merge conflicts

* refactor out apiconfig section

* add general settings section

* duplicate import

* add missing type

.changeset/blue-wasps-destroy.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-Remove the clsx / tailwind merge dependencies and replace with template literals

.changeset/chilled-paws-remain.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Fix ENAMETOOLONG when calling Claude Code

.changeset/friendly-flowers-drum.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-Migrate fetchUserCreditsData to protobus

.changeset/quick-moons-do.md

@@ -1,5 +0,0 @@
----
-"claude-dev": minor
----
-
-Migrate authStateChanged to protobus

.changeset/sharp-singers-tie.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": minor
+---
+
+Support Sonnet-4 in SAP AI Core provider

.changeset/strange-paws-relate.md

@@ -1,5 +0,0 @@
----
-"claude-dev": patch
----
-
-Add taskId as metadata to use from LiteLLM

.changeset/tall-elephants-sort.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Removed deleteNonFavoritedTasks, moved popup to extension, cleaned up deletion logic

.changeset/yellow-gifts-explain.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+host bridge migration - clipboard

.changeset/young-bulldogs-raise.md

@@ -0,0 +1,5 @@
+---
+"claude-dev": patch
+---
+
+Refactor task class, moving auto approve

.clinerules/protobuf-development.md

@@ -0,0 +1,89 @@
+# Cline Protobuf Development Guide
+
+This guide outlines how to add new gRPC endpoints for communication between the webview (frontend) and the extension host (backend).
+
+## Overview
+
+Cline uses [Protobuf](https://protobuf.dev/) to define a strongly-typed API, ensuring efficient and type-safe communication. All definitions are in the `/proto` directory. The compiler and plugins are included as project dependencies, so no manual installation is needed.
+
+## Key Concepts & Best Practices
+
+-   **File Structure**: Each feature domain should have its own `.proto` file (e.g., `account.proto`, `task.proto`).
+-   **Message Design**:
+    -   For simple, single-value data, use the shared types in `proto/common.proto` (e.g., `StringRequest`, `Empty`, `Int64Request`). This promotes consistency.
+    -   For complex data structures, define custom messages within the feature's `.proto` file (see `task.proto` for examples like `NewTaskRequest`).
+-   **Naming Conventions**:
+    -   Services: `PascalCaseService` (e.g., `AccountService`).
+    -   RPCs: `camelCase` (e.g., `accountEmailIdentified`).
+    -   Messages: `PascalCase` (e.g., `StringRequest`).
+-   **Streaming**: For server-to-client streaming, use the `stream` keyword on the response type. See `subscribeToAuthCallback` in `account.proto` for an example.
+
+---
+
+## 4-Step Development Workflow
+
+Here’s how to add a new RPC, using `scrollToSettings` as an example.
+
+### 1. Define the RPC in a `.proto` File
+
+Add your service method to the appropriate file in the `proto/` directory.
+
+**File: `proto/ui.proto`**
+```proto
+service UiService {
+  // ... other RPCs
+  // Scrolls to a specific settings section in the settings view
+  rpc scrollToSettings(StringRequest) returns (KeyValuePair);
+}
+```
+Here, we use the common `StringRequest` and `KeyValuePair` types.
+
+### 2. Compile Definitions
+
+After editing a `.proto` file, regenerate the TypeScript code. From the project root, run:
+```bash
+npm run protos
+```
+This command compiles all `.proto` files and outputs the generated code to `src/generated/` and `src/shared/`. Do not edit these generated files manually.
+
+### 3. Implement the Backend Handler
+
+Create the RPC implementation in the backend. Handlers are located in `src/core/controller/[service-name]/`.
+
+**File: `src/core/controller/ui/scrollToSettings.ts`**
+```typescript
+import { Controller } from ".."
+import { StringRequest, KeyValuePair } from "../../../shared/proto/common"
+
+/**
+ * Executes a scroll to settings action
+ * @param controller The controller instance
+ * @param request The request containing the ID of the settings section to scroll to
+ * @returns KeyValuePair with action and value fields for the UI to process
+ */
+export async function scrollToSettings(controller: Controller, request: StringRequest): Promise<KeyValuePair> {
+	return KeyValuePair.create({
+		key: "scrollToSettings",
+		value: request.value || "",
+	})
+}
+```
+
+### 4. Call the RPC from the Webview
+
+Call the new RPC from a React component in `webview-ui/`. The generated client makes this simple.
+
+**File: `webview-ui/src/components/browser/BrowserSettingsMenu.tsx`** (Example)
+```tsx
+import { UiServiceClient } from "../../../services/grpc"
+import { StringRequest } from "../../../../shared/proto/common"
+
+// ... inside a React component
+const handleMenuClick = async () => {
+    try {
+        await UiServiceClient.scrollToSettings(StringRequest.create({ value: "browser" }))
+    } catch (error) {
+        console.error("Error scrolling to browser settings:", error)
+    }
+}
+```

CHANGELOG.md

@@ -1,5 +1,19 @@
 # Changelog
 
+## [3.18.0]
+
+-   Optimized Cline to work with the Claude 4 family of models, resulting in improved performance, reliability, and new capabilities
+-   Added a new Gemini CLI provider that allows you to use your local Gemini CLI authentication to access Gemini models for free (Thanks @google-gemini!)
+-   Optimized Cline to work with the Gemini 2.5 family of models
+-   Updated the default and recommended model to Claude 4 Sonnet for the best performance
+-   Fix race condition in Plan/Act mode switching
+-   Improve robustness of search and replace parsing
+
+## [3.17.16]
+
+-   Fix Claude Code provider error handling for incomplete messages during long-running tasks (Thanks @BarreiroT!)
+-   Add taskId as metadata to LiteLLM API requests for better request tracing (Thanks @jorgegarciarey!)
+
 ## [3.17.15]
 
 -   Fix LiteLLM provider to properly respect selected model IDs when switching between Plan and Act modes (Thanks @sammcj!)

README.md

@@ -30,7 +30,7 @@ English | <a href="https://github.com/cline/cline/blob/main/locales/es/README.md
 </table>
 </div>
 
-Meet Cline, an AI assistant that can use your **CLI** a**N**d **E**ditor.
+Meet Cline (pronounced /klaɪn/, like "Klein"), an AI assistant that can use your **CLI** a**N**d **E**ditor.
 
 Thanks to [Claude 3.7 Sonnet's agentic coding capabilities](https://www.anthropic.com/claude/sonnet), Cline can handle complex software development tasks step-by-step. With tools that let him create & edit files, explore large projects, use the browser, and execute terminal commands (after you grant permission), he can assist you in ways that go beyond code completion or tech support. Cline can even use the Model Context Protocol (MCP) to create new tools and extend his own capabilities. While autonomous AI scripts traditionally run in sandboxed environments, this extension provides a human-in-the-loop GUI to approve every file change and terminal command, providing a safe and accessible way to explore the potential of agentic AI.
 

docs/docs.json

@@ -146,6 +146,7 @@
 				"group": "Provider Configuration",
 				"pages": [
 					"provider-config/anthropic",
+					"provider-config/claude-code",
 					"provider-config/aws-bedrock-with-credentials-authentication",
 					"provider-config/aws-bedrock-with-profile-authentication",
 					"provider-config/gcp-vertex-ai",

docs/exploring-clines-tools/new-task-tool.mdx

@@ -120,7 +120,7 @@ Example of context window usage over 50% with a 200K context window:
 # Context Window Usage
 
 105,000 / 200,000 tokens (53%)
-Model: anthropic/claude-3.7-sonnet (200K context window)
+Model: anthropic/claude-sonnet-4 (200K context window)
 \`\`\`
 
 **IMPORTANT**: When you see context window usage at or above 50%, you MUST:

docs/more-info/telemetry.mdx

@@ -6,7 +6,7 @@ title: "Telemetry"
 
 To help make Cline better for everyone, we collect anonymous usage data that helps us understand how developers are using our open-source AI coding agent. This feedback loop is crucial for improving Cline's capabilities and user experience.
 
-We use PostHog, an open-source analytics platform, for data collection and analysis. Our telemetry implementation is fully transparent - you can review the [source code](https://github.com/cline/cline/blob/main/src/services/telemetry/TelemetryService.ts) to see exactly what we track.
+We use PostHog, an open-source analytics platform, for data collection and analysis. Our telemetry implementation is fully transparent - you can review the [source code](https://github.com/cline/cline/blob/main/src/services/posthog/telemetry/TelemetryService.ts) to see exactly what we track.
 
 ### Tracking Policy
 
@@ -22,7 +22,7 @@ We collect basic anonymous usage data including:
 **System Context:** OS type and VS Code environment details\
 **UI Activity:** Navigation patterns and feature usage
 
-For complete transparency, you can inspect our [telemetry implementation](https://github.com/cline/cline/blob/main/src/services/telemetry/TelemetryService.ts) to see the exact events we track.
+For complete transparency, you can inspect our [telemetry implementation](https://github.com/cline/cline/blob/main/src/services/posthog/telemetry/TelemetryService.ts) to see the exact events we track.
 
 ### How to Opt Out
 

docs/provider-config/claude-code.mdx

@@ -0,0 +1,86 @@
+---
+title: "Claude Code"
+description: "Use your Claude Max subscription with Cline instead of paying per token. Learn how to set up and configure the Claude Code provider."
+---
+
+**Website:** [https://docs.anthropic.com/en/docs/claude-code/setup](https://docs.anthropic.com/en/docs/claude-code/setup)
+
+The Claude Code provider lets you use your existing Claude subscription with Cline. If you have Claude Max, this means you can use Claude in Cline without paying extra API costs.
+
+<Frame>
+	<img
+		src="https://storage.googleapis.com/cline_public_images/docs/assets/claude-code-use-opus.gif"
+		alt="Using the Claude Code provider in Cline with Opus model"
+	/>
+</Frame>
+
+## Setup
+
+First, you'll need to install and authenticate Claude Code on your system:
+
+1. **Install Claude Code**: Follow Anthropic's [official setup guide](https://docs.anthropic.com/en/docs/claude-code/setup) to install and authenticate the Claude CLI.
+
+2. **Configure in Cline**:
+    - Open Cline settings (⚙️ icon)
+    - Select **Claude Code** from the **API Provider** dropdown
+    - Set the path to your Claude CLI executable (usually just `claude` if it's in your PATH)
+
+<Frame>
+	<img
+		src="https://storage.googleapis.com/cline_public_images/docs/assets/claude-code-setup.gif"
+		alt="Setting up the Claude Code provider in Cline"
+	/>
+</Frame>
+
+### Finding your Claude Code path
+
+If you're not sure where Claude Code is installed:
+
+-   **macOS / Linux**: Run `which claude` in your terminal
+-   **Windows (Command Prompt)**: Run `where claude`
+-   **Windows (PowerShell)**: Run `Get-Command claude`
+
+## Supported Models
+
+The Claude Code provider supports these models:
+
+-   `claude-sonnet-4-20250514` (Recommended)
+-   `claude-opus-4-20250514`
+-   `claude-3-7-sonnet-20250219`
+-   `claude-3-5-sonnet-20241022`
+-   `claude-3-5-haiku-20241022`
+
+## How it works
+
+When you use Claude Code with Cline, here's what happens behind the scenes:
+
+Cline wraps the Claude Code CLI to handle your requests. Each time you send a message, Cline starts a new `claude` process, sends your conversation, and streams the response back. The AI reasoning comes from Claude Code, but all the actual file editing, terminal commands, and other tools are handled by Cline.
+
+The main difference you'll notice is that responses don't stream character-by-character like other providers. Instead, Claude Code processes your full request before sending back the complete response.
+
+## Limitations
+
+There are a few things to keep in mind with Claude Code:
+
+-   Images in your messages get converted to text placeholders since Claude Code doesn't support image uploads through the CLI
+-   Prompt caching isn't available with this provider
+-   Responses don't stream in real-time like other providers
+
+## Troubleshooting
+
+If you run into issues:
+
+**Authentication problems**: Make sure you're logged into Claude Code with your subscription account. Run `claude auth status` to check.
+
+**Path issues**: Double-check that the Claude CLI path in Cline's settings is correct. Try running `claude --version` in your terminal to verify it's working.
+
+**Still having trouble?** We're actively improving this integration. Report issues on our [GitHub](https://github.com/cline/cline/issues) or ask for help in our [Discord](https://discord.gg/cline).
+
+## Usage with subscriptions
+
+If you have a Claude Max subscription, your usage in Cline shows up as $0.00 in the billing interface since you're not paying additional API costs. Your usage still counts against your subscription limits, but you won't see per-token charges.
+
+For more details about using Claude Code with your subscription, check out Anthropic's documentation:
+
+-   [Claude Code Setup Guide](https://docs.anthropic.com/en/docs/claude-code/setup)
+-   [Using Claude Code with Pro/Max Plans](https://support.anthropic.com/en/articles/11145838-using-claude-code-with-your-pro-or-max-plan)

eslint-rules/__tests__/no-vscode-postmessage.test.ts

@@ -0,0 +1,74 @@
+const { RuleTester: VscodeRuleTester } = require("eslint")
+const vscodePostmessageRule = require("../no-vscode-postmessage")
+
+const vscodeRuleTester = new VscodeRuleTester({
+	parser: require.resolve("@typescript-eslint/parser"),
+	parserOptions: {
+		ecmaVersion: 2020,
+		sourceType: "module",
+		ecmaFeatures: {
+			jsx: true,
+		},
+	},
+})
+
+vscodeRuleTester.run("no-vscode-postmessage", vscodePostmessageRule, {
+	valid: [
+		// Should allow vscode.postMessage in grpc-client-base.ts
+		{
+			code: `vscode.postMessage({ type: "grpc_request", data: {} })`,
+			filename: "grpc-client-base.ts",
+		},
+		{
+			code: `vscode.postMessage({ type: "grpc_request_cancel" })`,
+			filename: "/path/to/grpc-client-base.ts",
+		},
+		// Should allow other vscode API calls
+		{
+			code: `vscode.window.showInformationMessage("Hello")`,
+			filename: "test.ts",
+		},
+		// Should allow postMessage calls on other objects
+		{
+			code: `window.postMessage({ type: "test" }, "*")`,
+			filename: "test.ts",
+		},
+		// Should allow variables named vscode but not calling postMessage
+		{
+			code: `const vscode = { other: "method" }; vscode.other()`,
+			filename: "test.ts",
+		},
+	],
+	invalid: [
+		// Should ban vscode.postMessage in regular files
+		{
+			code: `vscode.postMessage({ type: "test", data: {} })`,
+			filename: "test.ts",
+			errors: [
+				{
+					messageId: "useGrpcClient",
+				},
+			],
+		},
+		// Should ban vscode.postMessage in components
+		{
+			code: `vscode.postMessage({ type: "apiConfiguration", apiConfiguration })`,
+			filename: "ApiOptions.tsx",
+			errors: [
+				{
+					messageId: "useGrpcClient",
+				},
+			],
+		},
+		// Should ban vscode.postMessage in test files
+		{
+			code: `vscode.postMessage({ type: "newTask", text: message.text })`,
+			filename: "test.test.ts",
+			errors: [
+				{
+					messageId: "useGrpcClient",
+				},
+			],
+		},
+	],
+})

