandler {
 					}
 				}
 			}
+			// Handle token usage metadata
+			const { usageMetadata } = await streamingResult.response
+			if (usageMetadata) {
+				const { promptTokenCount = 0, candidatesTokenCount = 0 } = usageMetadata
+				yield {
+					type: "usage",
+					inputTokens: promptTokenCount,
+					outputTokens: candidatesTokenCount,
+					totalCost: calculateApiCostOpenAI(model.info, promptTokenCount, candidatesTokenCount, 0, 0),
+				}
+			}
 		}
 	}
 

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"
 
 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 (delta && "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

@@ -21,7 +21,9 @@ export const toolUseNames = [
 	"access_mcp_resource",
 	"ask_followup_question",
 	"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" | ...)
@@ -48,6 +50,7 @@ export const toolParamNames = [
 	"options",
 	"response",
 	"result",
+	"context",
 ] as const
 
 export type ToolParamName = (typeof toolParamNames)[number]

src/core/assistant-message/parse-assistant-message.ts

@@ -123,7 +123,7 @@ export function parseAssistantMessage(assistantMessage: string) {
 		contentBlocks.push(currentToolUse)
 	}
 
-	// Note: it doesnt matter if check for currentToolUse or currentTextContent, only one of them will be defined since only one can be partial at a time
+	// Note: it doesn't matter if check for currentToolUse or currentTextContent, only one of them will be defined since only one can be partial at a time
 	if (currentTextContent) {
 		// stream did not complete text content, add it as partial
 		contentBlocks.push(currentTextContent)

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

@@ -1,9 +1,9 @@
 import { Anthropic } from "@anthropic-ai/sdk"
-import { ClineApiReqInfo, ClineMessage } from "../../shared/ExtensionMessage"
-import { ApiHandler } from "../../api"
-import { OpenAiHandler } from "../../api/providers/openai"
+import { ClineApiReqInfo, ClineMessage } from "../../../shared/ExtensionMessage"
+import { ApiHandler } from "../../../api"
+import { getContextWindowInfo } from "./context-window-utils"
 
-export class ContextManager {
+class ContextManager {
 	getNewContextMessagesAndMetadata(
 		apiConversationHistory: Anthropic.Messages.MessageParam[],
 		clineMessages: ClineMessage[],
@@ -19,25 +19,7 @@ export class ContextManager {
 			if (previousRequest && previousRequest.text) {
 				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/ContextManager.ts

@@ -0,0 +1,834 @@
+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,
+	NO_FILE_READ = 1,
+	READ_FILE_TOOL = 2,
+	ALTER_FILE_TOOL = 3,
+	FILE_MENTION = 4,
+}
+
+// array of string values allows us to cover all changes for message types currently supported
+type MessageContent = string[]
+type MessageMetadata = string[][]
+
+// Type for a single context update
+type ContextUpdate = [number, string, MessageContent, MessageMetadata] // [timestamp, updateType, update, metadata]
+
+// Type for the serialized format of our nested maps
+type SerializedContextHistory = Array<
+	[
+		number, // messageIndex
+		[
+			number, // EditType (message type)
+			Array<
+				[
+					number, // blockIndex
+					ContextUpdate[], // updates array (now with 4 elements including metadata)
+				]
+			>,
+		],
+	]
+>
+
+export class ContextManager {
+	// mapping from the apiMessages outer index to the inner message index to a list of actual changes, ordered by timestamp
+	// timestamp is required in order to support full checkpointing, where the changes we apply need to be able to be undone when
+	// moving to an earlier conversation history checkpoint - this ordering intuitively allows for binary search on truncation
+	// there is also a number stored for each (EditType) which defines which message type it is, for custom handling
+
+	// format:  { outerIndex => [EditType, { innerIndex => [[timestamp, updateType, update], ...] }] }
+	// example: { 1 => { [0, 0 => [[<timestamp>, "text", "[NOTE] Some previous conversation history with the user has been removed ..."], ...] }] }
+	// the above example would be how we update the first assistant message to indicate we truncated text
+	private contextHistoryUpdates: Map<number, [number, Map<number, ContextUpdate[]>]>
+
+	constructor() {
+		this.contextHistoryUpdates = new Map()
+	}
+
+	/**
+	 * public function for loading contextHistoryUpdates from disk, if it exists
+	 */
+	async initializeContextHistory(taskDirectory: string) {
+		this.contextHistoryUpdates = await this.getSavedContextHistory(taskDirectory)
+	}
+
+	/**
+	 * get the stored context history updates from disk
+	 */
+	private async getSavedContextHistory(taskDirectory: string): Promise<Map<number, [number, Map<number, ContextUpdate[]>]>> {
+		try {
+			const filePath = path.join(taskDirectory, GlobalFileNames.contextHistory)
+			if (await fileExistsAtPath(filePath)) {
+				const data = await fs.readFile(filePath, "utf8")
+				const serializedUpdates = JSON.parse(data) as SerializedContextHistory
+
+				// Update to properly reconstruct the tuple structure
+				return new Map(
+					serializedUpdates.map(([messageIndex, [numberValue, innerMapArray]]) => [
+						messageIndex,
+						[numberValue, new Map(innerMapArray)],
+					]),
+				)
+			}
+		} catch (error) {
+			console.error("Failed to load context history:", error)
+		}
+		return new Map()
+	}
+
+	/**
+	 * save the context history updates to disk
+	 */
+	private async saveContextHistory(taskDirectory: string) {
+		try {
+			const serializedUpdates: SerializedContextHistory = Array.from(this.contextHistoryUpdates.entries()).map(
+				([messageIndex, [numberValue, innerMap]]) => [messageIndex, [numberValue, Array.from(innerMap.entries())]],
+			)
+
+			await fs.writeFile(
+				path.join(taskDirectory, GlobalFileNames.contextHistory),
+				JSON.stringify(serializedUpdates),
+				"utf8",
+			)
+		} catch (error) {
+			console.error("Failed to save context history:", error)
+		}
+	}
+
+	/**
+	 * primary entry point for getting up to date context & truncating when required
+	 */
+	async getNewContextMessagesAndMetadata(
+		apiConversationHistory: Anthropic.Messages.MessageParam[],
+		clineMessages: ClineMessage[],
+		api: ApiHandler,
+		conversationHistoryDeletedRange: [number, number] | undefined,
+		previousApiReqIndex: number,
+		taskDirectory: string,
+	) {
+		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 timestamp = previousRequest.ts
+				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
+					const keep = totalTokens / 2 > maxAllowedSize ? "quarter" : "half"
+
+					// we later check how many chars we trim to determine if we should still truncate history
+					let [anyContextUpdates, uniqueFileReadIndices] = this.applyContextOptimizations(
+						apiConversationHistory,
+						conversationHistoryDeletedRange ? conversationHistoryDeletedRange[1] + 1 : 2,
+						timestamp,
+					)
+
+					let needToTruncate = true
+					if (anyContextUpdates) {
+						// determine whether we've saved enough chars to not truncate
+						const charactersSavedPercentage = this.calculateContextOptimizationMetrics(
+							apiConversationHistory,
+							conversationHistoryDeletedRange,
+							uniqueFileReadIndices,
+						)
+						if (charactersSavedPercentage >= 0.3) {
+							needToTruncate = false
+						}
+					}
+
+					if (needToTruncate) {
+						// go ahead with truncation
+						anyContextUpdates = this.applyStandardContextTruncationNoticeChange(timestamp) || anyContextUpdates
+
+						// 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
+					}
+
+					// if we alter the context history, save the updated version to disk
+					if (anyContextUpdates) {
+						await this.saveContextHistory(taskDirectory)
+					}
+				}
+			}
+		}
+
+		const truncatedConversationHistory = this.getAndAlterTruncatedMessages(
+			apiConversationHistory,
+			conversationHistoryDeletedRange,
+		)
+
+		return {
+			conversationHistoryDeletedRange: conversationHistoryDeletedRange,
+			updatedConversationHistoryDeletedRange: updatedConversationHistoryDeletedRange,
+			truncatedConversationHistory: truncatedConversationHistory,
+		}
+	}
+
+	/**
+	 * get truncation range
+	 */
+	public getNextTruncationRange(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		currentDeletedRange: [number, number] | undefined,
+		keep: "half" | "quarter",
+	): [number, number] {
+		// We always keep the first user-assistant pairing, and truncate an even number of messages from there
+		const rangeStartIndex = 2 // index 0 and 1 are kept
+		const startOfRest = currentDeletedRange ? currentDeletedRange[1] + 1 : 2 // inclusive starting index
+
+		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 // inclusive ending index
+
+		// Make sure that the last message being removed is a assistant message, so the next message after the initial user-assistant pair is an assistant message. This preserves 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 !== "assistant") {
+			rangeEndIndex -= 1
+		}
+
+		// this is an inclusive range that will be removed from the conversation history
+		return [rangeStartIndex, rangeEndIndex]
+	}
+
+	/**
+	 * external interface to support old calls
+	 */
+	public getTruncatedMessages(
+		messages: Anthropic.Messages.MessageParam[],
+		deletedRange: [number, number] | undefined,
+	): Anthropic.Messages.MessageParam[] {
+		return this.getAndAlterTruncatedMessages(messages, deletedRange)
+	}
+
+	/**
+	 * apply all required truncation methods to the messages in context
+	 */
+	private getAndAlterTruncatedMessages(
+		messages: Anthropic.Messages.MessageParam[],
+		deletedRange: [number, number] | undefined,
+	): Anthropic.Messages.MessageParam[] {
+		if (messages.length <= 1) {
+			return messages
+		}
+
+		const updatedMessages = this.applyContextHistoryUpdates(messages, deletedRange ? deletedRange[1] + 1 : 2)
+
+		// OLD 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 updatedMessages
+	}
+
+	/**
+	 * applies deletedRange truncation and other alterations based on changes in this.contextHistoryUpdates
+	 */
+	private applyContextHistoryUpdates(
+		messages: Anthropic.Messages.MessageParam[],
+		startFromIndex: number,
+	): Anthropic.Messages.MessageParam[] {
+		// runtime is linear in length of user messages, if expecting a limited number of alterations, could be more optimal to loop over alterations
+
+		const firstChunk = messages.slice(0, 2) // get first user-assistant pair
+		const secondChunk = messages.slice(startFromIndex) // get remaining messages within context
+		const messagesToUpdate = [...firstChunk, ...secondChunk]
+
+		// we need the mapping from the local indices in messagesToUpdate to the global array of updates in this.contextHistoryUpdates
+		const originalIndices = [
+			...Array(2).keys(),
+			...Array(secondChunk.length)
+				.fill(0)
+				.map((_, i) => i + startFromIndex),
+		]
+
+		for (let arrayIndex = 0; arrayIndex < messagesToUpdate.length; arrayIndex++) {
+			const messageIndex = originalIndices[arrayIndex]
+
+			const innerTuple = this.contextHistoryUpdates.get(messageIndex)
+			if (!innerTuple) {
+				continue
+			}
+
+			// because we are altering this, we need a deep copy
+			messagesToUpdate[arrayIndex] = cloneDeep(messagesToUpdate[arrayIndex])
+
+			// Extract the map from the tuple
+			const innerMap = innerTuple[1]
+			for (const [blockIndex, changes] of innerMap) {
+				// apply the latest change among n changes - [timestamp, updateType, update]
+				const latestChange = changes[changes.length - 1]
+
+				if (latestChange[1] === "text") {
+					// only altering text for now
+					const message = messagesToUpdate[arrayIndex]
+
+					if (Array.isArray(message.content)) {
+						const block = message.content[blockIndex]
+						if (block && block.type === "text") {
+							block.text = latestChange[2][0]
+						}
+					}
+				}
+			}
+		}
+
+		return messagesToUpdate
+	}
+
+	/**
+	 * removes all context history updates that occurred after the specified timestamp and saves to disk
+	 */
+	async truncateContextHistory(timestamp: number, taskDirectory: string): Promise<void> {
+		this.truncateContextHistoryAtTimestamp(this.contextHistoryUpdates, timestamp)
+
+		// save the modified context history to disk
+		await this.saveContextHistory(taskDirectory)
+	}
+
+	/**
+	 * alters the context history to remove all alterations after a given timestamp
+	 * removes the index if there are no alterations there anymore, both outer and inner indices
+	 */
+	private truncateContextHistoryAtTimestamp(
+		contextHistory: Map<number, [number, Map<number, ContextUpdate[]>]>,
+		timestamp: number,
+	): void {
+		for (const [messageIndex, [_, innerMap]] of contextHistory) {
+			// track which blockIndices to delete
+			const blockIndicesToDelete: number[] = []
+
+			// loop over the innerIndices of the messages in this block
+			for (const [blockIndex, updates] of innerMap) {
+				// updates ordered by timestamp, so find cutoff point by iterating from right to left
+				let cutoffIndex = updates.length - 1
+				while (cutoffIndex >= 0 && updates[cutoffIndex][0] > timestamp) {
+					cutoffIndex--
+				}
+
+				// If we found updates to remove
+				if (cutoffIndex < updates.length - 1) {
+					// Modify the array in place to keep only updates up to cutoffIndex
+					updates.length = cutoffIndex + 1
+
+					// If no updates left after truncation, mark this block for deletion
+					if (updates.length === 0) {
+						blockIndicesToDelete.push(blockIndex)
+					}
+				}
+			}
+
+			// Remove empty blocks from inner map
+			for (const blockIndex of blockIndicesToDelete) {
+				innerMap.delete(blockIndex)
+			}
+
+			// If inner map is now empty, remove the message index from outer map
+			if (innerMap.size === 0) {
+				contextHistory.delete(messageIndex)
+			}
+		}
+	}
+
+	/**
+	 * applies the context optimization steps and returns whether any changes were made
+	 */
+	private applyContextOptimizations(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		startFromIndex: number,
+		timestamp: number,
+	): [boolean, Set<number>] {
+		const [fileReadUpdatesBool, uniqueFileReadIndices] = this.findAndPotentiallySaveFileReadContextHistoryUpdates(
+			apiMessages,
+			startFromIndex,
+			timestamp,
+		)
+
+		// true if any context optimization steps alter state
+		const contextHistoryUpdated = fileReadUpdatesBool
+
+		return [contextHistoryUpdated, uniqueFileReadIndices]
+	}
+
+	/**
+	 * if there is any truncation and there is no other alteration already set, alter the assistant message to indicate this occurred
+	 */
+	private applyStandardContextTruncationNoticeChange(timestamp: number): boolean {
+		if (!this.contextHistoryUpdates.has(1)) {
+			// first assistant message always at index 1
+			const innerMap = new Map<number, ContextUpdate[]>()
+			innerMap.set(0, [[timestamp, "text", [formatResponse.contextTruncationNotice()], []]])
+			this.contextHistoryUpdates.set(1, [0, innerMap]) // EditType is undefined for first assistant message
+			return true
+		}
+		return false
+	}
+
+	/**
+	 * wraps the logic for determining file reads to overwrite, and altering state
+	 * returns whether any updates were made (bool) and indices where updates were made
+	 */
+	private findAndPotentiallySaveFileReadContextHistoryUpdates(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		startFromIndex: number,
+		timestamp: number,
+	): [boolean, Set<number>] {
+		const [fileReadIndices, messageFilePaths] = this.getPossibleDuplicateFileReads(apiMessages, startFromIndex)
+		return this.applyFileReadContextHistoryUpdates(fileReadIndices, messageFilePaths, apiMessages, timestamp)
+	}
+
+	/**
+	 * generate a mapping from unique file reads from multiple tool calls to their outer index position(s)
+	 * also return additional metadata to support multiple file reads in file mention text blocks
+	 */
+	private getPossibleDuplicateFileReads(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		startFromIndex: number,
+	): [Map<string, [number, number, string, string][]>, Map<number, string[]>] {
+		// fileReadIndices: { fileName => [outerIndex, EditType, searchText, replaceText] }
+		// messageFilePaths: { outerIndex => [fileRead1, fileRead2, ..] }
+		// searchText in fileReadIndices is only required for file mention file-reads since there can be more than one file in the text
+		// searchText will be the empty string "" in the case that it's not required, for non-file mentions
+		// messageFilePaths is only used for file mentions as there can be multiple files read in the same text chunk
+
+		// for all text blocks per file, has info for updating the block
+		const fileReadIndices = new Map<string, [number, number, string, string][]>()
+
+		// for file mention text blocks, track all the unique files read
+		const messageFilePaths = new Map<number, string[]>()
+
+		for (let i = startFromIndex; i < apiMessages.length; i++) {
+			let thisExistingFileReads: string[] = []
+
+			if (this.contextHistoryUpdates.has(i)) {
+				const innerTuple = this.contextHistoryUpdates.get(i)
+
+				if (innerTuple) {
+					// safety check
+					const editType = innerTuple[0]
+
+					if (editType === EditType.FILE_MENTION) {
+						const innerMap = innerTuple[1]
+
+						const blockIndex = 1 // file mention blocks assumed to be at index 1
+						const blockUpdates = innerMap.get(blockIndex)
+
+						// if we have updated this text previously, we want to check whether the lists of files in the metadata are the same
+						if (blockUpdates && blockUpdates.length > 0) {
+							// the first list indicates the files we have replaced in this text, second list indicates all unique files in this text