eslint-rules/index.js

@@ -1,11 +1,13 @@
 // eslint-rules/index.js
 const noProtobufObjectLiterals = require("./no-protobuf-object-literals")
 const noGrpcClientObjectLiterals = require("./no-grpc-client-object-literals")
+const noVscodePostmessage = require("./no-vscode-postmessage")
 
 module.exports = {
 	rules: {
 		"no-protobuf-object-literals": noProtobufObjectLiterals,
 		"no-grpc-client-object-literals": noGrpcClientObjectLiterals,
+		"no-vscode-postmessage": noVscodePostmessage,
 	},
 	configs: {
 		recommended: {
@@ -13,6 +15,7 @@ module.exports = {
 			rules: {
 				"local/no-protobuf-object-literals": "error",
 				"local/no-grpc-client-object-literals": "error",
+				"local/no-vscode-postmessage": "error",
 			},
 		},
 	},

eslint-rules/no-vscode-postmessage.js

@@ -0,0 +1,61 @@
+const { ESLintUtils } = require("@typescript-eslint/utils")
+const path = require("path")
+
+const createRule = ESLintUtils.RuleCreator((name) => `https://cline.bot/eslint-rules/${name}`)
+
+module.exports = createRule({
+	name: "no-vscode-postmessage",
+	meta: {
+		type: "problem",
+		docs: {
+			description: "Ban vscode.postMessage() calls in favor of gRPC service clients, except in grpc-client-base.ts",
+			recommended: "error",
+		},
+		messages: {
+			useGrpcClient:
+				"Use gRPC service clients instead of vscode.postMessage().\n" +
+				"Example: AccountServiceClient.methodName(RequestType.create({...})) instead of vscode.postMessage({type: '...'}).\n" +
+				"Found: {{code}}",
+		},
+		schema: [],
+	},
+	defaultOptions: [],
+
+	create(context) {
+		// Check if current file is grpc-client-base.ts (exception case)
+		const filename = context.filename
+		const isGrpcClientBase = path.basename(filename) === "grpc-client-base.ts"
+
+		return {
+			// Detect vscode.postMessage calls
+			"CallExpression[callee.type='MemberExpression']"(node) {
+				// Skip if this is grpc-client-base.ts
+				if (isGrpcClientBase) {
+					return
+				}
+
+				const callee = node.callee
+
+				// Check for vscode.postMessage pattern
+				if (
+					callee.object &&
+					callee.object.type === "Identifier" &&
+					callee.object.name === "vscode" &&
+					callee.property &&
+					callee.property.name === "postMessage"
+				) {
+					const sourceCode = context.sourceCode
+					const callText = sourceCode.getText(node).trim()
+
+					context.report({
+						node,
+						messageId: "useGrpcClient",
+						data: {
+							code: callText,
+						},
+					})
+				}
+			},
+		}
+	},
+})

evals/.gitignore

@@ -19,4 +19,6 @@ diff_editing/test_outputs/
 .cache
 
 # Python bytecode cache
-*__pycache__/
+*__pycache__/
+
+diff-edits/cases.zip

evals/cli/src/commands/runDiffEval.ts

@@ -14,6 +14,8 @@ interface RunDiffEvalOptions {
 	testPath: string
 	outputPath: string
 	replay: boolean
+	replayRunId?: string
+	diffApplyFile?: string
 	maxCases?: number
 }
 
@@ -56,6 +58,14 @@ export async function runDiffEvalHandler(options: RunDiffEvalOptions) {
 		args.push("--replay")
 	}
 
+	if (options.replayRunId) {
+		args.push("--replay-run-id", options.replayRunId)
+	}
+
+	if (options.diffApplyFile) {
+		args.push("--diff-apply-file", options.diffApplyFile)
+	}
+
 	if (options.verbose) {
 		args.push("--verbose")
 	}

evals/cli/src/index.ts

@@ -93,6 +93,8 @@ program
 	.option("--thinking-budget <tokens>", "Set the thinking tokens budget", "0")
 	.option("--parallel", "Run tests in parallel", false)
 	.option("--replay", "Run evaluation from a pre-recorded LLM output, skipping the API call", false)
+	.option("--replay-run-id <run_id>", "The ID of the run to replay from the database")
+	.option("--diff-apply-file <filename>", "The name of the diff apply file to use for the replay")
 	.option("-v, --verbose", "Enable verbose logging", false)
 	.action(async (options) => {
 		try {

evals/diff-edits/ClineWrapper.ts

@@ -8,8 +8,9 @@ import {
 	parseAssistantMessageV3,
 	AssistantMessageContent,
 } from "./parsing/parse-assistant-message-06-06-25" // "../../src/core/assistant-message"
-import { constructNewFileContent as constructNewFileContentV1, constructNewFileContentV2 } from "./diff-apply/diff-06-06-25"
-import { constructNewFileContent as constructNewFileContentV3 } from "../../src/core/assistant-message/diff" // this defaults to the new v1 when called
+import { constructNewFileContent as constructNewFileContent_06_06_25 } from "./diff-apply/diff-06-06-25"
+import { constructNewFileContent as constructNewFileContent_06_23_25 } from "./diff-apply/diff-06-23-25"
+import { constructNewFileContent as constructNewFileContent_06_25_25 } from "./diff-apply/diff-06-25-25"
 
 type ParseAssistantMessageFn = (message: string) => AssistantMessageContent[]
 type ConstructNewFileContentFn = (diff: string, original: string, strict: boolean) => Promise<string>
@@ -21,9 +22,9 @@ const parsingFunctions: Record<string, ParseAssistantMessageFn> = {
 }
 
 const diffEditingFunctions: Record<string, ConstructNewFileContentFn> = {
-	constructNewFileContentV1: constructNewFileContentV1,
-	constructNewFileContentV2: constructNewFileContentV2,
-	constructNewFileContentV3: constructNewFileContentV3, // position invariant diff
+	"diff-06-06-25": constructNewFileContent_06_06_25,
+	"diff-06-23-25": constructNewFileContent_06_23_25,
+	"diff-06-25-25": constructNewFileContent_06_25_25,
 }
 
 import { TestInput, TestResult, ExtractedToolCall } from "./types"
@@ -151,6 +152,7 @@ export async function runSingleEvaluation(input: TestInput): Promise<TestResult>
 			diffEditFunction,
 			thinkingBudgetTokens,
 			originalDiffEditToolCallMessage,
+			diffApplyFile,
 		} = input
 
 		const requiredParams = {
@@ -176,7 +178,7 @@ export async function runSingleEvaluation(input: TestInput): Promise<TestResult>
 		}
 
 		const parseAssistantMessage = parsingFunctions[parsingFunction]
-		const constructNewFileContent = diffEditingFunctions[diffEditFunction]
+		const constructNewFileContent = diffEditingFunctions[diffApplyFile || diffEditFunction]
 
 		if (!parseAssistantMessage || !constructNewFileContent) {
 			return {

evals/diff-edits/README.md

@@ -63,9 +63,22 @@ For example, if we ask for 5 valid attempts per test case, the system will keep
 
 This way, if we're comparing two models and one gets a 10% success rate on its valid diff edit attempts, and another gets 90%, we have a much clearer picture of their actual diff-generating capabilities. It avoids muddying the waters with attempts where the model didn't even try to perform the specific action we're evaluating. This approach helps us isolate and measure the diff-editing skill more directly, despite the non-deterministic nature of these models.
 
-## Some known edge cases
+## Replays
 
-I noticed that some of the current conversation jsons in the `./cases` folder are a little big bogus. Here's a running list of these areas:
+You can also use the replay argument to replay a previous benchmark run. This is super useful for iterating on our diffing algorithms without having to re-run expensive and time-consuming LLM calls.
 
-- ~~What if the conversation json was using a model with a massive context window, like Google Gemini's 1M context window, and we're now re-rolling that case on a smaller context window model like claude 4 (200k) or grok-3 (128k)? It's def gonna fail, and we shouldn't just keep trying. We should have a smart system for selecting which cases we can use given the arguments passed in. For example, if we pass in claude and grok with `max cases = 2`, we shouldn't just pick the first two jsons in the folder. We should go through, use a tokenizer, and make sure it would fit with some padding like 20k tokens. Use tiktoken. 20k padding will be sufficient even though different models tokenize differently.~~
-- There are some weird jsons, where something weird happen, where essentially there's a fluke. Maybe the user was using an extremely dumb model that just hallucinated a fake filepath. Now when we try to reroll that case, we never get a valid case. This can easily be handled by making sure that the file is present before selecting that eval for testing. By "file is present" I mean, that file_contents is present in the eval. Additionally, we should in the streamlit dashboard show cases where getting a valid attempt is a challenge, so we can review those cases more easily and throw them out if they're bogus. Maybe a special tab/page in the dashboard for this purpose. Across all runs / cases, what the most consistently problematic cases are. Pop one open to see the case formatted json with all the user/assistant turns.
+When you run an evaluation, every detail is stored in the database—including the raw, unmodified output from the model. The replay feature takes advantage of this by pulling that raw output and feeding it into a *different* diffing algorithm. This lets you isolate the performance of the diffing logic itself. We can see if a new algorithm is better at applying the exact same set of diffs that a model generated in a previous run.
+
+This process is blazingly fast and free, as it completely bypasses the need to make new API calls. It ensures a true apples-to-apples comparison between diffing strategies, since the model's output—the "ground truth" for the evaluation—remains identical.
+
+Here’s an example of how you would replay a previous run with a new diffing algorithm:
+
+```shell
+cd evals && npm run diff-eval -- --replay-run-id 9902189e-63a8-4210-a4fc-fe59e2eaf2c2 --diff-apply-file diff-06-23-25 --verbose
+```
+
+In this command:
+-   `--replay-run-id` specifies the original run we want to use as our ground truth.
+-   `--diff-apply-file` tells the script to use the new diffing logic from the `diff-06-23-25.ts` file.
+
+The script will then create a new run in the database that mirrors the original, but with the results of applying the new diffing algorithm. This allows for a direct comparison in the dashboard, helping us quickly see which of our diffing strategies is the most robust.

evals/diff-edits/TestRunner.ts

@@ -1,4 +1,9 @@
 import { runSingleEvaluation, TestInput, TestResult } from "./ClineWrapper"
+import { parseAssistantMessageV2, AssistantMessageContent } from "./parsing/parse-assistant-message-06-06-25"
+import { constructNewFileContent as constructNewFileContent_06_06_25 } from "./diff-apply/diff-06-06-25"
+import { constructNewFileContent as constructNewFileContent_06_23_25 } from "./diff-apply/diff-06-23-25"
+import { constructNewFileContent as constructNewFileContent_06_25_25 } from "./diff-apply/diff-06-25-25"
+import { constructNewFileContent as constructNewFileContentV3 } from "../../src/core/assistant-message/diff"
 import { basicSystemPrompt } from "./prompts/basicSystemPrompt-06-06-25"
 import { claude4SystemPrompt } from "./prompts/claude4SystemPrompt-06-06-25"
 import { formatResponse } from "./helpers"
@@ -18,6 +23,10 @@ import {
 	insertResult,
 	DatabaseClient,
 	CreateResultInput,
+	getResultsByRun,
+	getCaseById,
+	getFileByHash,
+	getBenchmarkRun,
 } from "./database"
 
 // Load environment variables from .env file
@@ -184,6 +193,78 @@ class NodeTestRunner {
 		log(isVerbose, `✓ Created ${this.caseIdMap.size} database case records`);
 	}
 
+	/**
+	 * Store replay result in database, copying original data but with new diffing results
+	 */
+	async storeReplayResultInDatabase(replayResult: TestResult, originalResult: any, testId: string, newCaseId: string): Promise<void> {
+		if (!this.currentRunId || !this.processingFunctionsHash) {
+			return; // Skip if database not initialized
+		}
+
+		try {
+			// Map error string to error enum (simple mapping)
+			const errorEnum = this.mapErrorToEnum(replayResult.error);
+
+			// Store diff edit content if available
+			let fileEditedHash: string | undefined;
+			if (replayResult.diffEdit) {
+				fileEditedHash = await upsertFile({
+					filepath: `diff-edit-${testId}`,
+					content: replayResult.diffEdit
+				});
+			}
+
+			// Calculate basic metrics from diff edit if available
+			let numEdits = 0;
+			let numLinesAdded = 0;
+			let numLinesDeleted = 0;
+			
+			if (replayResult.diffEdit) {
+				// Simple parsing to count edits - count SEARCH/REPLACE blocks
+				const searchBlocks = (replayResult.diffEdit.match(/------- SEARCH/g) || []).length;
+				numEdits = searchBlocks;
+				
+				// Count added/deleted lines (rough approximation)
+				const lines = replayResult.diffEdit.split('\n');
+				for (const line of lines) {
+					if (line.startsWith('+') && !line.startsWith('+++')) {
+						numLinesAdded++;
+					} else if (line.startsWith('-') && !line.startsWith('---')) {
+						numLinesDeleted++;
+					}
+				}
+			}
+
+			// Copy original result data but update replay-specific fields
+			const resultInput: CreateResultInput = {
+				run_id: this.currentRunId, // New run ID
+				case_id: newCaseId, // New case ID
+				model_id: originalResult.model_id, // Copy from original
+				processing_functions_hash: this.processingFunctionsHash, // New processing functions
+				succeeded: replayResult.success && (replayResult.diffEditSuccess ?? false), // New result
+				error_enum: errorEnum, // New error if any
+				num_edits: numEdits || originalResult.num_edits, // New or original
+				num_lines_deleted: numLinesDeleted || originalResult.num_lines_deleted, // New or original
+				num_lines_added: numLinesAdded || originalResult.num_lines_added, // New or original
+				// Copy timing and cost data from original (since we didn't make API calls)
+				time_to_first_token_ms: originalResult.time_to_first_token_ms,
+				time_to_first_edit_ms: originalResult.time_to_first_edit_ms,
+				time_round_trip_ms: originalResult.time_round_trip_ms,
+				cost_usd: originalResult.cost_usd,
+				completion_tokens: originalResult.completion_tokens,
+				// Use original model output (since we're replaying)
+				raw_model_output: originalResult.raw_model_output,
+				file_edited_hash: fileEditedHash || originalResult.file_edited_hash,
+				parsed_tool_call_json: replayResult.toolCalls ? JSON.stringify(replayResult.toolCalls) : originalResult.parsed_tool_call_json
+			};
+
+			await insertResult(resultInput);
+		} catch (error) {
+			console.error(`Failed to store replay result in database for ${testId}:`, error);
+			// Continue execution - don't fail the test run
+		}
+	}
+
 	/**
 	 * Store test result in database
 	 */
@@ -394,6 +475,143 @@ class NodeTestRunner {
 		}
 	}
 