+							// if they are equal then we have replaced all the files in this text already, and can ignore further processing
+							if (
+								blockUpdates[blockUpdates.length - 1][3][0].length ===
+								blockUpdates[blockUpdates.length - 1][3][1].length
+							) {
+								continue
+							}
+							// otherwise there are still file reads here we can overwrite, so still need to process this text chunk
+							// to do so we need to keep track of which files we've already replaced so we don't replace them again
+							else {
+								thisExistingFileReads = blockUpdates[blockUpdates.length - 1][3][0]
+							}
+						}
+					} else {
+						// for all other cases we can assume that we dont need to check this again
+						continue
+					}
+				}
+			}
+
+			const message = apiMessages[i]
+			if (message.role === "user" && Array.isArray(message.content) && message.content.length > 0) {
+				const firstBlock = message.content[0]
+				if (firstBlock.type === "text") {
+					const matchTup = this.parsePotentialToolCall(firstBlock.text)
+					let foundNormalFileRead = false
+					if (matchTup) {
+						if (matchTup[0] === "read_file") {
+							this.handleReadFileToolCall(i, matchTup[1], fileReadIndices)
+							foundNormalFileRead = true
+						} else if (matchTup[0] === "replace_in_file" || matchTup[0] === "write_to_file") {
+							if (message.content.length > 1) {
+								const secondBlock = message.content[1]
+								if (secondBlock.type === "text") {
+									this.handlePotentialFileChangeToolCalls(i, matchTup[1], secondBlock.text, fileReadIndices)
+									foundNormalFileRead = true
+								}
+							}
+						}
+					}
+
+					// file mentions can happen in most other user message blocks
+					if (!foundNormalFileRead) {
+						if (message.content.length > 1) {
+							const secondBlock = message.content[1]
+							if (secondBlock.type === "text") {
+								const [hasFileRead, filePaths] = this.handlePotentialFileMentionCalls(
+									i,
+									secondBlock.text,
+									fileReadIndices,
+									thisExistingFileReads, // file reads we've already replaced in this text in the latest version of this updated text
+								)
+								if (hasFileRead) {
+									messageFilePaths.set(i, filePaths) // all file paths in this string
+								}
+							}
+						}
+					}
+				}
+			}
+		}
+
+		return [fileReadIndices, messageFilePaths]
+	}
+
+	/**
+	 * handles potential file content mentions in text blocks
+	 * there will not be more than one of the same file read in a text block
+	 */
+	private handlePotentialFileMentionCalls(
+		i: number,
+		secondBlockText: string,
+		fileReadIndices: Map<string, [number, number, string, string][]>,
+		thisExistingFileReads: string[],
+	): [boolean, string[]] {
+		const pattern = new RegExp(`<file_content path="([^"]*)">([\\s\\S]*?)</file_content>`, "g")
+
+		let foundMatch = false
+		const filePaths: string[] = []
+
+		let match
+		while ((match = pattern.exec(secondBlockText)) !== null) {
+			foundMatch = true
+
+			const filePath = match[1]
+			filePaths.push(filePath) // we will record all unique paths from file mentions in this text
+
+			// we can assume that thisExistingFileReads does not have many entries
+			if (!thisExistingFileReads.includes(filePath)) {
+				// meaning we haven't already replaced this file read
+
+				const entireMatch = match[0] // The entire matched string
+
+				// Create the replacement text - keep the tags but replace the content
+				const replacementText = `<file_content path="${filePath}">${formatResponse.duplicateFileReadNotice()}</file_content>`
+
+				const indices = fileReadIndices.get(filePath) || []
+				indices.push([i, EditType.FILE_MENTION, entireMatch, replacementText])
+				fileReadIndices.set(filePath, indices)
+			}
+		}
+
+		return [foundMatch, filePaths]
+	}
+
+	/**
+	 * parses specific tool call formats, returns null if no acceptable format is found
+	 */
+	private parsePotentialToolCall(text: string): [string, string] | null {
+		const match = text.match(/^\[([^\s]+) for '([^']+)'\] Result:$/)
+
+		if (!match) {
+			return null
+		}
+
+		return [match[1], match[2]]
+	}
+
+	/**
+	 * file_read tool call always pastes the file, so this is always a hit
+	 */
+	private handleReadFileToolCall(
+		i: number,
+		filePath: string,
+		fileReadIndices: Map<string, [number, number, string, string][]>,
+	) {
+		const indices = fileReadIndices.get(filePath) || []
+		indices.push([i, EditType.READ_FILE_TOOL, "", formatResponse.duplicateFileReadNotice()])
+		fileReadIndices.set(filePath, indices)
+	}
+
+	/**
+	 * write_to_file and replace_in_file tool output are handled similarly
+	 */
+	private handlePotentialFileChangeToolCalls(
+		i: number,
+		filePath: string,
+		secondBlockText: string,
+		fileReadIndices: Map<string, [number, number, string, string][]>,
+	) {
+		const pattern = new RegExp(`(<final_file_content path="[^"]*">)[\\s\\S]*?(</final_file_content>)`)
+
+		// check if this exists in the text, it won't exist if the user rejects the file change for example
+		if (pattern.test(secondBlockText)) {
+			const replacementText = secondBlockText.replace(pattern, `$1 ${formatResponse.duplicateFileReadNotice()} $2`)
+			const indices = fileReadIndices.get(filePath) || []
+			indices.push([i, EditType.ALTER_FILE_TOOL, "", replacementText])
+			fileReadIndices.set(filePath, indices)
+		}
+	}
+
+	/**
+	 * alter all occurrences of file read operations and track which messages were updated
+	 * returns the outer index of messages we alter, to count number of changes
+	 */
+	private applyFileReadContextHistoryUpdates(
+		fileReadIndices: Map<string, [number, number, string, string][]>,
+		messageFilePaths: Map<number, string[]>,
+		apiMessages: Anthropic.Messages.MessageParam[],
+		timestamp: number,
+	): [boolean, Set<number>] {
+		let didUpdate = false
+		const updatedMessageIndices = new Set<number>() // track which messages we update on this round
+		const fileMentionUpdates = new Map<number, [string, string[]]>()
+
+		for (const [filePath, indices] of fileReadIndices.entries()) {
+			// Only process if there are multiple reads of the same file, else we will want to keep the latest read of the file
+			if (indices.length > 1) {
+				// Process all but the last index, as we will keep that instance of the file read
+				for (let i = 0; i < indices.length - 1; i++) {
+					const messageIndex = indices[i][0]
+					const messageType = indices[i][1] // EditType value
+					const searchText = indices[i][2] // search text (for file mentions, else empty string)
+					const messageString = indices[i][3] // what we will replace the string with
+
+					didUpdate = true
+					updatedMessageIndices.add(messageIndex)
+
+					// for single-fileread text we can set the updates here
+					// for potential multi-fileread text we need to determine all changes & iteratively update the text prior to saving the final change
+					if (messageType === EditType.FILE_MENTION) {
+						if (!fileMentionUpdates.has(messageIndex)) {
+							// Get base text either from existing updates or from apiMessages
+							let baseText = ""
+							let prevFilesReplaced: string[] = []
+
+							const innerTuple = this.contextHistoryUpdates.get(messageIndex)
+							if (innerTuple) {
+								const blockUpdates = innerTuple[1].get(1) // assumed index=1 for file mention filereads
+								if (blockUpdates && blockUpdates.length > 0) {
+									baseText = blockUpdates[blockUpdates.length - 1][2][0] // index 0 of MessageContent
+									prevFilesReplaced = blockUpdates[blockUpdates.length - 1][3][0] // previously overwritten file reads in this text
+								}
+							}
+
+							// can assume that this content will exist, otherwise it would not have been in fileReadIndices
+							const messageContent = apiMessages[messageIndex]?.content
+							if (!baseText && Array.isArray(messageContent) && messageContent.length > 1) {
+								const contentBlock = messageContent[1] // assume index=1 for all text to replace for file mention filereads
+								if (contentBlock.type === "text") {
+									baseText = contentBlock.text
+								}
+							}
+
+							// prevFilesReplaced keeps track of the previous file reads we've replace in this string, empty array if none
+							fileMentionUpdates.set(messageIndex, [baseText, prevFilesReplaced])
+						}
+
+						// Replace searchText with messageString for all file reads we need to replace in this text
+						if (searchText) {
+							const currentTuple = fileMentionUpdates.get(messageIndex) || ["", []]
+							if (currentTuple[0]) {
+								// safety check
+								// replace this text chunk
+								const updatedText = currentTuple[0].replace(searchText, messageString)
+
+								// add the newly added filePath read
+								const updatedFileReads = currentTuple[1]
+								updatedFileReads.push(filePath)
+
+								fileMentionUpdates.set(messageIndex, [updatedText, updatedFileReads])
+							}
+						}
+					} else {
+						let innerTuple = this.contextHistoryUpdates.get(messageIndex)
+						let innerMap: Map<number, ContextUpdate[]>
+
+						if (!innerTuple) {
+							innerMap = new Map<number, ContextUpdate[]>()
+							this.contextHistoryUpdates.set(messageIndex, [messageType, innerMap])
+						} else {
+							innerMap = innerTuple[1]
+						}
+
+						// block index for file reads from read_file, write_to_file, replace_in_file tools is 1
+						const blockIndex = 1
+
+						const updates = innerMap.get(blockIndex) || []
+
+						// metadata array is empty for non-file mention occurrences
+						updates.push([timestamp, "text", [messageString], []])
+
+						innerMap.set(blockIndex, updates)
+					}
+				}
+			}
+		}
+
+		// apply file mention updates to contextHistoryUpdates
+		// in fileMentionUpdates, filePathsUpdated includes all the file paths which are updated in the latest version of this altered text
+		for (const [messageIndex, [updatedText, filePathsUpdated]] of fileMentionUpdates.entries()) {
+			let innerTuple = this.contextHistoryUpdates.get(messageIndex)
+			let innerMap: Map<number, ContextUpdate[]>
+
+			if (!innerTuple) {
+				innerMap = new Map<number, ContextUpdate[]>()
+				this.contextHistoryUpdates.set(messageIndex, [EditType.FILE_MENTION, innerMap])
+			} else {
+				innerMap = innerTuple[1]
+			}
+
+			const blockIndex = 1 // we only consider the block index of 1 for file mentions
+			const updates = innerMap.get(blockIndex) || []
+
+			// filePathsUpdated includes changes done previously to this timestamp, and right now
+			if (messageFilePaths.has(messageIndex)) {
+				const allFileReads = messageFilePaths.get(messageIndex)
+				if (allFileReads) {
+					// safety check
+					// we gather all the file reads possible in this text from messageFilePaths
+					// filePathsUpdated from fileMentionUpdates stores all the files reads we have replaced now & previously
+					updates.push([timestamp, "text", [updatedText], [filePathsUpdated, allFileReads]])
+					innerMap.set(blockIndex, updates)
+				}
+			}
+		}
+
+		return [didUpdate, updatedMessageIndices]
+	}
+
+	/**
+	 * count total characters in messages and total savings within this range
+	 */
+	private countCharactersAndSavingsInRange(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		startIndex: number,
+		endIndex: number,
+		uniqueFileReadIndices: Set<number>,
+	): { totalCharacters: number; charactersSaved: number } {
+		let totalCharCount = 0
+		let totalCharactersSaved = 0
+
+		for (let i = startIndex; i < endIndex; i++) {
+			// looping over the outer indices of messages
+			const message = apiMessages[i]
+
+			if (!message.content) {
+				continue
+			}
+
+			// hasExistingAlterations checks whether the outer idnex has any changes
+			// hasExistingAlterations will also include the alterations we just made
+			const hasExistingAlterations = this.contextHistoryUpdates.has(i)
+			const hasNewAlterations = uniqueFileReadIndices.has(i)
+
+			if (Array.isArray(message.content)) {
+				for (let blockIndex = 0; blockIndex < message.content.length; blockIndex++) {
+					// looping over inner indices of messages
+					const block = message.content[blockIndex]
+
+					if (block.type === "text" && block.text) {
+						// true if we just altered it, or it was altered before
+						if (hasExistingAlterations) {
+							const innerTuple = this.contextHistoryUpdates.get(i)
+							const updates = innerTuple?.[1].get(blockIndex) // updated text for this inner index
+
+							if (updates && updates.length > 0) {
+								// exists if we have an update for the message at this index
+								const latestUpdate = updates[updates.length - 1]
+
+								// if block was just altered, then calculate savings
+								if (hasNewAlterations) {
+									let originalTextLength
+									if (updates.length > 1) {
+										originalTextLength = updates[updates.length - 2][2][0].length // handles case if we have multiple updates for same text block
+									} else {
+										originalTextLength = block.text.length
+									}
+
+									const newTextLength = latestUpdate[2][0].length // replacement text
+									totalCharactersSaved += originalTextLength - newTextLength
+
+									totalCharCount += originalTextLength
+								} else {
+									// meaning there was an update to this text previously, but we didn't just alter it
+									totalCharCount += latestUpdate[2][0].length
+								}
+							} else {
+								// reach here if there was one inner index with an update, but now we are at a different index, so updates is not defined
+								totalCharCount += block.text.length
+							}
+						} else {
+							// reach here if there's no alterations for this outer index, meaning each inner index won't have any changes either
+							totalCharCount += block.text.length
+						}
+					} else if (block.type === "image" && block.source) {
+						if (block.source.type === "base64" && block.source.data) {
+							totalCharCount += block.source.data.length
+						}
+					}
+				}
+			}
+		}
+
+		return { totalCharacters: totalCharCount, charactersSaved: totalCharactersSaved }
+	}
+
+	/**
+	 * count total percentage character savings across in-range conversation
+	 */
+	private calculateContextOptimizationMetrics(
+		apiMessages: Anthropic.Messages.MessageParam[],
+		conversationHistoryDeletedRange: [number, number] | undefined,
+		uniqueFileReadIndices: Set<number>,
+	): number {
+		// count for first user-assistant message pair
+		const firstChunkResult = this.countCharactersAndSavingsInRange(apiMessages, 0, 2, uniqueFileReadIndices)
+
+		// count for the remaining in-range messages
+		const secondChunkResult = this.countCharactersAndSavingsInRange(
+			apiMessages,
+			conversationHistoryDeletedRange ? conversationHistoryDeletedRange[1] + 1 : 2,
+			apiMessages.length,
+			uniqueFileReadIndices,
+		)
+
+		const totalCharacters = firstChunkResult.totalCharacters + secondChunkResult.totalCharacters
+		const totalCharactersSaved = firstChunkResult.charactersSaved + secondChunkResult.charactersSaved
+
+		const percentCharactersSaved = totalCharacters === 0 ? 0 : totalCharactersSaved / totalCharacters
+
+		return percentCharactersSaved
+	}
+}

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 +1,175 @@
+import path from "path"
+import { ensureRulesDirectoryExists, GlobalFileNames } from "../../../storage/disk"
+import { fileExistsAtPath, isDirectory, readDirectory } from "../../../../utils/fs"
+import { formatResponse } from "../../../prompts/responses"
+import fs from "fs/promises"
+import { ClineRulesToggles } from "../../../../shared/cline-rules"
+import { getGlobalState, getWorkspaceState, updateGlobalState, updateWorkspaceState } from "../../../storage/state"
+import * as vscode from "vscode"
+
+export const getGlobalClineRules = async (globalClineRulesFilePath: string, toggles: ClineRulesToggles) => {
+	if (await fileExistsAtPath(globalClineRulesFilePath)) {
+		if (await isDirectory(globalClineRulesFilePath)) {
+			try {
+				const rulesFilePaths = await readDirectory(globalClineRulesFilePath)
+				const rulesFilesTotalContent = await getClineRulesFilesTotalContent(
+					rulesFilePaths,
+					globalClineRulesFilePath,
+					toggles,
+				)
+				if (rulesFilesTotalContent) {
+					const clineRulesFileInstructions = formatResponse.clineRulesGlobalDirectoryInstructions(
+						globalClineRulesFilePath,
+						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, toggles: ClineRulesToggles) => {
+	const clineRulesFilePath = path.resolve(cwd, GlobalFileNames.clineRules)
+
+	let clineRulesFileInstructions: string | undefined
+
+	if (await fileExistsAtPath(clineRulesFilePath)) {
+		if (await isDirectory(clineRulesFilePath)) {
+			try {
+				const rulesFilePaths = await readDirectory(clineRulesFilePath)
+				const rulesFilesTotalContent = await getClineRulesFilesTotalContent(rulesFilePaths, cwd, toggles)
+				if (rulesFilesTotalContent) {
+					clineRulesFileInstructions = formatResponse.clineRulesLocalDirectoryInstructions(cwd, rulesFilesTotalContent)
+				}
+			} catch {
+				console.error(`Failed to read .clinerules directory at ${clineRulesFilePath}`)
+			}
+		} else {
+			try {
+				if (clineRulesFilePath in toggles && toggles[clineRulesFilePath] !== false) {
+					const ruleFileContent = (await fs.readFile(clineRulesFilePath, "utf8")).trim()
+					if (ruleFileContent) {
+						clineRulesFileInstructions = formatResponse.clineRulesLocalFileInstructions(cwd, ruleFileContent)
+					}
+				}
+			} catch {
+				console.error(`Failed to read .clinerules file at ${clineRulesFilePath}`)
+			}
+		}
+	}
+
+	return clineRulesFileInstructions
+}
+
+const getClineRulesFilesTotalContent = async (rulesFilePaths: string[], basePath: string, toggles: ClineRulesToggles) => {
+	const ruleFilesTotalContent = await Promise.all(
+		rulesFilePaths.map(async (filePath) => {
+			const ruleFilePath = path.resolve(basePath, filePath)
+			const ruleFilePathRelative = path.relative(basePath, ruleFilePath)
+
+			if (ruleFilePath in toggles && toggles[ruleFilePath] === false) {
+				return null
+			}
+
+			return `${ruleFilePathRelative}\n` + (await fs.readFile(ruleFilePath, "utf8")).trim()
+		}),
+	).then((contents) => contents.filter(Boolean).join("\n\n"))
+	return ruleFilesTotalContent
+}
+
+export async function synchronizeRuleToggles(
+	rulesDirectoryPath: string,
+	currentToggles: ClineRulesToggles,
+): Promise<ClineRulesToggles> {
+	// Create a copy of toggles to modify
+	const updatedToggles = { ...currentToggles }
+
+	try {
+		const pathExists = await fileExistsAtPath(rulesDirectoryPath)
+
+		if (pathExists) {
+			const isDir = await isDirectory(rulesDirectoryPath)
+
+			if (isDir) {
+				// DIRECTORY CASE
+				const filePaths = await readDirectory(rulesDirectoryPath)
+				const existingRulePaths = new Set<string>()
+
+				for (const filePath of filePaths) {
+					const ruleFilePath = path.resolve(rulesDirectoryPath, filePath)
+					existingRulePaths.add(ruleFilePath)
+
+					const pathHasToggle = ruleFilePath in updatedToggles
+					if (!pathHasToggle) {
+						updatedToggles[ruleFilePath] = true
+					}
+				}
+
+				// Clean up toggles for non-existent files
+				for (const togglePath in updatedToggles) {
+					const pathExists = existingRulePaths.has(togglePath)
+					if (!pathExists) {
+						delete updatedToggles[togglePath]
+					}
+				}
+			} else {
+				// FILE CASE
+				// Add toggle for this file
+				const pathHasToggle = rulesDirectoryPath in updatedToggles
+				if (!pathHasToggle) {
+					updatedToggles[rulesDirectoryPath] = true
+				}
+
+				// Remove toggles for any other paths
+				for (const togglePath in updatedToggles) {
+					if (togglePath !== rulesDirectoryPath) {
+						delete updatedToggles[togglePath]
+					}
+				}
+			}
+		} else {
+			// PATH DOESN'T EXIST CASE
+			// Clear all toggles since the path doesn't exist
+			for (const togglePath in updatedToggles) {
+				delete updatedToggles[togglePath]
+			}
+		}
+	} catch (error) {
+		console.error(`Failed to synchronize rule toggles for path: ${rulesDirectoryPath}`, error)
+	}
+
+	return updatedToggles
+}
+
+export async function refreshClineRulesToggles(
+	context: vscode.ExtensionContext,
+	workingDirectory: string,
+): Promise<{
+	globalToggles: ClineRulesToggles
+	localToggles: ClineRulesToggles
+}> {
+	// Global toggles
+	const globalClineRulesToggles = ((await getGlobalState(context, "globalClineRulesToggles")) as ClineRulesToggles) || {}
+	const globalClineRulesFilePath = await ensureRulesDirectoryExists()
+	const updatedGlobalToggles = await synchronizeRuleToggles(globalClineRulesFilePath, globalClineRulesToggles)
+	await updateGlobalState(context, "globalClineRulesToggles", updatedGlobalToggles)
+
+	// Local toggles
+	const localClineRulesToggles = ((await getWorkspaceState(context, "localClineRulesToggles")) as ClineRulesToggles) || {}
+	const localClineRulesFilePath = path.resolve(workingDirectory, GlobalFileNames.clineRules)
+	const updatedLocalToggles = await synchronizeRuleToggles(localClineRulesFilePath, localClineRulesToggles)
+	await updateWorkspaceState(context, "localClineRulesToggles", updatedLocalToggles)
+
+	return {
+		globalToggles: updatedGlobalToggles,
+		localToggles: updatedLocalToggles,
+	}
+}