+	async runDatabaseReplay(replayRunId: string, diffApplyFile: string, isVerbose: boolean) {
+		log(isVerbose, `Starting database replay for run_id: ${replayRunId}`)
+		log(isVerbose, `Using diff apply file: ${diffApplyFile}`)
+
+		// 1. Get the correct diffing function
+		const diffEditingFunctions: Record<string, any> = {
+			"diff-06-06-25": constructNewFileContent_06_06_25,
+			"diff-06-23-25": constructNewFileContent_06_23_25,
+			"diff-06-25-25": constructNewFileContent_06_25_25,
+			constructNewFileContentV3: constructNewFileContentV3,
+		}
+		const constructNewFileContent = diffEditingFunctions[diffApplyFile]
+
+		if (!constructNewFileContent) {
+			throw new Error(`Could not find diff apply function for: ${diffApplyFile}`)
+		}
+		log(isVerbose, `Successfully loaded diff apply function: ${diffApplyFile}`)
+
+		// 2. Fetch original run data
+		const originalResults = await getResultsByRun(replayRunId)
+		if (originalResults.length === 0) {
+			throw new Error(`No results found for run_id: ${replayRunId}`)
+		}
+		log(isVerbose, `Found ${originalResults.length} results to replay.`)
+
+		const originalRun = await getBenchmarkRun(replayRunId)
+		if (!originalRun) {
+			throw new Error(`Could not find original run with id ${replayRunId}`)
+		}
+
+		// 3. Create a new benchmark run for the replay
+		const replayRunDescription = `Replay of run ${replayRunId} using ${diffApplyFile}`
+		this.currentRunId = await createBenchmarkRun({
+			description: replayRunDescription,
+			system_prompt_hash: originalRun.system_prompt_hash,
+		})
+		log(isVerbose, `Created new run for replay: ${this.currentRunId}`)
+
+		// 4. Set up processing functions for the new run
+		this.processingFunctionsHash = await upsertProcessingFunctions({
+			name: `replay-${diffApplyFile}`,
+			parsing_function: "parseAssistantMessageV2",
+			diff_edit_function: diffApplyFile,
+		})
+
+		// 5. Process each result from the original run
+		let replayedCount = 0
+		const caseIdMirror: Map<string, string> = new Map()
+
+		for (const originalResult of originalResults) {
+			// 5a. Basic validation to ensure we can even process this
+			if (!originalResult.case_id) {
+				log(isVerbose, `Skipping result ${originalResult.result_id} due to missing case_id.`)
+				continue
+			}
+
+			// 5b. Mirror the case for the new run, reusing if already created
+			let newCaseId = caseIdMirror.get(originalResult.case_id)
+			if (!newCaseId) {
+				const originalCase = await getCaseById(originalResult.case_id)
+				if (!originalCase) {
+					log(isVerbose, `Skipping result ${originalResult.result_id} because original case could not be found.`)
+					continue
+				}
+				newCaseId = await createCase({
+					run_id: this.currentRunId,
+					description: `Replay of case ${originalCase.case_id} from run ${replayRunId}`,
+					system_prompt_hash: originalCase.system_prompt_hash,
+					task_id: originalCase.task_id,
+					tokens_in_context: originalCase.tokens_in_context,
+					file_hash: originalCase.file_hash,
+				})
+				caseIdMirror.set(originalResult.case_id, newCaseId)
+			}
+
+			// 5c. Determine if the original attempt was a "valid attempt"
+			const isValidOriginalAttempt = originalResult.error_enum === null || originalResult.error_enum === 3 // 3 is diff_edit_error
+
+			const newResultInput: CreateResultInput = {
+				...(originalResult as any),
+				run_id: this.currentRunId,
+				case_id: newCaseId,
+				processing_functions_hash: this.processingFunctionsHash,
+			}
+			delete (newResultInput as any).result_id
+
+			if (isValidOriginalAttempt) {
+				// This was a valid attempt. Re-run the diff algorithm.
+				const originalCase = await getCaseById(originalResult.case_id)
+				if (!originalCase) {
+					log(isVerbose, `  [WARN] Replay for result ${originalResult.result_id}: Could not find original case. Copying original result.`)
+					newResultInput.succeeded = originalResult.succeeded
+					newResultInput.error_enum = originalResult.error_enum
+				} else {
+					const originalFile = originalCase.file_hash ? await getFileByHash(originalCase.file_hash) : null
+					const parsedToolCall = originalResult.parsed_tool_call_json ? JSON.parse(originalResult.parsed_tool_call_json)[0] : null
+					const diffContent = parsedToolCall?.input?.diff
+
+					if (originalFile && diffContent) {
+						let diffSuccess = false
+						try {
+							await constructNewFileContent(diffContent, originalFile.content, true)
+							diffSuccess = true
+							log(isVerbose, `  [OK] Replay for task ${originalCase.task_id}: Diff applied successfully.`)
+						} catch (e) {
+							diffSuccess = false
+							log(isVerbose, `  [FAIL] Replay for task ${originalCase.task_id}: New diff algorithm failed.`)
+						}
+						newResultInput.succeeded = diffSuccess
+						newResultInput.error_enum = diffSuccess ? undefined : 3 // 3 = diff_edit_error
+					} else {
+						// Something is wrong with the ground truth data, just copy it.
+						log(
+							isVerbose,
+							`  [WARN] Replay for task ${originalCase.task_id}: Valid original attempt but missing file or diff content. Copying original result.`,
+						)
+						newResultInput.succeeded = originalResult.succeeded
+						newResultInput.error_enum = originalResult.error_enum
+					}
+				}
+			} else {
+				// This was not a valid attempt. Just copy the original result's outcome.
+				log(isVerbose, `  [SKIP] Replay for task ${originalResult.case_id}: Invalid original attempt. Copying original result.`)
+				newResultInput.succeeded = originalResult.succeeded
+				newResultInput.error_enum = originalResult.error_enum
+			}
+
+			await insertResult(newResultInput)
+			replayedCount++
+		}
+
+		log(isVerbose, `\n✓ Database replay completed successfully.`)
+		log(isVerbose, `  Total original results: ${originalResults.length}`)
+		log(isVerbose, `  Total replayed results: ${replayedCount}`)
+		log(isVerbose, `  New run ID: ${this.currentRunId}`)
+	}
+
 	/**
 	 * Run a single test example
 	 */
@@ -420,6 +638,7 @@ class NodeTestRunner {
 			diffEditFunction: testConfig.diff_edit_function,
 			thinkingBudgetTokens: testConfig.thinking_tokens_budget,
 			originalDiffEditToolCallMessage: testConfig.replay ? testCase.original_diff_edit_tool_call_message : undefined,
+			diffApplyFile: testConfig.diff_apply_file,
 		}
 
 		if (isVerbose) {
@@ -712,6 +931,8 @@ async function main() {
 		.option("--thinking-budget <tokens>", "Set the thinking tokens budget", "0")
 		.option("--parallel", "Run tests in parallel", false)
 		.option("--replay", "Run evaluation from a pre-recorded LLM output, skipping the API call", false)
+		.option("--replay-run-id <run_id>", "The ID of the run to replay from the database")
+		.option("--diff-apply-file <filename>", "The name of the diff apply file to use for the replay")
 		.option("-v, --verbose", "Enable verbose logging", false)
 		.option("--max-concurrency <number>", "Maximum number of parallel requests", "80")
 
@@ -732,6 +953,17 @@ async function main() {
 	}
 
 	const validAttemptsPerCase = parseInt(options.validAttemptsPerCase, 10);
+
+	const runner = new NodeTestRunner(options.replay || !!options.replayRunId)
+
+	if (options.replayRunId) {
+		if (!options.diffApplyFile) {
+			console.error("Error: --diff-apply-file is required when using --replay-run-id")
+			process.exit(1)
+		}
+		await runner.runDatabaseReplay(options.replayRunId, options.diffApplyFile, isVerbose)
+		return
+	}
 	
 	try {
 		const startTime = Date.now()

evals/diff-edits/dashboard/app.py

@@ -420,21 +420,33 @@ def render_model_comparison_cards(model_performance):
                 metric_col1, metric_col2, metric_col3, metric_col4 = st.columns(4)
                 
                 with metric_col1:
-                    st.metric("Avg Latency", f"{model['avg_round_trip_ms']:.0f}ms")
+                    if pd.notna(model['avg_round_trip_ms']):
+                        st.metric("Avg Latency", f"{model['avg_round_trip_ms']:.0f}ms")
+                    else:
+                        st.metric("Avg Latency", "N/A")
                 
                 with metric_col2:
-                    st.metric("Avg Cost", f"${model['avg_cost']:.4f}")
+                    if pd.notna(model['avg_cost']):
+                        st.metric("Avg Cost", f"${model['avg_cost']:.4f}")
+                    else:
+                        st.metric("Avg Cost", "N/A")
                 
                 with metric_col3:
                     st.metric("Valid Results", f"{model['total_results']}")
                 
                 with metric_col4:
-                    st.metric("First Token", f"{model['avg_first_token_ms']:.0f}ms")
+                    if pd.notna(model['avg_first_token_ms']):
+                        st.metric("First Token", f"{model['avg_first_token_ms']:.0f}ms")
+                    else:
+                        st.metric("First Token", "N/A")
             
             with col2:
                 st.write("")  # Add some spacing
                 if st.button(f"Drill Down", key=f"drill_{model['model_id']}", use_container_width=True):
                     st.session_state.drill_down_model = model['model_id']
+                    # Update URL with model_id for drill down
+                    st.query_params["model_id"] = model['model_id']
+                    st.rerun()
             
             st.divider()  # Add a divider between models
 
@@ -835,6 +847,11 @@ def main():
     if 'selected_run_id' not in st.session_state:
         st.session_state.selected_run_id = None
     
+    # Handle URL parameters for direct linking
+    query_params = st.query_params
+    url_run_id = query_params.get("run_id")
+    url_model_id = query_params.get("model_id")
+    
     # Load all runs for sidebar
     all_runs = load_all_runs()
     
@@ -842,6 +859,18 @@ def main():
         st.error("No evaluation runs found in the database.")
         st.stop()
     
+    # Set initial run selection from URL or default to latest
+    if url_run_id and url_run_id in all_runs['run_id'].values:
+        if st.session_state.selected_run_id != url_run_id:
+            st.session_state.selected_run_id = url_run_id
+            st.session_state.drill_down_model = None  # Reset drill down when changing runs via URL
+    elif st.session_state.selected_run_id is None:
+        st.session_state.selected_run_id = all_runs.iloc[0]['run_id']  # Default to latest
+    
+    # Set drill down model from URL
+    if url_model_id and st.session_state.selected_run_id == url_run_id:
+        st.session_state.drill_down_model = url_model_id
+    
     # Sidebar for run selection
     with st.sidebar:
         st.markdown("## 📊 Evaluation Runs")
@@ -887,6 +916,10 @@ def main():
         if run_ids[selected_run_idx] != st.session_state.selected_run_id:
             st.session_state.selected_run_id = run_ids[selected_run_idx]
             st.session_state.drill_down_model = None  # Reset drill down when changing runs
+            # Update URL with new run_id
+            st.query_params["run_id"] = st.session_state.selected_run_id
+            if "model_id" in st.query_params:
+                del st.query_params["model_id"]  # Clear model_id when changing runs
             st.rerun()
         
         # Show run details in sidebar
@@ -897,6 +930,57 @@ def main():
         st.markdown(f"**Created:** {selected_run['created_at']}")
         if selected_run['description']:
             st.markdown(f"**Description:** {selected_run['description']}")
+        
+        # Show shareable URL
+        st.markdown("---")
+        st.markdown("### 🔗 Share This View")
+        
+        # Build current URL
+        # Dynamically derive the base URL
+        server_address = st.server.server_address if hasattr(st.server, 'server_address') else "localhost"
+        server_port = st.server.server_port if hasattr(st.server, 'server_port') else "8501"
+        base_url = f"http://{server_address}:{server_port}"
+        current_url = f"{base_url}/?run_id={st.session_state.selected_run_id}"
+        if st.session_state.drill_down_model:
+            current_url += f"&model_id={st.session_state.drill_down_model}"
+        
+        st.markdown("**Current URL:**")
+        st.code(current_url, language=None)
+        
+        # Copy button using HTML/JS
+        copy_button_html = f"""
+        <button onclick="copyToClipboard('{current_url}')" style="
+            padding: 8px 16px; 
+            border-radius: 5px; 
+            border: 1px solid #ccc; 
+            background: #f0f2f6;
+            cursor: pointer;
+            font-size: 14px;
+            margin-top: 5px;
+        ">📋 Copy Link</button>
+        <script>
+            function copyToClipboard(text) {{
+                navigator.clipboard.writeText(text).then(function() {{
+                    // Success feedback
+                    event.target.innerText = '✅ Copied!';
+                    event.target.style.backgroundColor = '#d4edda';
+                    setTimeout(() => {{ 
+                        event.target.innerText = '📋 Copy Link'; 
+                        event.target.style.backgroundColor = '#f0f2f6';
+                    }}, 2000);
+                }}, function(err) {{
+                    // Error feedback
+                    event.target.innerText = '❌ Failed';
+                    event.target.style.backgroundColor = '#f8d7da';
+                    setTimeout(() => {{ 
+                        event.target.innerText = '📋 Copy Link'; 
+                        event.target.style.backgroundColor = '#f0f2f6';
+                    }}, 2000);
+                }});
+            }}
+        </script>
+        """
+        st.components.v1.html(copy_button_html, height=50)
     
     # Load data for selected run
     current_run, model_performance = load_run_comparison(st.session_state.selected_run_id)
@@ -914,6 +998,9 @@ def main():
         with col1:
             if st.button("Back to Overview", use_container_width=True):
                 st.session_state.drill_down_model = None
+                # Clear model_id from URL when going back to overview
+                if "model_id" in st.query_params:
+                    del st.query_params["model_id"]
                 st.rerun()
         
         render_detailed_analysis(current_run['run_id'], st.session_state.drill_down_model)

evals/diff-edits/database.md

@@ -33,7 +33,7 @@ This is the most granular and important table in the database. A `result` repres
 
 -   **Purpose**: To store the detailed outcome of every single model attempt, providing the raw data for all quantitative and qualitative analysis.
 -   **Key Columns**:
-    -   `result_id`: A unique identifier for the individual attempt.
+    -   `result_id`: The primary key for the result.
     -   `run_id`, `case_id`, `model_id`, `processing_functions_hash`: A set of foreign keys that precisely situate this result within the context of a specific run, case, model, and set of helper functions.
     -   `succeeded`: A boolean indicating if the generated diff was applied successfully.
     -   `error_enum`: A numeric code representing the specific type of error if the attempt failed (e.g., `1` for `no_tool_calls`, `7` for `wrong_file_edited`).
@@ -82,3 +82,15 @@ This relational schema provides a powerful foundation for sophisticated analysis
 -   "Which of our internal diffing algorithms is the most robust against a wide range of model-generated edits?"
 
 Ultimately, this data model enables us to move from simply *measuring* performance to truly *understanding* it, providing the insights needed to build more capable and reliable AI engineering systems.
+
+---
+
+## Viewing the Full Schema
+
+To see the most up-to-date and detailed schema for the database, you can use the `sqlite3` command-line tool. From the `evals/diff-edits` directory, run the following command:
+
+```bash
+sqlite3 evals.db .schema
+```
+
+This will print the complete `CREATE TABLE` statements for all tables in the database, providing a definitive reference for the database structure.

evals/diff-edits/diff-apply/diff-06-23-25.ts

@@ -0,0 +1,827 @@
+const SEARCH_BLOCK_START = "------- SEARCH"
+const SEARCH_BLOCK_END = "======="
+const REPLACE_BLOCK_END = "+++++++ REPLACE"
+
+const SEARCH_BLOCK_CHAR = "-"
+const REPLACE_BLOCK_CHAR = "+"
+const LEGACY_SEARCH_BLOCK_CHAR = "<"
+const LEGACY_REPLACE_BLOCK_CHAR = ">"
+
+// Replace the exact string constants with flexible regex patterns
+const SEARCH_BLOCK_START_REGEX = /^[-]{3,} SEARCH$/
+const SEARCH_BLOCK_END_REGEX = /^[=]{3,}$/
+const REPLACE_BLOCK_END_REGEX = /^[+]{3,} REPLACE$/
+const LEGACY_SEARCH_BLOCK_START_REGEX = /^[<]{3,} SEARCH$/
+const LEGACY_REPLACE_BLOCK_END_REGEX = /^[>]{3,} REPLACE$/
+
+// Helper functions to check if a line matches the flexible patterns
+function isSearchBlockStart(line: string): boolean {
+	return SEARCH_BLOCK_START_REGEX.test(line) || LEGACY_SEARCH_BLOCK_START_REGEX.test(line)
+}
+
+function isSearchBlockEnd(line: string): boolean {
+	return SEARCH_BLOCK_END_REGEX.test(line)
+}
+
+function isReplaceBlockEnd(line: string): boolean {
+	return REPLACE_BLOCK_END_REGEX.test(line) || LEGACY_REPLACE_BLOCK_END_REGEX.test(line)
+}
+
+/**
+ * Attempts a line-trimmed fallback match for the given search content in the original content.
+ * It tries to match `searchContent` lines against a block of lines in `originalContent` starting
+ * from `lastProcessedIndex`. Lines are matched by trimming leading/trailing whitespace and ensuring
+ * they are identical afterwards.
+ *
+ * Returns [matchIndexStart, matchIndexEnd] if found, or false if not found.
+ */
+function lineTrimmedFallbackMatch(originalContent: string, searchContent: string, startIndex: number): [number, number] | false {
+	// Split both contents into lines
+	const originalLines = originalContent.split("\n")
+	const searchLines = searchContent.split("\n")
+
+	// Trim trailing empty line if exists (from the trailing \n in searchContent)
+	if (searchLines[searchLines.length - 1] === "") {
+		searchLines.pop()
+	}
+
+	// Find the line number where startIndex falls
+	let startLineNum = 0
+	let currentIndex = 0
+	while (currentIndex < startIndex && startLineNum < originalLines.length) {
+		currentIndex += originalLines[startLineNum].length + 1 // +1 for \n
+		startLineNum++
+	}
+
+	// For each possible starting position in original content
+	for (let i = startLineNum; i <= originalLines.length - searchLines.length; i++) {
+		let matches = true
+
+		// Try to match all search lines from this position
+		for (let j = 0; j < searchLines.length; j++) {
+			const originalTrimmed = originalLines[i + j].trim()
+			const searchTrimmed = searchLines[j].trim()
+
+			if (originalTrimmed !== searchTrimmed) {
+				matches = false
+				break
+			}
+		}
+
+		// If we found a match, calculate the exact character positions
+		if (matches) {
+			// Find start character index
+			let matchStartIndex = 0
+			for (let k = 0; k < i; k++) {
+				matchStartIndex += originalLines[k].length + 1 // +1 for \n
+			}
+
+			// Find end character index
+			let matchEndIndex = matchStartIndex
+			for (let k = 0; k < searchLines.length; k++) {
+				matchEndIndex += originalLines[i + k].length + 1 // +1 for \n
+			}
+
+			return [matchStartIndex, matchEndIndex]
+		}
+	}
+
+	return false
+}
+
+/**
+ * Attempts to match blocks of code by using the first and last lines as anchors.
+ * This is a third-tier fallback strategy that helps match blocks where we can identify
+ * the correct location by matching the beginning and end, even if the exact content
+ * differs slightly.
+ *
+ * The matching strategy:
+ * 1. Only attempts to match blocks of 3 or more lines to avoid false positives
+ * 2. Extracts from the search content:
+ *    - First line as the "start anchor"
+ *    - Last line as the "end anchor"
+ * 3. For each position in the original content:
+ *    - Checks if the next line matches the start anchor
+ *    - If it does, jumps ahead by the search block size
+ *    - Checks if that line matches the end anchor
+ *    - All comparisons are done after trimming whitespace
+ *
+ * This approach is particularly useful for matching blocks of code where:
+ * - The exact content might have minor differences
+ * - The beginning and end of the block are distinctive enough to serve as anchors
+ * - The overall structure (number of lines) remains the same
+ *
+ * @param originalContent - The full content of the original file
+ * @param searchContent - The content we're trying to find in the original file
+ * @param startIndex - The character index in originalContent where to start searching
+ * @returns A tuple of [startIndex, endIndex] if a match is found, false otherwise
+ */
+function blockAnchorFallbackMatch(originalContent: string, searchContent: string, startIndex: number): [number, number] | false {
+	const originalLines = originalContent.split("\n")
+	const searchLines = searchContent.split("\n")
+
+	// Only use this approach for blocks of 3+ lines
+	if (searchLines.length < 3) {
+		return false
+	}
+
+	// Trim trailing empty line if exists
+	if (searchLines[searchLines.length - 1] === "") {
+		searchLines.pop()
+	}
+
+	const firstLineSearch = searchLines[0].trim()
+	const lastLineSearch = searchLines[searchLines.length - 1].trim()
+	const searchBlockSize = searchLines.length
+
+	// Find the line number where startIndex falls
+	let startLineNum = 0
+	let currentIndex = 0
+	while (currentIndex < startIndex && startLineNum < originalLines.length) {
+		currentIndex += originalLines[startLineNum].length + 1
+		startLineNum++
+	}
+
+	// Look for matching start and end anchors
+	for (let i = startLineNum; i <= originalLines.length - searchBlockSize; i++) {
+		// Check if first line matches
+		if (originalLines[i].trim() !== firstLineSearch) {
+			continue
+		}
+
+		// Check if last line matches at the expected position
+		if (originalLines[i + searchBlockSize - 1].trim() !== lastLineSearch) {
+			continue
+		}
+
+		// Calculate exact character positions
+		let matchStartIndex = 0
+		for (let k = 0; k < i; k++) {
+			matchStartIndex += originalLines[k].length + 1
+		}
+
+		let matchEndIndex = matchStartIndex
+		for (let k = 0; k < searchBlockSize; k++) {
+			matchEndIndex += originalLines[i + k].length + 1
+		}
+
+		return [matchStartIndex, matchEndIndex]
+	}
+
+	return false
+}
+
+/**
+ * This function reconstructs the file content by applying a streamed diff (in a
+ * specialized SEARCH/REPLACE block format) to the original file content. It is designed
+ * to handle both incremental updates and the final resulting file after all chunks have
+ * been processed.
+ *
+ * The diff format is a custom structure that uses three markers to define changes:
+ *
+ *   ------- SEARCH
+ *   [Exact content to find in the original file]
+ *   =======
+ *   [Content to replace with]
+ *   +++++++ REPLACE
+ *
+ * Behavior and Assumptions:
+ * 1. The file is processed chunk-by-chunk. Each chunk of `diffContent` may contain
+ *    partial or complete SEARCH/REPLACE blocks. By calling this function with each
+ *    incremental chunk (with `isFinal` indicating the last chunk), the final reconstructed
+ *    file content is produced.
+ *
+ * 2. Matching Strategy (in order of attempt):
+ *    a. Exact Match: First attempts to find the exact SEARCH block text in the original file
+ *    b. Line-Trimmed Match: Falls back to line-by-line comparison ignoring leading/trailing whitespace
+ *    c. Block Anchor Match: For blocks of 3+ lines, tries to match using first/last lines as anchors
+ *    If all matching strategies fail, an error is thrown.
+ *
+ * 3. Empty SEARCH Section:
+ *    - If SEARCH is empty and the original file is empty, this indicates creating a new file
+ *      (pure insertion).
+ *    - If SEARCH is empty and the original file is not empty, this indicates a complete
+ *      file replacement (the entire original content is considered matched and replaced).
+ *
+ * 4. Applying Changes:
+ *    - Before encountering the "=======" marker, lines are accumulated as search content.
+ *    - After "=======" and before ">>>>>>> REPLACE", lines are accumulated as replacement content.
+ *    - Once the block is complete (">>>>>>> REPLACE"), the matched section in the original
+ *      file is replaced with the accumulated replacement lines, and the position in the original
+ *      file is advanced.
+ *
+ * 5. Incremental Output:
+ *    - As soon as the match location is found and we are in the REPLACE section, each new
+ *      replacement line is appended to the result so that partial updates can be viewed
+ *      incrementally.
+ *
+ * 6. Partial Markers:
+ *    - If the final line of the chunk looks like it might be part of a marker but is not one
+ *      of the known markers, it is removed. This prevents incomplete or partial markers
+ *      from corrupting the output.
+ *
+ * 7. Finalization:
+ *    - Once all chunks have been processed (when `isFinal` is true), any remaining original
+ *      content after the last replaced section is appended to the result.
+ *    - Trailing newlines are not forcibly added. The code tries to output exactly what is specified.
+ *
+ * Errors:
+ * - If the search block cannot be matched using any of the available matching strategies,
+ *   an error is thrown.
+ */
+export async function constructNewFileContent(
+	diffContent: string,
+	originalContent: string,
+	isFinal: boolean,
+	version: "v1" | "v2" = "v1",
+): Promise<string> {
+	const constructor = constructNewFileContentVersionMapping[version]
+	if (!constructor) {
+		throw new Error(`Invalid version '${version}' for file content constructor`)
+	}
+	return constructor(diffContent, originalContent, isFinal)
+}
+
+const constructNewFileContentVersionMapping: Record<
+	string,
+	(diffContent: string, originalContent: string, isFinal: boolean) => Promise<string>
+> = {
+	v1: constructNewFileContentV1,
+	v2: constructNewFileContentV2,
+} as const
+
+async function constructNewFileContentV1(diffContent: string, originalContent: string, isFinal: boolean): Promise<string> {
+	let result = ""
+	let lastProcessedIndex = 0
+
+	let currentSearchContent = ""
+	let currentReplaceContent = ""
+	let inSearch = false
+	let inReplace = false
+
+	let searchMatchIndex = -1
+	let searchEndIndex = -1
+
+	// Track all replacements to handle out-of-order edits
+	let replacements: Array<{ start: number; end: number; content: string }> = []
+	let pendingOutOfOrderReplacement = false
+
+	let lines = diffContent.split("\n")
+
+	// If the last line looks like a partial marker but isn't recognized,
+	// remove it because it might be incomplete.
+	const lastLine = lines[lines.length - 1]
+	if (
+		lines.length > 0 &&
+		(lastLine.startsWith(SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith("=") ||
+			lastLine.startsWith(REPLACE_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_REPLACE_BLOCK_CHAR)) &&
+		!isSearchBlockStart(lastLine) &&
+		!isSearchBlockEnd(lastLine) &&
+		!isReplaceBlockEnd(lastLine)
+	) {
+		lines.pop()
+	}
+
+	for (const line of lines) {
+		if (isSearchBlockStart(line)) {
+			inSearch = true
+			currentSearchContent = ""
+			currentReplaceContent = ""
+			continue
+		}
+
+		if (isSearchBlockEnd(line)) {
+			inSearch = false
+			inReplace = true
+
+			// Remove trailing linebreak for adding the === marker
+			// if (currentSearchContent.endsWith("\r\n")) {
+			// 	currentSearchContent = currentSearchContent.slice(0, -2)
+			// } else if (currentSearchContent.endsWith("\n")) {
+			// 	currentSearchContent = currentSearchContent.slice(0, -1)
+			// }
+
+			if (!currentSearchContent) {
+				// Empty search block
+				if (originalContent.length === 0) {
+					// New file scenario: nothing to match, just start inserting
+					searchMatchIndex = 0
+					searchEndIndex = 0
+				} else {
+					// ERROR: Empty search block with non-empty file indicates malformed SEARCH marker
+					throw new Error(
+						"Empty SEARCH block detected with non-empty file. This usually indicates a malformed SEARCH marker.\n" +
+							"Please ensure your SEARCH marker follows the correct format:\n" +
+							"- Use '------- SEARCH' (7+ dashes + space + SEARCH)\n",
+					)
+				}
+			} else {
+				// Add check for inefficient full-file search
+				// if (currentSearchContent.trim() === originalContent.trim()) {
+				// 	throw new Error(
+				// 		"The SEARCH block contains the entire file content. Please either:\n" +
+				// 			"1. Use an empty SEARCH block to replace the entire file, or\n" +
+				// 			"2. Make focused changes to specific parts of the file that need modification.",
+				// 	)
+				// }
+
+				// Exact search match scenario
+				const exactIndex = originalContent.indexOf(currentSearchContent, lastProcessedIndex)
+				if (exactIndex !== -1) {
+					searchMatchIndex = exactIndex
+					searchEndIndex = exactIndex + currentSearchContent.length
+				} else {
+					// Attempt fallback line-trimmed matching
+					const lineMatch = lineTrimmedFallbackMatch(originalContent, currentSearchContent, lastProcessedIndex)
+					if (lineMatch) {
+						;[searchMatchIndex, searchEndIndex] = lineMatch
+					} else {
+						// Try block anchor fallback for larger blocks
+						const blockMatch = blockAnchorFallbackMatch(originalContent, currentSearchContent, lastProcessedIndex)
+						if (blockMatch) {
+							;[searchMatchIndex, searchEndIndex] = blockMatch
+						} else {
+							// Last resort: search the entire file from the beginning
+							const fullFileIndex = originalContent.indexOf(currentSearchContent, 0)
+							if (fullFileIndex !== -1) {
+								// Found in the file - could be out of order
+								searchMatchIndex = fullFileIndex
+								searchEndIndex = fullFileIndex + currentSearchContent.length
+								if (searchMatchIndex < lastProcessedIndex) {
+									pendingOutOfOrderReplacement = true
+								}
+							} else {
+								throw new Error(
+									`The SEARCH block:\n${currentSearchContent.trimEnd()}\n...does not match anything in the file.`,
+								)
+							}
+						}
+					}
+				}
+			}
+
+			// Check if this is an out-of-order replacement
+			if (searchMatchIndex < lastProcessedIndex) {
+				pendingOutOfOrderReplacement = true
+			}
+
+			// For in-order replacements, output everything up to the match location
+			if (!pendingOutOfOrderReplacement) {
+				result += originalContent.slice(lastProcessedIndex, searchMatchIndex)
+			}
+			continue
+		}
+
+		if (isReplaceBlockEnd(line)) {
+			// Finished one replace block
+
+			// Store this replacement
+			replacements.push({
+				start: searchMatchIndex,
+				end: searchEndIndex,
+				content: currentReplaceContent,
+			})
+
+			// If this was an in-order replacement, advance lastProcessedIndex
+			if (!pendingOutOfOrderReplacement) {
+				lastProcessedIndex = searchEndIndex
+			}
+
+			// Reset for next block
+			inSearch = false
+			inReplace = false
+			currentSearchContent = ""
+			currentReplaceContent = ""
+			searchMatchIndex = -1
+			searchEndIndex = -1
+			pendingOutOfOrderReplacement = false
+			continue
+		}
+
+		// Accumulate content for search or replace
+		// (currentReplaceContent is not being used for anything right now since we directly append to result.)