src/core/controller/browser/getBrowserConnectionInfo.ts

@@ -0,0 +1,47 @@
+import { BrowserConnectionInfo } from "../../../shared/proto/browser"
+import { EmptyRequest } from "../../../shared/proto/common"
+import { Controller } from "../index"
+import { getAllExtensionState } from "../../storage/state"
+
+/**
+ * Get information about the current browser connection
+ * @param controller The controller instance
+ * @param request The request message
+ * @returns The browser connection info
+ */
+export async function getBrowserConnectionInfo(controller: Controller, request: EmptyRequest): Promise<BrowserConnectionInfo> {
+	try {
+		// Get browser settings from extension state
+		const { browserSettings } = await getAllExtensionState(controller.context)
+
+		// Check if there's an active browser session by using the controller's handleWebviewMessage approach
+		// This is similar to what's done in controller/index.ts for the "getBrowserConnectionInfo" message
+		if (controller.task?.browserSession) {
+			// Access the browser session through the controller's task property
+			// Using indexer notation to access private property
+			const browserSession = controller.task.browserSession
+			const connectionInfo = browserSession.getConnectionInfo()
+
+			// Convert from BrowserSession.BrowserConnectionInfo to proto.BrowserConnectionInfo
+			return {
+				isConnected: connectionInfo.isConnected,
+				isRemote: connectionInfo.isRemote,
+				host: connectionInfo.host || "", // Ensure host is never undefined
+			}
+		}
+
+		// Fallback to browser settings if no active browser session
+		return {
+			isConnected: false,
+			isRemote: !!browserSettings.remoteBrowserEnabled,
+			host: browserSettings.remoteBrowserHost || "",
+		}
+	} catch (error: unknown) {
+		console.error("Error getting browser connection info:", error)
+		return {
+			isConnected: false,
+			isRemote: false,
+			host: "",
+		}
+	}
+}