+		// (We artificially add a linebreak since we split on \n at the beginning. In order to not include a trailing linebreak in the final search/result blocks we need to remove it before using them. This allows for partial line matches to be correctly identified.)
+		// NOTE: search/replace blocks must be arranged in the order they appear in the file due to how we build the content using lastProcessedIndex. We also cannot strip the trailing newline since for non-partial lines it would remove the linebreak from the original content. (If we remove end linebreak from search, then we'd also have to remove it from replace but we can't know if it's a partial line or not since the model may be using the line break to indicate the end of the block rather than as part of the search content.) We require the model to output full lines in order for our fallbacks to work as well.
+		if (inSearch) {
+			currentSearchContent += line + "\n"
+		} else if (inReplace) {
+			currentReplaceContent += line + "\n"
+			// Only output replacement lines immediately for in-order replacements
+			if (searchMatchIndex !== -1 && !pendingOutOfOrderReplacement) {
+				result += line + "\n"
+			}
+		}
+	}
+
+	// If this is the final chunk, we need to apply all replacements and build the final result
+	if (isFinal) {
+		// Handle the case where we're still in replace mode when processing ends
+		// and this is the final chunk - treat it as if we encountered the REPLACE marker
+		if (inReplace && searchMatchIndex !== -1) {
+			// Store this replacement
+			replacements.push({
+				start: searchMatchIndex,
+				end: searchEndIndex,
+				content: currentReplaceContent,
+			})
+
+			// If this was an in-order replacement, advance lastProcessedIndex
+			if (!pendingOutOfOrderReplacement) {
+				lastProcessedIndex = searchEndIndex
+			}
+
+			// Reset state
+			inSearch = false
+			inReplace = false
+			currentSearchContent = ""
+			currentReplaceContent = ""
+			searchMatchIndex = -1
+			searchEndIndex = -1
+			pendingOutOfOrderReplacement = false
+		}
+		// end of handling missing replace marker
+
+		// Sort replacements by start position
+		replacements.sort((a, b) => a.start - b.start)
+
+		// Rebuild the entire result by applying all replacements
+		result = ""
+		let currentPos = 0
+
+		for (const replacement of replacements) {
+			// Add original content up to this replacement
+			result += originalContent.slice(currentPos, replacement.start)
+			// Add the replacement content
+			result += replacement.content
+			// Move position to after the replaced section
+			currentPos = replacement.end
+		}
+
+		// Add any remaining original content
+		result += originalContent.slice(currentPos)
+	}
+
+	return result
+}
+
+enum ProcessingState {
+	Idle = 0,
+	StateSearch = 1 << 0,
+	StateReplace = 1 << 1,
+}
+
+class NewFileContentConstructor {
+	private originalContent: string
+	private isFinal: boolean
+	private state: number
+	private pendingNonStandardLines: string[]
+	private result: string
+	private lastProcessedIndex: number
+	private currentSearchContent: string
+	private currentReplaceContent: string
+	private searchMatchIndex: number
+	private searchEndIndex: number
+
+	constructor(originalContent: string, isFinal: boolean) {
+		this.originalContent = originalContent
+		this.isFinal = isFinal
+		this.pendingNonStandardLines = []
+		this.result = ""
+		this.lastProcessedIndex = 0
+		this.state = ProcessingState.Idle
+		this.currentSearchContent = ""
+		this.currentReplaceContent = ""
+		this.searchMatchIndex = -1
+		this.searchEndIndex = -1
+	}
+
+	private resetForNextBlock() {
+		// Reset for next block
+		this.state = ProcessingState.Idle
+		this.currentSearchContent = ""
+		this.currentReplaceContent = ""
+		this.searchMatchIndex = -1
+		this.searchEndIndex = -1
+	}
+
+	private findLastMatchingLineIndex(regx: RegExp, lineLimit: number) {
+		for (let i = lineLimit; i > 0; ) {
+			i--
+			if (this.pendingNonStandardLines[i].match(regx)) {
+				return i
+			}
+		}
+		return -1
+	}
+
+	private updateProcessingState(newState: ProcessingState) {
+		const isValidTransition =
+			(this.state === ProcessingState.Idle && newState === ProcessingState.StateSearch) ||
+			(this.state === ProcessingState.StateSearch && newState === ProcessingState.StateReplace)
+
+		if (!isValidTransition) {
+			throw new Error(
+				`Invalid state transition.\n` +
+					"Valid transitions are:\n" +
+					"- Idle → StateSearch\n" +
+					"- StateSearch → StateReplace",
+			)
+		}
+
+		this.state |= newState
+	}
+
+	private isStateActive(state: ProcessingState): boolean {
+		return (this.state & state) === state
+	}
+
+	private activateReplaceState() {
+		this.updateProcessingState(ProcessingState.StateReplace)
+	}
+
+	private activateSearchState() {
+		this.updateProcessingState(ProcessingState.StateSearch)
+		this.currentSearchContent = ""
+		this.currentReplaceContent = ""
+	}
+
+	private isSearchingActive(): boolean {
+		return this.isStateActive(ProcessingState.StateSearch)
+	}
+
+	private isReplacingActive(): boolean {
+		return this.isStateActive(ProcessingState.StateReplace)
+	}
+
+	private hasPendingNonStandardLines(pendingNonStandardLineLimit: number): boolean {
+		return this.pendingNonStandardLines.length - pendingNonStandardLineLimit < this.pendingNonStandardLines.length
+	}
+
+	public processLine(line: string) {
+		this.internalProcessLine(line, true, this.pendingNonStandardLines.length)
+	}
+
+	public getResult() {
+		// If this is the final chunk, append any remaining original content
+		if (this.isFinal && this.lastProcessedIndex < this.originalContent.length) {
+			this.result += this.originalContent.slice(this.lastProcessedIndex)
+		}
+		if (this.isFinal && this.state !== ProcessingState.Idle) {
+			throw new Error("File processing incomplete - SEARCH/REPLACE operations still active during finalization")
+		}
+		return this.result
+	}
+
+	private internalProcessLine(
+		line: string,
+		canWritependingNonStandardLines: boolean,
+		pendingNonStandardLineLimit: number,
+	): number {
+		let removeLineCount = 0
+		if (isSearchBlockStart(line)) {
+			removeLineCount = this.trimPendingNonStandardTrailingEmptyLines(pendingNonStandardLineLimit)
+			if (removeLineCount > 0) {
+				pendingNonStandardLineLimit = pendingNonStandardLineLimit - removeLineCount
+			}
+			if (this.hasPendingNonStandardLines(pendingNonStandardLineLimit)) {
+				this.tryFixSearchReplaceBlock(pendingNonStandardLineLimit)
+				canWritependingNonStandardLines && (this.pendingNonStandardLines.length = 0)
+			}
+			this.activateSearchState()
+		} else if (isSearchBlockEnd(line)) {
+			// 校验非标内容
+			if (!this.isSearchingActive()) {
+				this.tryFixSearchBlock(pendingNonStandardLineLimit)
+				canWritependingNonStandardLines && (this.pendingNonStandardLines.length = 0)
+			}
+			this.activateReplaceState()
+			this.beforeReplace()
+		} else if (isReplaceBlockEnd(line)) {
+			if (!this.isReplacingActive()) {
+				this.tryFixReplaceBlock(pendingNonStandardLineLimit)
+				canWritependingNonStandardLines && (this.pendingNonStandardLines.length = 0)
+			}
+			this.lastProcessedIndex = this.searchEndIndex
+			this.resetForNextBlock()
+		} else {
+			// Accumulate content for search or replace
+			// (currentReplaceContent is not being used for anything right now since we directly append to result.)
+			// (We artificially add a linebreak since we split on \n at the beginning. In order to not include a trailing linebreak in the final search/result blocks we need to remove it before using them. This allows for partial line matches to be correctly identified.)
+			// NOTE: search/replace blocks must be arranged in the order they appear in the file due to how we build the content using lastProcessedIndex. We also cannot strip the trailing newline since for non-partial lines it would remove the linebreak from the original content. (If we remove end linebreak from search, then we'd also have to remove it from replace but we can't know if it's a partial line or not since the model may be using the line break to indicate the end of the block rather than as part of the search content.) We require the model to output full lines in order for our fallbacks to work as well.
+			if (this.isReplacingActive()) {
+				this.currentReplaceContent += line + "\n"
+				// Output replacement lines immediately if we know the insertion point
+				if (this.searchMatchIndex !== -1) {
+					this.result += line + "\n"
+				}
+			} else if (this.isSearchingActive()) {
+				this.currentSearchContent += line + "\n"
+			} else {
+				let appendToPendingNonStandardLines = canWritependingNonStandardLines
+				if (appendToPendingNonStandardLines) {
+					// 处理非标内容
+					this.pendingNonStandardLines.push(line)
+				}
+			}
+		}
+		return removeLineCount
+	}
+
+	private beforeReplace() {
+		// Remove trailing linebreak for adding the === marker
+		// if (currentSearchContent.endsWith("\r\n")) {
+		// 	currentSearchContent = currentSearchContent.slice(0, -2)
+		// } else if (currentSearchContent.endsWith("\n")) {
+		// 	currentSearchContent = currentSearchContent.slice(0, -1)
+		// }
+
+		if (!this.currentSearchContent) {
+			// Empty search block
+			if (this.originalContent.length === 0) {
+				// New file scenario: nothing to match, just start inserting
+				this.searchMatchIndex = 0
+				this.searchEndIndex = 0
+			} else {
+				// Complete file replacement scenario: treat the entire file as matched
+				this.searchMatchIndex = 0
+				this.searchEndIndex = this.originalContent.length
+			}
+		} else {
+			// Add check for inefficient full-file search
+			// if (currentSearchContent.trim() === originalContent.trim()) {
+			// 	throw new Error(
+			// 		"The SEARCH block contains the entire file content. Please either:\n" +
+			// 			"1. Use an empty SEARCH block to replace the entire file, or\n" +
+			// 			"2. Make focused changes to specific parts of the file that need modification.",
+			// 	)
+			// }
+			// Exact search match scenario
+			const exactIndex = this.originalContent.indexOf(this.currentSearchContent, this.lastProcessedIndex)
+			if (exactIndex !== -1) {
+				this.searchMatchIndex = exactIndex
+				this.searchEndIndex = exactIndex + this.currentSearchContent.length
+			} else {
+				// Attempt fallback line-trimmed matching
+				const lineMatch = lineTrimmedFallbackMatch(
+					this.originalContent,
+					this.currentSearchContent,
+					this.lastProcessedIndex,
+				)
+				if (lineMatch) {
+					;[this.searchMatchIndex, this.searchEndIndex] = lineMatch
+				} else {
+					// Try block anchor fallback for larger blocks
+					const blockMatch = blockAnchorFallbackMatch(
+						this.originalContent,
+						this.currentSearchContent,
+						this.lastProcessedIndex,
+					)
+					if (blockMatch) {
+						;[this.searchMatchIndex, this.searchEndIndex] = blockMatch
+					} else {
+						throw new Error(
+							`The SEARCH block:\n${this.currentSearchContent.trimEnd()}\n...does not match anything in the file.`,
+						)
+					}
+				}
+			}
+		}
+		if (this.searchMatchIndex < this.lastProcessedIndex) {
+			throw new Error(
+				`The SEARCH block:\n${this.currentSearchContent.trimEnd()}\n...matched an incorrect content in the file.`,
+			)
+		}
+		// Output everything up to the match location
+		this.result += this.originalContent.slice(this.lastProcessedIndex, this.searchMatchIndex)
+	}
+
+	private tryFixSearchBlock(lineLimit: number): number {
+		let removeLineCount = 0
+		if (lineLimit < 0) {
+			lineLimit = this.pendingNonStandardLines.length
+		}
+		if (!lineLimit) {
+			throw new Error("Invalid SEARCH/REPLACE block structure - no lines available to process")
+		}
+		let searchTagRegexp = /^([-]{3,}|[<]{3,}) SEARCH$/
+		const searchTagIndex = this.findLastMatchingLineIndex(searchTagRegexp, lineLimit)
+		if (searchTagIndex !== -1) {
+			let fixLines = this.pendingNonStandardLines.slice(searchTagIndex, lineLimit)
+			fixLines[0] = SEARCH_BLOCK_START
+			for (const line of fixLines) {
+				removeLineCount += this.internalProcessLine(line, false, searchTagIndex)
+			}
+		} else {
+			throw new Error(
+				`Invalid REPLACE marker detected - could not find matching SEARCH block starting from line ${searchTagIndex + 1}`,
+			)
+		}
+		return removeLineCount
+	}
+
+	private tryFixReplaceBlock(lineLimit: number): number {
+		let removeLineCount = 0
+		if (lineLimit < 0) {
+			lineLimit = this.pendingNonStandardLines.length
+		}
+		if (!lineLimit) {
+			throw new Error()
+		}
+		let replaceBeginTagRegexp = /^[=]{3,}$/
+		const replaceBeginTagIndex = this.findLastMatchingLineIndex(replaceBeginTagRegexp, lineLimit)
+		if (replaceBeginTagIndex !== -1) {
+			// // 校验非标内容
+			// if (!this.isSearchingActive()) {
+			// 	removeLineCount += this.tryFixSearchBlock(replaceBeginTagIndex)
+			// }
+			let fixLines = this.pendingNonStandardLines.slice(replaceBeginTagIndex - removeLineCount, lineLimit - removeLineCount)
+			fixLines[0] = SEARCH_BLOCK_END
+			for (const line of fixLines) {
+				removeLineCount += this.internalProcessLine(line, false, replaceBeginTagIndex - removeLineCount)
+			}
+		} else {
+			throw new Error(`Malformed REPLACE block - missing valid separator after line ${replaceBeginTagIndex + 1}`)
+		}
+		return removeLineCount
+	}
+
+	private tryFixSearchReplaceBlock(lineLimit: number): number {
+		let removeLineCount = 0
+		if (lineLimit < 0) {
+			lineLimit = this.pendingNonStandardLines.length
+		}
+		if (!lineLimit) {
+			throw new Error()
+		}
+
+		let replaceEndTagRegexp = /^([+]{3,}|[>]{3,}) REPLACE$/
+		const replaceEndTagIndex = this.findLastMatchingLineIndex(replaceEndTagRegexp, lineLimit)
+		const likeReplaceEndTag = replaceEndTagIndex === lineLimit - 1
+		if (likeReplaceEndTag) {
+			// // 校验非标内容
+			// if (!this.isReplacingActive()) {
+			// 	removeLineCount += this.tryFixReplaceBlock(replaceEndTagIndex)
+			// }
+			let fixLines = this.pendingNonStandardLines.slice(replaceEndTagIndex - removeLineCount, lineLimit - removeLineCount)
+			fixLines[fixLines.length - 1] = REPLACE_BLOCK_END
+			for (const line of fixLines) {
+				removeLineCount += this.internalProcessLine(line, false, replaceEndTagIndex - removeLineCount)
+			}
+		} else {
+			throw new Error("Malformed SEARCH/REPLACE block structure: Missing valid closing REPLACE marker")
+		}
+		return removeLineCount
+	}
+
+	/**
+	 * Removes trailing empty lines from the pendingNonStandardLines array
+	 * @param lineLimit - The index to start checking from (exclusive).
+	 *                    Removes empty lines from lineLimit-1 backwards.
+	 * @returns The number of empty lines removed
+	 */
+	private trimPendingNonStandardTrailingEmptyLines(lineLimit: number): number {
+		let removedCount = 0
+		let i = Math.min(lineLimit, this.pendingNonStandardLines.length) - 1
+
+		while (i >= 0 && this.pendingNonStandardLines[i].trim() === "") {
+			this.pendingNonStandardLines.pop()
+			removedCount++
+			i--
+		}
+
+		return removedCount
+	}
+}
+
+export async function constructNewFileContentV2(diffContent: string, originalContent: string, isFinal: boolean): Promise<string> {
+	let newFileContentConstructor = new NewFileContentConstructor(originalContent, isFinal)
+
+	let lines = diffContent.split("\n")
+
+	// If the last line looks like a partial marker but isn't recognized,
+	// remove it because it might be incomplete.
+	const lastLine = lines[lines.length - 1]
+	if (
+		lines.length > 0 &&
+		(lastLine.startsWith(SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith("=") ||
+			lastLine.startsWith(REPLACE_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_REPLACE_BLOCK_CHAR)) &&
+		lastLine !== SEARCH_BLOCK_START &&
+		lastLine !== SEARCH_BLOCK_END &&
+		lastLine !== REPLACE_BLOCK_END
+	) {
+		lines.pop()
+	}
+
+	for (const line of lines) {
+		newFileContentConstructor.processLine(line)
+	}
+
+	let result = newFileContentConstructor.getResult()
+	return result
+}

evals/diff-edits/diff-apply/diff-06-25-25.ts

@@ -0,0 +1,829 @@
+const SEARCH_BLOCK_START = "------- SEARCH"
+const SEARCH_BLOCK_END = "======="
+const REPLACE_BLOCK_END = "+++++++ REPLACE"
+
+const SEARCH_BLOCK_CHAR = "-"
+const REPLACE_BLOCK_CHAR = "+"
+const LEGACY_SEARCH_BLOCK_CHAR = "<"
+const LEGACY_REPLACE_BLOCK_CHAR = ">"
+
+// Replace the exact string constants with flexible regex patterns
+const SEARCH_BLOCK_START_REGEX = /^[-]{3,} SEARCH>?$/
+const LEGACY_SEARCH_BLOCK_START_REGEX = /^[<]{3,} SEARCH>?$/
+
+const SEARCH_BLOCK_END_REGEX = /^[=]{3,}$/
+
+const REPLACE_BLOCK_END_REGEX = /^[+]{3,} REPLACE>?$/
+const LEGACY_REPLACE_BLOCK_END_REGEX = /^[>]{3,} REPLACE>?$/
+
+// Helper functions to check if a line matches the flexible patterns
+function isSearchBlockStart(line: string): boolean {
+	return SEARCH_BLOCK_START_REGEX.test(line) || LEGACY_SEARCH_BLOCK_START_REGEX.test(line)
+}
+
+function isSearchBlockEnd(line: string): boolean {
+	return SEARCH_BLOCK_END_REGEX.test(line)
+}
+
+function isReplaceBlockEnd(line: string): boolean {
+	return REPLACE_BLOCK_END_REGEX.test(line) || LEGACY_REPLACE_BLOCK_END_REGEX.test(line)
+}
+
+/**
+ * Attempts a line-trimmed fallback match for the given search content in the original content.
+ * It tries to match `searchContent` lines against a block of lines in `originalContent` starting
+ * from `lastProcessedIndex`. Lines are matched by trimming leading/trailing whitespace and ensuring
+ * they are identical afterwards.
+ *
+ * Returns [matchIndexStart, matchIndexEnd] if found, or false if not found.
+ */
+function lineTrimmedFallbackMatch(originalContent: string, searchContent: string, startIndex: number): [number, number] | false {
+	// Split both contents into lines
+	const originalLines = originalContent.split("\n")
+	const searchLines = searchContent.split("\n")
+
+	// Trim trailing empty line if exists (from the trailing \n in searchContent)
+	if (searchLines[searchLines.length - 1] === "") {
+		searchLines.pop()
+	}
+
+	// Find the line number where startIndex falls
+	let startLineNum = 0
+	let currentIndex = 0
+	while (currentIndex < startIndex && startLineNum < originalLines.length) {
+		currentIndex += originalLines[startLineNum].length + 1 // +1 for \n
+		startLineNum++
+	}
+
+	// For each possible starting position in original content
+	for (let i = startLineNum; i <= originalLines.length - searchLines.length; i++) {
+		let matches = true
+
+		// Try to match all search lines from this position
+		for (let j = 0; j < searchLines.length; j++) {
+			const originalTrimmed = originalLines[i + j].trim()
+			const searchTrimmed = searchLines[j].trim()
+
+			if (originalTrimmed !== searchTrimmed) {
+				matches = false
+				break
+			}
+		}
+
+		// If we found a match, calculate the exact character positions
+		if (matches) {
+			// Find start character index
+			let matchStartIndex = 0
+			for (let k = 0; k < i; k++) {
+				matchStartIndex += originalLines[k].length + 1 // +1 for \n
+			}
+
+			// Find end character index
+			let matchEndIndex = matchStartIndex
+			for (let k = 0; k < searchLines.length; k++) {
+				matchEndIndex += originalLines[i + k].length + 1 // +1 for \n
+			}
+
+			return [matchStartIndex, matchEndIndex]
+		}
+	}
+
+	return false
+}
+
+/**
+ * Attempts to match blocks of code by using the first and last lines as anchors.
+ * This is a third-tier fallback strategy that helps match blocks where we can identify
+ * the correct location by matching the beginning and end, even if the exact content
+ * differs slightly.
+ *
+ * The matching strategy:
+ * 1. Only attempts to match blocks of 3 or more lines to avoid false positives
+ * 2. Extracts from the search content:
+ *    - First line as the "start anchor"
+ *    - Last line as the "end anchor"
+ * 3. For each position in the original content:
+ *    - Checks if the next line matches the start anchor
+ *    - If it does, jumps ahead by the search block size
+ *    - Checks if that line matches the end anchor
+ *    - All comparisons are done after trimming whitespace
+ *
+ * This approach is particularly useful for matching blocks of code where:
+ * - The exact content might have minor differences
+ * - The beginning and end of the block are distinctive enough to serve as anchors
+ * - The overall structure (number of lines) remains the same
+ *
+ * @param originalContent - The full content of the original file
+ * @param searchContent - The content we're trying to find in the original file
+ * @param startIndex - The character index in originalContent where to start searching
+ * @returns A tuple of [startIndex, endIndex] if a match is found, false otherwise
+ */
+function blockAnchorFallbackMatch(originalContent: string, searchContent: string, startIndex: number): [number, number] | false {
+	const originalLines = originalContent.split("\n")
+	const searchLines = searchContent.split("\n")
+
+	// Only use this approach for blocks of 3+ lines
+	if (searchLines.length < 3) {
+		return false
+	}
+
+	// Trim trailing empty line if exists
+	if (searchLines[searchLines.length - 1] === "") {
+		searchLines.pop()
+	}
+
+	const firstLineSearch = searchLines[0].trim()
+	const lastLineSearch = searchLines[searchLines.length - 1].trim()
+	const searchBlockSize = searchLines.length
+
+	// Find the line number where startIndex falls
+	let startLineNum = 0
+	let currentIndex = 0
+	while (currentIndex < startIndex && startLineNum < originalLines.length) {
+		currentIndex += originalLines[startLineNum].length + 1
+		startLineNum++
+	}
+
+	// Look for matching start and end anchors
+	for (let i = startLineNum; i <= originalLines.length - searchBlockSize; i++) {
+		// Check if first line matches
+		if (originalLines[i].trim() !== firstLineSearch) {
+			continue
+		}
+
+		// Check if last line matches at the expected position
+		if (originalLines[i + searchBlockSize - 1].trim() !== lastLineSearch) {
+			continue
+		}
+
+		// Calculate exact character positions
+		let matchStartIndex = 0
+		for (let k = 0; k < i; k++) {
+			matchStartIndex += originalLines[k].length + 1
+		}
+
+		let matchEndIndex = matchStartIndex
+		for (let k = 0; k < searchBlockSize; k++) {
+			matchEndIndex += originalLines[i + k].length + 1
+		}
+
+		return [matchStartIndex, matchEndIndex]
+	}
+
+	return false
+}
+
+/**
+ * This function reconstructs the file content by applying a streamed diff (in a
+ * specialized SEARCH/REPLACE block format) to the original file content. It is designed
+ * to handle both incremental updates and the final resulting file after all chunks have
+ * been processed.
+ *
+ * The diff format is a custom structure that uses three markers to define changes:
+ *
+ *   ------- SEARCH
+ *   [Exact content to find in the original file]
+ *   =======
+ *   [Content to replace with]
+ *   +++++++ REPLACE
+ *
+ * Behavior and Assumptions:
+ * 1. The file is processed chunk-by-chunk. Each chunk of `diffContent` may contain
+ *    partial or complete SEARCH/REPLACE blocks. By calling this function with each
+ *    incremental chunk (with `isFinal` indicating the last chunk), the final reconstructed
+ *    file content is produced.
+ *
+ * 2. Matching Strategy (in order of attempt):
+ *    a. Exact Match: First attempts to find the exact SEARCH block text in the original file
+ *    b. Line-Trimmed Match: Falls back to line-by-line comparison ignoring leading/trailing whitespace
+ *    c. Block Anchor Match: For blocks of 3+ lines, tries to match using first/last lines as anchors
+ *    If all matching strategies fail, an error is thrown.
+ *
+ * 3. Empty SEARCH Section:
+ *    - If SEARCH is empty and the original file is empty, this indicates creating a new file
+ *      (pure insertion).
+ *    - If SEARCH is empty and the original file is not empty, this indicates a complete
+ *      file replacement (the entire original content is considered matched and replaced).
+ *
+ * 4. Applying Changes:
+ *    - Before encountering the "=======" marker, lines are accumulated as search content.
+ *    - After "=======" and before ">>>>>>> REPLACE", lines are accumulated as replacement content.
+ *    - Once the block is complete (">>>>>>> REPLACE"), the matched section in the original
+ *      file is replaced with the accumulated replacement lines, and the position in the original
+ *      file is advanced.
+ *
+ * 5. Incremental Output:
+ *    - As soon as the match location is found and we are in the REPLACE section, each new
+ *      replacement line is appended to the result so that partial updates can be viewed
+ *      incrementally.
+ *
+ * 6. Partial Markers:
+ *    - If the final line of the chunk looks like it might be part of a marker but is not one
+ *      of the known markers, it is removed. This prevents incomplete or partial markers
+ *      from corrupting the output.
+ *
+ * 7. Finalization:
+ *    - Once all chunks have been processed (when `isFinal` is true), any remaining original
+ *      content after the last replaced section is appended to the result.
+ *    - Trailing newlines are not forcibly added. The code tries to output exactly what is specified.
+ *
+ * Errors:
+ * - If the search block cannot be matched using any of the available matching strategies,
+ *   an error is thrown.
+ */
+export async function constructNewFileContent(
+	diffContent: string,
+	originalContent: string,
+	isFinal: boolean,
+	version: "v1" | "v2" = "v1",
+): Promise<string> {
+	const constructor = constructNewFileContentVersionMapping[version]
+	if (!constructor) {
+		throw new Error(`Invalid version '${version}' for file content constructor`)
+	}
+	return constructor(diffContent, originalContent, isFinal)
+}
+
+const constructNewFileContentVersionMapping: Record<
+	string,
+	(diffContent: string, originalContent: string, isFinal: boolean) => Promise<string>
+> = {
+	v1: constructNewFileContentV1,
+	v2: constructNewFileContentV2,
+} as const
+
+async function constructNewFileContentV1(diffContent: string, originalContent: string, isFinal: boolean): Promise<string> {
+	let result = ""
+	let lastProcessedIndex = 0
+
+	let currentSearchContent = ""
+	let currentReplaceContent = ""
+	let inSearch = false
+	let inReplace = false
+
+	let searchMatchIndex = -1
+	let searchEndIndex = -1
+
+	// Track all replacements to handle out-of-order edits
+	let replacements: Array<{ start: number; end: number; content: string }> = []
+	let pendingOutOfOrderReplacement = false
+
+	let lines = diffContent.split("\n")
+
+	// If the last line looks like a partial marker but isn't recognized,
+	// remove it because it might be incomplete.
+	const lastLine = lines[lines.length - 1]
+	if (
+		lines.length > 0 &&
+		(lastLine.startsWith(SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith("=") ||
+			lastLine.startsWith(REPLACE_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_REPLACE_BLOCK_CHAR)) &&
+		!isSearchBlockStart(lastLine) &&
+		!isSearchBlockEnd(lastLine) &&
+		!isReplaceBlockEnd(lastLine)
+	) {
+		lines.pop()
+	}
+
+	for (const line of lines) {
+		if (isSearchBlockStart(line)) {
+			inSearch = true
+			currentSearchContent = ""
+			currentReplaceContent = ""
+			continue
+		}
+
+		if (isSearchBlockEnd(line)) {
+			inSearch = false
+			inReplace = true
+
+			// Remove trailing linebreak for adding the === marker
+			// if (currentSearchContent.endsWith("\r\n")) {
+			// 	currentSearchContent = currentSearchContent.slice(0, -2)
+			// } else if (currentSearchContent.endsWith("\n")) {
+			// 	currentSearchContent = currentSearchContent.slice(0, -1)
+			// }
+
+			if (!currentSearchContent) {
+				// Empty search block
+				if (originalContent.length === 0) {
+					// New file scenario: nothing to match, just start inserting
+					searchMatchIndex = 0
+					searchEndIndex = 0
+				} else {
+					// ERROR: Empty search block with non-empty file indicates malformed SEARCH marker
+					throw new Error(
+						"Empty SEARCH block detected with non-empty file. This usually indicates a malformed SEARCH marker.\n" +
+							"Please ensure your SEARCH marker follows the correct format:\n" +
+							"- Use '------- SEARCH' (7+ dashes + space + SEARCH)\n",
+					)
+				}
+			} else {
+				// Add check for inefficient full-file search
+				// if (currentSearchContent.trim() === originalContent.trim()) {
+				// 	throw new Error(
+				// 		"The SEARCH block contains the entire file content. Please either:\n" +
+				// 			"1. Use an empty SEARCH block to replace the entire file, or\n" +
+				// 			"2. Make focused changes to specific parts of the file that need modification.",
+				// 	)
+				// }
+
+				// Exact search match scenario
+				const exactIndex = originalContent.indexOf(currentSearchContent, lastProcessedIndex)
+				if (exactIndex !== -1) {
+					searchMatchIndex = exactIndex
+					searchEndIndex = exactIndex + currentSearchContent.length
+				} else {
+					// Attempt fallback line-trimmed matching
+					const lineMatch = lineTrimmedFallbackMatch(originalContent, currentSearchContent, lastProcessedIndex)
+					if (lineMatch) {
+						;[searchMatchIndex, searchEndIndex] = lineMatch
+					} else {
+						// Try block anchor fallback for larger blocks
+						const blockMatch = blockAnchorFallbackMatch(originalContent, currentSearchContent, lastProcessedIndex)
+						if (blockMatch) {
+							;[searchMatchIndex, searchEndIndex] = blockMatch
+						} else {
+							// Last resort: search the entire file from the beginning
+							const fullFileIndex = originalContent.indexOf(currentSearchContent, 0)
+							if (fullFileIndex !== -1) {
+								// Found in the file - could be out of order
+								searchMatchIndex = fullFileIndex
+								searchEndIndex = fullFileIndex + currentSearchContent.length
+								if (searchMatchIndex < lastProcessedIndex) {
+									pendingOutOfOrderReplacement = true
+								}
+							} else {
+								throw new Error(
+									`The SEARCH block:\n${currentSearchContent.trimEnd()}\n...does not match anything in the file.`,
+								)
+							}
+						}
+					}
+				}
+			}
+
+			// Check if this is an out-of-order replacement
+			if (searchMatchIndex < lastProcessedIndex) {
+				pendingOutOfOrderReplacement = true
+			}
+
+			// For in-order replacements, output everything up to the match location
+			if (!pendingOutOfOrderReplacement) {
+				result += originalContent.slice(lastProcessedIndex, searchMatchIndex)
+			}
+			continue
+		}
+
+		if (isReplaceBlockEnd(line)) {
+			// Finished one replace block
+
+			// Store this replacement
+			replacements.push({
+				start: searchMatchIndex,
+				end: searchEndIndex,
+				content: currentReplaceContent,
+			})
+
+			// If this was an in-order replacement, advance lastProcessedIndex
+			if (!pendingOutOfOrderReplacement) {
+				lastProcessedIndex = searchEndIndex
+			}
+
+			// Reset for next block
+			inSearch = false
+			inReplace = false
+			currentSearchContent = ""
+			currentReplaceContent = ""
+			searchMatchIndex = -1
+			searchEndIndex = -1
+			pendingOutOfOrderReplacement = false
+			continue
+		}
+
+		// Accumulate content for search or replace
+		// (currentReplaceContent is not being used for anything right now since we directly append to result.)
+		// (We artificially add a linebreak since we split on \n at the beginning. In order to not include a trailing linebreak in the final search/result blocks we need to remove it before using them. This allows for partial line matches to be correctly identified.)