src/core/controller/browser/index.ts

@@ -0,0 +1,15 @@
+import { createServiceRegistry, ServiceMethodHandler } from "../grpc-service"
+import { registerAllMethods } from "./methods"
+
+// Create browser service registry
+const browserService = createServiceRegistry("browser")
+
+// Export the method handler type and registration function
+export type BrowserMethodHandler = ServiceMethodHandler
+export const registerMethod = browserService.registerMethod
+
+// Export the request handler
+export const handleBrowserServiceRequest = browserService.handleRequest
+
+// Register all browser methods
+registerAllMethods()

src/core/controller/browser/methods.ts

@@ -0,0 +1,14 @@
+// AUTO-GENERATED FILE - DO NOT MODIFY DIRECTLY
+// Generated by proto/build-proto.js
+
+// Import all method implementations
+import { registerMethod } from "./index"
+import { getBrowserConnectionInfo } from "./getBrowserConnectionInfo"
+import { testBrowserConnection } from "./testBrowserConnection"
+
+// Register all browser service methods
+export function registerAllMethods(): void {
+	// Register each method with the registry
+	registerMethod("getBrowserConnectionInfo", getBrowserConnectionInfo)
+	registerMethod("testBrowserConnection", testBrowserConnection)
+}

src/core/controller/browser/testBrowserConnection.ts

@@ -0,0 +1,63 @@
+import { BrowserConnection } from "../../../shared/proto/browser"
+import { StringRequest } from "../../../shared/proto/common"
+import { Controller } from "../index"
+import { getAllExtensionState } from "../../storage/state"
+import { BrowserSession } from "../../../services/browser/BrowserSession"
+import { discoverChromeInstances } from "../../../services/browser/BrowserDiscovery"
+
+/**
+ * Test connection to a browser instance
+ * @param controller The controller instance
+ * @param request The request message
+ * @returns The browser connection result
+ */
+export async function testBrowserConnection(controller: Controller, request: StringRequest): Promise<BrowserConnection> {
+	try {
+		const { browserSettings } = await getAllExtensionState(controller.context)
+		const browserSession = new BrowserSession(controller.context, browserSettings)
+		const text = request.value || ""
+
+		// If no text is provided, try auto-discovery
+		if (!text) {
+			try {
+				const discoveredHost = await discoverChromeInstances()
+				if (discoveredHost) {
+					// Test the connection to the discovered host
+					const result = await browserSession.testConnection(discoveredHost)
+					return {
+						success: result.success,
+						message: `Auto-discovered and tested connection to Chrome at ${discoveredHost}: ${result.message}`,
+						endpoint: result.endpoint || "",
+					}
+				} else {
+					return {
+						success: false,
+						message:
+							"No Chrome instances found on the network. Make sure Chrome is running with remote debugging enabled (--remote-debugging-port=9222).",
+						endpoint: "",
+					}
+				}
+			} catch (error) {
+				return {
+					success: false,
+					message: `Error during auto-discovery: ${error instanceof Error ? error.message : String(error)}`,
+					endpoint: "",
+				}
+			}
+		} else {
+			// Test the provided URL
+			const result = await browserSession.testConnection(text)
+			return {
+				success: result.success,
+				message: result.message,
+				endpoint: result.endpoint || "",
+			}
+		}
+	} catch (error) {
+		return {
+			success: false,
+			message: `Error testing connection: ${error instanceof Error ? error.message : String(error)}`,
+			endpoint: "",
+		}
+	}
+}

src/core/controller/grpc-handler.ts

@@ -0,0 +1,81 @@
+import { Controller } from "./index"
+import { handleBrowserServiceRequest } from "./browser/index"
+import { ExtensionMessage } from "../../shared/ExtensionMessage"
+
+/**
+ * Handles gRPC requests from the webview
+ */
+export class GrpcHandler {
+	constructor(private controller: Controller) {}
+
+	/**
+	 * Handle a gRPC request from the webview
+	 * @param service The service name
+	 * @param method The method name
+	 * @param message The request message
+	 * @param requestId The request ID for response correlation
+	 * @returns The response message or error
+	 */
+	async handleRequest(
+		service: string,
+		method: string,
+		message: any,
+		requestId: string,
+	): Promise<{
+		message?: any
+		error?: string
+		request_id: string
+	}> {
+		try {
+			// Handle BrowserService requests
+			if (service === "cline.BrowserService") {
+				return {
+					message: await handleBrowserServiceRequest(this.controller, method, message),
+					request_id: requestId,
+				}
+			}
+
+			throw new Error(`Unknown service: ${service}`)
+		} catch (error) {
+			return {
+				error: error instanceof Error ? error.message : String(error),
+				request_id: requestId,
+			}
+		}
+	}
+}
+
+/**
+ * Handle a gRPC request from the webview
+ * @param controller The controller instance
+ * @param request The gRPC request
+ */
+export async function handleGrpcRequest(
+	controller: Controller,
+	request: {
+		service: string
+		method: string
+		message: any
+		request_id: string
+	},
+) {
+	try {
+		const grpcHandler = new GrpcHandler(controller)
+		const response = await grpcHandler.handleRequest(request.service, request.method, request.message, request.request_id)
+
+		// Send the response back to the webview
+		await controller.postMessageToWebview({
+			type: "grpc_response",
+			grpc_response: response,
+		})
+	} catch (error) {
+		// Send error response
+		await controller.postMessageToWebview({
+			type: "grpc_response",
+			grpc_response: {
+				error: error instanceof Error ? error.message : String(error),
+				request_id: request.request_id,
+			},
+		})
+	}
+}

src/core/controller/grpc-service.ts

@@ -0,0 +1,65 @@
+import { Controller } from "./index"
+
+/**
+ * Generic type for service method handlers
+ */
+export type ServiceMethodHandler = (controller: Controller, message: any) => Promise<any>
+
+/**
+ * Generic service registry for gRPC services
+ */
+export class ServiceRegistry {
+	private serviceName: string
+	private methodRegistry: Record<string, ServiceMethodHandler> = {}
+
+	/**
+	 * Create a new service registry
+	 * @param serviceName The name of the service (used for logging)
+	 */
+	constructor(serviceName: string) {
+		this.serviceName = serviceName
+	}
+
+	/**
+	 * Register a method handler
+	 * @param methodName The name of the method to register
+	 * @param handler The handler function for the method
+	 */
+	registerMethod(methodName: string, handler: ServiceMethodHandler): void {
+		this.methodRegistry[methodName] = handler
+		console.log(`Registered ${this.serviceName} method: ${methodName}`)
+	}
+
+	/**
+	 * Handle a service request
+	 * @param controller The controller instance
+	 * @param method The method name
+	 * @param message The request message
+	 * @returns The response message
+	 */
+	async handleRequest(controller: Controller, method: string, message: any): Promise<any> {
+		const handler = this.methodRegistry[method]
+
+		if (!handler) {
+			throw new Error(`Unknown ${this.serviceName} method: ${method}`)
+		}
+
+		return handler(controller, message)
+	}
+}
+
+/**
+ * Create a service registry factory function
+ * @param serviceName The name of the service
+ * @returns An object with register and handle functions
+ */
+export function createServiceRegistry(serviceName: string) {
+	const registry = new ServiceRegistry(serviceName)
+
+	return {
+		registerMethod: (methodName: string, handler: ServiceMethodHandler) => registry.registerMethod(methodName, handler),
+
+		handleRequest: (controller: Controller, method: string, message: any) =>
+			registry.handleRequest(controller, method, message),
+	}
+}

src/core/controller/index.ts

@@ -1,13 +1,13 @@
 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"
+import { handleGrpcRequest } from "./grpc-handler"
 import { buildApiHandler } from "../../api"
 import { cleanupLegacyCheckpoints } from "../../integrations/checkpoints/CheckpointMigration"
 import { downloadTask } from "../../integrations/misc/export-markdown"
@@ -17,10 +17,12 @@ 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"
@@ -30,22 +32,24 @@ 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,
 	getSecret,
+	getWorkspaceState,
 	resetExtensionState,
 	storeSecret,
 	updateApiConfiguration,
 	updateGlobalState,
+	updateWorkspaceState,
 } from "../storage/state"
-import { WebviewProvider } from "../webview"
-import { GlobalFileNames } from "../storage/disk"
-import { searchWorkspaceFiles } from "../../services/search/file-search"
-import { getWorkspacePath } from "../../utils/path"
+import { Task, cwd } from "../task"
+import { ClineRulesToggles } from "../../shared/cline-rules"
+import { refreshClineRulesToggles } from "../context/instructions/user-instructions/cline-rules"
 
 /*
 https://github.com/microsoft/vscode-webview-ui-toolkit-samples/blob/main/default/weather-webview/src/providers/WeatherViewProvider.ts
@@ -54,25 +58,37 @@ 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>
+	task?: Task
+	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) => {
@@ -95,11 +111,8 @@ 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")
@@ -122,12 +135,19 @@ 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,
@@ -135,29 +155,20 @@ 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)
 	}
 
 	/**
@@ -248,6 +259,13 @@ export class Controller {
 					telemetryService.updateTelemetryState(isOptedIn)
 				})
 				break
+			case "showChatView": {
+				this.postMessageToWebview({
+					type: "action",
+					action: "chatButtonClicked",
+				})
+				break
+			}
 			case "newTask":
 				// Code that should run in response to the hello message command
 				//vscode.window.showInformationMessage(message.text!)
@@ -257,7 +275,7 @@ 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) {
@@ -279,6 +297,12 @@ export class Controller {
 				break
 			case "browserSettings":
 				if (message.browserSettings) {
+					// remoteBrowserEnabled now means "enable remote browser connection"
+					// commenting out since this is being done in BrowserSettingsSection updateRemoteBrowserEnabled
+					// if (!message.browserSettings.remoteBrowserEnabled) {
+					// 	// If disabling remote browser connection, clear the remoteBrowserHost
+					// 	message.browserSettings.remoteBrowserHost = undefined
+					// }
 					await updateGlobalState(this.context, "browserSettings", message.browserSettings)
 					if (this.task) {
 						this.task.browserSettings = message.browserSettings
@@ -287,6 +311,41 @@ export class Controller {
 					await this.postStateToWebview()
 				}
 				break
+			case "discoverBrowser":
+				try {
+					const discoveredHost = await discoverChromeInstances()
+
+					if (discoveredHost) {
+						// Don't update the remoteBrowserHost state when auto-discovering
+						// This way we don't override the user's preference
+
+						// Test the connection to get the endpoint
+						const { browserSettings } = await getAllExtensionState(this.context)
+						const browserSession = new BrowserSession(this.context, browserSettings)
+						const result = await browserSession.testConnection(discoveredHost)
+
+						// Send the result back to the webview
+						await this.postMessageToWebview({
+							type: "browserConnectionResult",
+							success: true,
+							text: `Successfully discovered and connected to Chrome at ${discoveredHost}`,
+							endpoint: result.endpoint,
+						})
+					} else {
+						await this.postMessageToWebview({
+							type: "browserConnectionResult",
+							success: false,
+							text: "No Chrome instances found on the network. Make sure Chrome is running with remote debugging enabled (--remote-debugging-port=9222).",
+						})
+					}
+				} catch (error) {
+					await this.postMessageToWebview({
+						type: "browserConnectionResult",
+						success: false,
+						text: `Error discovering browser: ${error instanceof Error ? error.message : String(error)}`,
+					})
+				}
+				break
 			case "togglePlanActMode":
 				if (message.chatSettings) {
 					await this.togglePlanActModeWithChatSettings(message.chatSettings, message.chatContent)
@@ -299,11 +358,11 @@ export class Controller {
 					text: message.text,
 				})
 				break
-			// case "relaunchChromeDebugMode":
-			// 	if (this.task) {
-			// 		this.task.browserSession.relaunchChromeDebugMode()
-			// 	}
-			// 	break
+			case "relaunchChromeDebugMode":
+				const { browserSettings } = await getAllExtensionState(this.context)
+				const browserSession = new BrowserSession(this.context, browserSettings)
+				await browserSession.relaunchChromeDebugMode(this)
+				break
 			case "askResponse":
 				this.task?.handleWebviewAskResponse(message.askResponse!, message.text, message.images)
 				break
@@ -362,11 +421,18 @@ 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)
 				this.postMessageToWebview({ type: "openAiModels", openAiModels })
 				break
+			case "refreshClineRules":
+				await refreshClineRulesToggles(this.context, cwd)
+				await this.postStateToWebview()
+				break
 			case "openImage":
 				openImage(message.text!)
 				break
@@ -394,7 +460,7 @@ export class Controller {
 				break
 			}
 			case "checkpointRestore": {
-				await this.cancelTask() // we cannot alter message history say if the task is active, as it could be in the middle of editing a file or running a command, which expect the ask to be responded to rather than being superceded by a new message eg add deleted_api_reqs
+				await this.cancelTask() // we cannot alter message history say if the task is active, as it could be in the middle of editing a file or running a command, which expect the ask to be responded to rather than being superseded by a new message eg add deleted_api_reqs
 				// cancel task waits for any open editor to be reverted and starts a new cline instance
 				if (message.number) {
 					// wait for messages to be loaded
@@ -404,7 +470,7 @@ export class Controller {
 						console.error("Failed to init new cline instance")
 					})
 					// NOTE: cancelTask awaits abortTask, which awaits diffViewProvider.revertChanges, which reverts any edited files, allowing us to reset to a checkpoint rather than running into a state where the revertChanges function is called alongside or after the checkpoint reset
-					await this.task?.restoreCheckpoint(message.number, message.text! as ClineCheckpointRestore)
+					await this.task?.restoreCheckpoint(message.number, message.text! as ClineCheckpointRestore, message.offset)
 				}
 				break
 			}
@@ -450,7 +516,7 @@ 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": {
@@ -472,14 +538,7 @@ export class Controller {
 						await this.togglePlanActModeWithChatSettings({ mode: "act" })
 					}
 