+		// NOTE: search/replace blocks must be arranged in the order they appear in the file due to how we build the content using lastProcessedIndex. We also cannot strip the trailing newline since for non-partial lines it would remove the linebreak from the original content. (If we remove end linebreak from search, then we'd also have to remove it from replace but we can't know if it's a partial line or not since the model may be using the line break to indicate the end of the block rather than as part of the search content.) We require the model to output full lines in order for our fallbacks to work as well.
+		if (inSearch) {
+			currentSearchContent += line + "\n"
+		} else if (inReplace) {
+			currentReplaceContent += line + "\n"
+			// Only output replacement lines immediately for in-order replacements
+			if (searchMatchIndex !== -1 && !pendingOutOfOrderReplacement) {
+				result += line + "\n"
+			}
+		}
+	}
+
+	// If this is the final chunk, we need to apply all replacements and build the final result
+	if (isFinal) {
+		// Handle the case where we're still in replace mode when processing ends
+		// and this is the final chunk - treat it as if we encountered the REPLACE marker
+		if (inReplace && searchMatchIndex !== -1) {
+			// Store this replacement
+			replacements.push({
+				start: searchMatchIndex,
+				end: searchEndIndex,
+				content: currentReplaceContent,
+			})
+
+			// If this was an in-order replacement, advance lastProcessedIndex
+			if (!pendingOutOfOrderReplacement) {
+				lastProcessedIndex = searchEndIndex
+			}
+
+			// Reset state
+			inSearch = false
+			inReplace = false
+			currentSearchContent = ""
+			currentReplaceContent = ""
+			searchMatchIndex = -1
+			searchEndIndex = -1
+			pendingOutOfOrderReplacement = false
+		}
+		// end of handling missing replace marker
+
+		// Sort replacements by start position
+		replacements.sort((a, b) => a.start - b.start)
+
+		// Rebuild the entire result by applying all replacements
+		result = ""
+		let currentPos = 0
+
+		for (const replacement of replacements) {
+			// Add original content up to this replacement
+			result += originalContent.slice(currentPos, replacement.start)
+			// Add the replacement content
+			result += replacement.content
+			// Move position to after the replaced section
+			currentPos = replacement.end
+		}
+
+		// Add any remaining original content
+		result += originalContent.slice(currentPos)
+	}
+
+	return result
+}
+
+enum ProcessingState {
+	Idle = 0,
+	StateSearch = 1 << 0,
+	StateReplace = 1 << 1,
+}
+
+class NewFileContentConstructor {
+	private originalContent: string
+	private isFinal: boolean
+	private state: number
+	private pendingNonStandardLines: string[]
+	private result: string
+	private lastProcessedIndex: number
+	private currentSearchContent: string
+	private currentReplaceContent: string
+	private searchMatchIndex: number
+	private searchEndIndex: number
+
+	constructor(originalContent: string, isFinal: boolean) {
+		this.originalContent = originalContent
+		this.isFinal = isFinal
+		this.pendingNonStandardLines = []
+		this.result = ""
+		this.lastProcessedIndex = 0
+		this.state = ProcessingState.Idle
+		this.currentSearchContent = ""
+		this.currentReplaceContent = ""
+		this.searchMatchIndex = -1
+		this.searchEndIndex = -1
+	}
+
+	private resetForNextBlock() {
+		// Reset for next block
+		this.state = ProcessingState.Idle
+		this.currentSearchContent = ""
+		this.currentReplaceContent = ""
+		this.searchMatchIndex = -1
+		this.searchEndIndex = -1
+	}
+
+	private findLastMatchingLineIndex(regx: RegExp, lineLimit: number) {
+		for (let i = lineLimit; i > 0; ) {
+			i--
+			if (this.pendingNonStandardLines[i].match(regx)) {
+				return i
+			}
+		}
+		return -1
+	}
+
+	private updateProcessingState(newState: ProcessingState) {
+		const isValidTransition =
+			(this.state === ProcessingState.Idle && newState === ProcessingState.StateSearch) ||
+			(this.state === ProcessingState.StateSearch && newState === ProcessingState.StateReplace)
+
+		if (!isValidTransition) {
+			throw new Error(
+				`Invalid state transition.\n` +
+					"Valid transitions are:\n" +
+					"- Idle → StateSearch\n" +
+					"- StateSearch → StateReplace",
+			)
+		}
+
+		this.state |= newState
+	}
+
+	private isStateActive(state: ProcessingState): boolean {
+		return (this.state & state) === state
+	}
+
+	private activateReplaceState() {
+		this.updateProcessingState(ProcessingState.StateReplace)
+	}
+
+	private activateSearchState() {
+		this.updateProcessingState(ProcessingState.StateSearch)
+		this.currentSearchContent = ""
+		this.currentReplaceContent = ""
+	}
+
+	private isSearchingActive(): boolean {
+		return this.isStateActive(ProcessingState.StateSearch)
+	}
+
+	private isReplacingActive(): boolean {
+		return this.isStateActive(ProcessingState.StateReplace)
+	}
+
+	private hasPendingNonStandardLines(pendingNonStandardLineLimit: number): boolean {
+		return this.pendingNonStandardLines.length - pendingNonStandardLineLimit < this.pendingNonStandardLines.length
+	}
+
+	public processLine(line: string) {
+		this.internalProcessLine(line, true, this.pendingNonStandardLines.length)
+	}
+
+	public getResult() {
+		// If this is the final chunk, append any remaining original content
+		if (this.isFinal && this.lastProcessedIndex < this.originalContent.length) {
+			this.result += this.originalContent.slice(this.lastProcessedIndex)
+		}
+		if (this.isFinal && this.state !== ProcessingState.Idle) {
+			throw new Error("File processing incomplete - SEARCH/REPLACE operations still active during finalization")
+		}
+		return this.result
+	}
+
+	private internalProcessLine(
+		line: string,
+		canWritependingNonStandardLines: boolean,
+		pendingNonStandardLineLimit: number,
+	): number {
+		let removeLineCount = 0
+		if (isSearchBlockStart(line)) {
+			removeLineCount = this.trimPendingNonStandardTrailingEmptyLines(pendingNonStandardLineLimit)
+			if (removeLineCount > 0) {
+				pendingNonStandardLineLimit = pendingNonStandardLineLimit - removeLineCount
+			}
+			if (this.hasPendingNonStandardLines(pendingNonStandardLineLimit)) {
+				this.tryFixSearchReplaceBlock(pendingNonStandardLineLimit)
+				canWritependingNonStandardLines && (this.pendingNonStandardLines.length = 0)
+			}
+			this.activateSearchState()
+		} else if (isSearchBlockEnd(line)) {
+			// 校验非标内容
+			if (!this.isSearchingActive()) {
+				this.tryFixSearchBlock(pendingNonStandardLineLimit)
+				canWritependingNonStandardLines && (this.pendingNonStandardLines.length = 0)
+			}
+			this.activateReplaceState()
+			this.beforeReplace()
+		} else if (isReplaceBlockEnd(line)) {
+			if (!this.isReplacingActive()) {
+				this.tryFixReplaceBlock(pendingNonStandardLineLimit)
+				canWritependingNonStandardLines && (this.pendingNonStandardLines.length = 0)
+			}
+			this.lastProcessedIndex = this.searchEndIndex
+			this.resetForNextBlock()
+		} else {
+			// Accumulate content for search or replace
+			// (currentReplaceContent is not being used for anything right now since we directly append to result.)
+			// (We artificially add a linebreak since we split on \n at the beginning. In order to not include a trailing linebreak in the final search/result blocks we need to remove it before using them. This allows for partial line matches to be correctly identified.)
+			// NOTE: search/replace blocks must be arranged in the order they appear in the file due to how we build the content using lastProcessedIndex. We also cannot strip the trailing newline since for non-partial lines it would remove the linebreak from the original content. (If we remove end linebreak from search, then we'd also have to remove it from replace but we can't know if it's a partial line or not since the model may be using the line break to indicate the end of the block rather than as part of the search content.) We require the model to output full lines in order for our fallbacks to work as well.
+			if (this.isReplacingActive()) {
+				this.currentReplaceContent += line + "\n"
+				// Output replacement lines immediately if we know the insertion point
+				if (this.searchMatchIndex !== -1) {
+					this.result += line + "\n"
+				}
+			} else if (this.isSearchingActive()) {
+				this.currentSearchContent += line + "\n"
+			} else {
+				let appendToPendingNonStandardLines = canWritependingNonStandardLines
+				if (appendToPendingNonStandardLines) {
+					// 处理非标内容
+					this.pendingNonStandardLines.push(line)
+				}
+			}
+		}
+		return removeLineCount
+	}
+
+	private beforeReplace() {
+		// Remove trailing linebreak for adding the === marker
+		// if (currentSearchContent.endsWith("\r\n")) {
+		// 	currentSearchContent = currentSearchContent.slice(0, -2)
+		// } else if (currentSearchContent.endsWith("\n")) {
+		// 	currentSearchContent = currentSearchContent.slice(0, -1)
+		// }
+
+		if (!this.currentSearchContent) {
+			// Empty search block
+			if (this.originalContent.length === 0) {
+				// New file scenario: nothing to match, just start inserting
+				this.searchMatchIndex = 0
+				this.searchEndIndex = 0
+			} else {
+				// Complete file replacement scenario: treat the entire file as matched
+				this.searchMatchIndex = 0
+				this.searchEndIndex = this.originalContent.length
+			}
+		} else {
+			// Add check for inefficient full-file search
+			// if (currentSearchContent.trim() === originalContent.trim()) {
+			// 	throw new Error(
+			// 		"The SEARCH block contains the entire file content. Please either:\n" +
+			// 			"1. Use an empty SEARCH block to replace the entire file, or\n" +
+			// 			"2. Make focused changes to specific parts of the file that need modification.",
+			// 	)
+			// }
+			// Exact search match scenario
+			const exactIndex = this.originalContent.indexOf(this.currentSearchContent, this.lastProcessedIndex)
+			if (exactIndex !== -1) {
+				this.searchMatchIndex = exactIndex
+				this.searchEndIndex = exactIndex + this.currentSearchContent.length
+			} else {
+				// Attempt fallback line-trimmed matching
+				const lineMatch = lineTrimmedFallbackMatch(
+					this.originalContent,
+					this.currentSearchContent,
+					this.lastProcessedIndex,
+				)
+				if (lineMatch) {
+					;[this.searchMatchIndex, this.searchEndIndex] = lineMatch
+				} else {
+					// Try block anchor fallback for larger blocks
+					const blockMatch = blockAnchorFallbackMatch(
+						this.originalContent,
+						this.currentSearchContent,
+						this.lastProcessedIndex,
+					)
+					if (blockMatch) {
+						;[this.searchMatchIndex, this.searchEndIndex] = blockMatch
+					} else {
+						throw new Error(
+							`The SEARCH block:\n${this.currentSearchContent.trimEnd()}\n...does not match anything in the file.`,
+						)
+					}
+				}
+			}
+		}
+		if (this.searchMatchIndex < this.lastProcessedIndex) {
+			throw new Error(
+				`The SEARCH block:\n${this.currentSearchContent.trimEnd()}\n...matched an incorrect content in the file.`,
+			)
+		}
+		// Output everything up to the match location
+		this.result += this.originalContent.slice(this.lastProcessedIndex, this.searchMatchIndex)
+	}
+
+	private tryFixSearchBlock(lineLimit: number): number {
+		let removeLineCount = 0
+		if (lineLimit < 0) {
+			lineLimit = this.pendingNonStandardLines.length
+		}
+		if (!lineLimit) {
+			throw new Error("Invalid SEARCH/REPLACE block structure - no lines available to process")
+		}
+		let searchTagRegexp = /^([-]{3,}|[<]{3,}) SEARCH$/
+		const searchTagIndex = this.findLastMatchingLineIndex(searchTagRegexp, lineLimit)
+		if (searchTagIndex !== -1) {
+			let fixLines = this.pendingNonStandardLines.slice(searchTagIndex, lineLimit)
+			fixLines[0] = SEARCH_BLOCK_START
+			for (const line of fixLines) {
+				removeLineCount += this.internalProcessLine(line, false, searchTagIndex)
+			}
+		} else {
+			throw new Error(
+				`Invalid REPLACE marker detected - could not find matching SEARCH block starting from line ${searchTagIndex + 1}`,
+			)
+		}
+		return removeLineCount
+	}
+
+	private tryFixReplaceBlock(lineLimit: number): number {
+		let removeLineCount = 0
+		if (lineLimit < 0) {
+			lineLimit = this.pendingNonStandardLines.length
+		}
+		if (!lineLimit) {
+			throw new Error()
+		}
+		let replaceBeginTagRegexp = /^[=]{3,}$/
+		const replaceBeginTagIndex = this.findLastMatchingLineIndex(replaceBeginTagRegexp, lineLimit)
+		if (replaceBeginTagIndex !== -1) {
+			// // 校验非标内容
+			// if (!this.isSearchingActive()) {
+			// 	removeLineCount += this.tryFixSearchBlock(replaceBeginTagIndex)
+			// }
+			let fixLines = this.pendingNonStandardLines.slice(replaceBeginTagIndex - removeLineCount, lineLimit - removeLineCount)
+			fixLines[0] = SEARCH_BLOCK_END
+			for (const line of fixLines) {
+				removeLineCount += this.internalProcessLine(line, false, replaceBeginTagIndex - removeLineCount)
+			}
+		} else {
+			throw new Error(`Malformed REPLACE block - missing valid separator after line ${replaceBeginTagIndex + 1}`)
+		}
+		return removeLineCount
+	}
+
+	private tryFixSearchReplaceBlock(lineLimit: number): number {
+		let removeLineCount = 0
+		if (lineLimit < 0) {
+			lineLimit = this.pendingNonStandardLines.length
+		}
+		if (!lineLimit) {
+			throw new Error()
+		}
+
+		let replaceEndTagRegexp = /^([+]{3,}|[>]{3,}) REPLACE$/
+		const replaceEndTagIndex = this.findLastMatchingLineIndex(replaceEndTagRegexp, lineLimit)
+		const likeReplaceEndTag = replaceEndTagIndex === lineLimit - 1
+		if (likeReplaceEndTag) {
+			// // 校验非标内容
+			// if (!this.isReplacingActive()) {
+			// 	removeLineCount += this.tryFixReplaceBlock(replaceEndTagIndex)
+			// }
+			let fixLines = this.pendingNonStandardLines.slice(replaceEndTagIndex - removeLineCount, lineLimit - removeLineCount)
+			fixLines[fixLines.length - 1] = REPLACE_BLOCK_END
+			for (const line of fixLines) {
+				removeLineCount += this.internalProcessLine(line, false, replaceEndTagIndex - removeLineCount)
+			}
+		} else {
+			throw new Error("Malformed SEARCH/REPLACE block structure: Missing valid closing REPLACE marker")
+		}
+		return removeLineCount
+	}
+
+	/**
+	 * Removes trailing empty lines from the pendingNonStandardLines array
+	 * @param lineLimit - The index to start checking from (exclusive).
+	 *                    Removes empty lines from lineLimit-1 backwards.
+	 * @returns The number of empty lines removed
+	 */
+	private trimPendingNonStandardTrailingEmptyLines(lineLimit: number): number {
+		let removedCount = 0
+		let i = Math.min(lineLimit, this.pendingNonStandardLines.length) - 1
+
+		while (i >= 0 && this.pendingNonStandardLines[i].trim() === "") {
+			this.pendingNonStandardLines.pop()
+			removedCount++
+			i--
+		}
+
+		return removedCount
+	}
+}
+
+export async function constructNewFileContentV2(diffContent: string, originalContent: string, isFinal: boolean): Promise<string> {
+	let newFileContentConstructor = new NewFileContentConstructor(originalContent, isFinal)
+
+	let lines = diffContent.split("\n")
+
+	// If the last line looks like a partial marker but isn't recognized,
+	// remove it because it might be incomplete.
+	const lastLine = lines[lines.length - 1]
+	if (
+		lines.length > 0 &&
+		(lastLine.startsWith(SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_SEARCH_BLOCK_CHAR) ||
+			lastLine.startsWith("=") ||
+			lastLine.startsWith(REPLACE_BLOCK_CHAR) ||
+			lastLine.startsWith(LEGACY_REPLACE_BLOCK_CHAR)) &&
+		lastLine !== SEARCH_BLOCK_START &&
+		lastLine !== SEARCH_BLOCK_END &&
+		lastLine !== REPLACE_BLOCK_END
+	) {
+		lines.pop()
+	}
+
+	for (const line of lines) {
+		newFileContentConstructor.processLine(line)
+	}
+
+	let result = newFileContentConstructor.getResult()
+	return result
+}