-					// 2. Enable MCP settings if disabled
-					// Enable MCP mode if disabled
-					const mcpConfig = vscode.workspace.getConfiguration("cline.mcp")
-					if (mcpConfig.get<string>("mode") !== "full") {
-						await mcpConfig.update("mode", "full", true)
-					}
-
-					// 3. download MCP
+					// 2. download MCP
 					await this.downloadMcp(message.mcpId)
 				}
 				break
@@ -551,6 +610,30 @@ export class Controller {
 				}
 				break
 			}
+			case "toggleClineRule": {
+				const { isGlobal, rulePath, enabled } = message
+				if (rulePath && typeof enabled === "boolean" && typeof isGlobal === "boolean") {
+					if (isGlobal) {
+						const toggles =
+							((await getGlobalState(this.context, "globalClineRulesToggles")) as ClineRulesToggles) || {}
+						toggles[rulePath] = enabled
+						await updateGlobalState(this.context, "globalClineRulesToggles", toggles)
+					} else {
+						const toggles =
+							((await getWorkspaceState(this.context, "localClineRulesToggles")) as ClineRulesToggles) || {}
+						toggles[rulePath] = enabled
+						await updateWorkspaceState(this.context, "localClineRulesToggles", toggles)
+					}
+					await this.postStateToWebview()
+				} else {
+					console.error("toggleClineRule: Missing or invalid parameters", {
+						rulePath,
+						isGlobal: typeof isGlobal === "boolean" ? isGlobal : `Invalid: ${typeof isGlobal}`,
+						enabled: typeof enabled === "boolean" ? enabled : `Invalid: ${typeof enabled}`,
+					})
+				}
+				break
+			}
 			case "requestTotalTasksSize": {
 				this.refreshTotalTasksSize()
 				break
@@ -623,6 +706,13 @@ export class Controller {
 				})
 				break
 			}
+			case "scrollToSettings": {
+				await this.postMessageToWebview({
+					type: "scrollToSettings",
+					text: message.text,
+				})
+				break
+			}
 			case "telemetrySetting": {
 				if (message.telemetrySetting) {
 					await this.updateTelemetrySetting(message.telemetrySetting)
@@ -663,6 +753,57 @@ export class Controller {
 				this.postMessageToWebview({ type: "relinquishControl" })
 				break
 			}
+			case "getDetectedChromePath": {
+				try {
+					const { browserSettings } = await getAllExtensionState(this.context)
+					const browserSession = new BrowserSession(this.context, browserSettings)
+					const { path, isBundled } = await browserSession.getDetectedChromePath()
+					await this.postMessageToWebview({
+						type: "detectedChromePath",
+						text: path,
+						isBundled,
+					})
+				} catch (error) {
+					console.error("Error getting detected Chrome path:", error)
+				}
+				break
+			}
+			case "getRelativePaths": {
+				if (message.uris && message.uris.length > 0) {
+					const resolvedPaths = await Promise.all(
+						message.uris.map(async (uriString) => {
+							try {
+								const fileUri = vscode.Uri.parse(uriString, true)
+								const relativePath = vscode.workspace.asRelativePath(fileUri, false)
+
+								if (path.isAbsolute(relativePath)) {
+									console.warn(`Dropped file ${relativePath} is outside the workspace. Sending original path.`)
+									return fileUri.fsPath.replace(/\\/g, "/")
+								} else {
+									let finalPath = "/" + relativePath.replace(/\\/g, "/")
+									try {
+										const stat = await vscode.workspace.fs.stat(fileUri)
+										if (stat.type === vscode.FileType.Directory) {
+											finalPath += "/"
+										}
+									} catch (statError) {
+										console.error(`Error stating file ${fileUri.fsPath}:`, statError)
+									}
+									return finalPath
+								}
+							} catch (error) {
+								console.error(`Error calculating relative path for ${uriString}:`, error)
+								return null
+							}
+						}),
+					)
+					await this.postMessageToWebview({
+						type: "relativePathsResponse",
+						paths: resolvedPaths,
+					})
+				}
+				break
+			}
 			case "searchFiles": {
 				const workspacePath = getWorkspacePath()
 
@@ -706,6 +847,33 @@ 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
+			}
+			case "grpc_request": {
+				if (message.grpc_request) {
+					await handleGrpcRequest(this, message.grpc_request)
+				}
+				break
+			}
 			// Add more switch case statements here as more webview message commands
 			// are created within the webview context (i.e. inside media/main.js)
 		}
@@ -731,6 +899,7 @@ export class Controller {
 			previousModeModelInfo: newModelInfo,
 			previousModeVsCodeLmModelSelector: newVsCodeLmModelSelector,
 			previousModeThinkingBudgetTokens: newThinkingBudgetTokens,
+			previousModeReasoningEffort: newReasoningEffort,
 			planActSeparateModelsSetting,
 		} = await getAllExtensionState(this.context)
 
@@ -740,6 +909,7 @@ 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":
@@ -749,6 +919,7 @@ export class Controller {
 				case "openai-native":
 				case "qwen":
 				case "deepseek":
+				case "xai":
 					await updateGlobalState(this.context, "previousModeModelId", apiConfiguration.apiModelId)
 					break
 				case "openrouter":
@@ -779,13 +950,21 @@ 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":
@@ -795,6 +974,7 @@ export class Controller {
 					case "openai-native":
 					case "qwen":
 					case "deepseek":
+					case "xai":
 						await updateGlobalState(this.context, "apiModelId", newModelId)
 						break
 					case "openrouter":
@@ -820,6 +1000,7 @@ export class Controller {
 						break
 					case "requesty":
 						await updateGlobalState(this.context, "requestyModelId", newModelId)
+						await updateGlobalState(this.context, "requestyModelInfo", newModelInfo)
 						break
 				}
 
@@ -874,7 +1055,7 @@ 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
 		}
 	}
@@ -887,61 +1068,6 @@ 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() {
@@ -1183,15 +1309,17 @@ export class Controller {
 
 			// Create task with context from README and added guidelines for MCP server installation
 			const task = `Set up the MCP server from ${mcpDetails.githubUrl} while adhering to these MCP server installation rules:
+- Start by loading the MCP documentation.
 - Use "${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",
@@ -1235,7 +1363,7 @@ 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 ${apiKey}` }
 			}
@@ -1416,6 +1544,52 @@ Here is the project's README to help you get started:\n\n${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) {
@@ -1479,9 +1653,7 @@ 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)
 	}
@@ -1521,6 +1693,8 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 		taskDirPath: string
 		apiConversationHistoryFilePath: string
 		uiMessagesFilePath: string
+		contextHistoryFilePath: string
+		taskMetadataFilePath: string
 		apiConversationHistory: Anthropic.MessageParam[]
 	}> {
 		const history = ((await getGlobalState(this.context, "taskHistory")) as HistoryItem[] | undefined) || []
@@ -1529,6 +1703,8 @@ Here is the project's README to help you get started:\n\n${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"))
@@ -1537,12 +1713,14 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 					taskDirPath,
 					apiConversationHistoryFilePath,
 					uiMessagesFilePath,
+					contextHistoryFilePath,
+					taskMetadataFilePath,
 					apiConversationHistory,
 				}
 			}
 		}
 		// if we tried to get a task that doesn't exist, remove it from state
-		// FIXME: this seems to happen sometimes when the json file doesnt save to disk for some reason
+		// FIXME: this seems to happen sometimes when the json file doesn't save to disk for some reason
 		await this.deleteTaskFromState(id)
 		throw new Error("Task not found")
 	}
@@ -1551,7 +1729,7 @@ Here is the project's README to help you get started:\n\n${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",
@@ -1608,22 +1786,28 @@ 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
@@ -1668,8 +1852,12 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			mcpMarketplaceEnabled,
 			telemetrySetting,
 			planActSeparateModelsSetting,
+			globalClineRulesToggles,
 		} = await getAllExtensionState(this.context)
 
+		const localClineRulesToggles =
+			((await getWorkspaceState(this.context, "localClineRulesToggles")) as ClineRulesToggles) || {}
+
 		return {
 			version: this.context.extension?.packageJSON?.version ?? "",
 			apiConfiguration,
@@ -1692,6 +1880,8 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 			telemetrySetting,
 			planActSeparateModelsSetting,
 			vscMachineId: vscode.env.machineId,
+			globalClineRulesToggles: globalClineRulesToggles || {},
+			localClineRulesToggles: localClineRulesToggles || {},
 		}
 	}
 
@@ -1713,7 +1903,7 @@ Here is the project's README to help you get started:\n\n${mcpDetails.readmeCont
 	// conversation history to send in API requests
 
 	/*
-	It seems that some API messages do not comply with vscode state requirements. Either the Anthropic library is manipulating these values somehow in the backend in a way thats creating cyclic references, or the API returns a function or a Symbol as part of the message content.
+	It seems that some API messages do not comply with vscode state requirements. Either the Anthropic library is manipulating these values somehow in the backend in a way that's creating cyclic references, or the API returns a function or a Symbol as part of the message content.
 	VSCode docs about state: "The value must be JSON-stringifyable ... value — A value. MUST not contain cyclic references."
 	For now we'll store the conversation history in memory, and if we need to store in state directly we'd need to do a manual conversion to ensure proper json stringification.
 	*/

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) {