evals/diff-edits/types.ts

@@ -33,6 +33,7 @@ export interface TestConfig {
 	diff_edit_function: string
 	thinking_tokens_budget: number
 	replay: boolean
+	diff_apply_file?: string
 }
 
 export interface SystemPromptDetails {
@@ -100,4 +101,5 @@ export interface TestInput {
 	diffEditFunction: string
 	thinkingBudgetTokens: number
 	originalDiffEditToolCallMessage?: string
+	diffApplyFile?: string
 }

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": "3.17.15",
+	"version": "3.18.0",
 	"icon": "assets/icons/icon.png",
 	"engines": {
 		"vscode": "^1.84.0"
@@ -332,13 +332,14 @@
 		"package": "npm run check-types && npm run build:webview && npm run lint && node esbuild.js --production",
 		"protos": "node proto/build-proto.js && node scripts/generate-server-setup.mjs && node scripts/generate-host-bridge-client.mjs",
 		"postprotos": "prettier src/shared/proto src/core/controller src/hosts/ webview-ui/src/services src/generated --write --log-level warn",
+		"clean": "rimraf dist dist-standalone webview-ui/build src/generated out/",
 		"compile-tests": "node ./scripts/build-tests.js",
 		"watch-tests": "tsc -p . -w --outDir out",
-		"pretest": "npm run compile-tests && npm run compile && npm run compile-standalone && npm run lint",
 		"check-types": "npm run protos && tsc --noEmit",
 		"lint": "eslint src --ext ts && eslint webview-ui/src --ext ts && buf lint && cd webview-ui && npm run lint",
 		"format": "prettier . --check",
 		"format:fix": "prettier . --write",
+		"pretest": "npm run compile && npm run compile-tests && npm run compile-standalone && npm run lint",
 		"test": "npm-run-all test:unit test:integration",
 		"test:ci": "node scripts/test-ci.js",
 		"test:integration": "vscode-test",
@@ -396,6 +397,7 @@
 		"prettier": "^3.3.3",
 		"protoc-gen-ts": "^0.8.7",
 		"proxyquire": "^2.1.3",
+		"rimraf": "^6.0.1",
 		"should": "^13.2.3",
 		"sinon": "^19.0.2",
 		"ts-node": "^10.9.2",
@@ -440,6 +442,7 @@
 		"fzf": "^0.5.2",
 		"get-folder-size": "^5.0.0",
 		"globby": "^14.0.2",
+		"google-auth-library": "^10.1.0",
 		"grpc-health-check": "^2.0.2",
 		"iconv-lite": "^0.6.3",
 		"ignore": "^7.0.3",

proto/build-proto-config.js

@@ -0,0 +1,30 @@
+// Configuration file for protocol buffer build scripts
+// Contains service name mappings used by both build-proto.js and build-go-proto.js
+
+// List of gRPC services
+// To add a new service, simply add it to this map and run the build scripts
+// The service handler will be automatically discovered and used by grpc-handler.ts
+export const serviceNameMap = {
+	account: "cline.AccountService",
+	browser: "cline.BrowserService",
+	checkpoints: "cline.CheckpointsService",
+	file: "cline.FileService",
+	mcp: "cline.McpService",
+	state: "cline.StateService",
+	task: "cline.TaskService",
+	web: "cline.WebService",
+	models: "cline.ModelsService",
+	slash: "cline.SlashService",
+	ui: "cline.UiService",
+	// Add new services here - no other code changes needed!
+}
+
+// List of host gRPC services (IDE API bridge)
+// These services are implemented in the IDE extension and called by the standalone Cline Core
+export const hostServiceNameMap = {
+	uri: "host.UriService",
+	watch: "host.WatchService",
+	workspace: "host.WorkspaceService",
+	env: "host.EnvService",
+	// Add new host services here
+}

proto/build-proto.js

@@ -9,16 +9,18 @@ import chalk from "chalk"
 import os from "os"
 
 import { createRequire } from "module"
+import { serviceNameMap, hostServiceNameMap } from "./build-proto-config.js"
+
 const require = createRequire(import.meta.url)
 const PROTOC = path.join(require.resolve("grpc-tools"), "../bin/protoc")
 
 const SCRIPT_DIR = path.dirname(fileURLToPath(import.meta.url))
 const ROOT_DIR = path.resolve(SCRIPT_DIR, "..")
 
-const TS_OUT_DIR = path.join(ROOT_DIR, "src/shared/proto")
-const GRPC_JS_OUT_DIR = path.join(ROOT_DIR, "src/generated/grpc-js")
-const NICE_JS_OUT_DIR = path.join(ROOT_DIR, "src/generated/nice-grpc")
-const DESCRIPTOR_OUT_DIR = path.join(ROOT_DIR, "dist-standalone/proto")
+const TS_OUT_DIR = path.join(ROOT_DIR, "src", "shared", "proto")
+const GRPC_JS_OUT_DIR = path.join(ROOT_DIR, "src", "generated", "grpc-js")
+const NICE_JS_OUT_DIR = path.join(ROOT_DIR, "src", "generated", "nice-grpc")
+const DESCRIPTOR_OUT_DIR = path.join(ROOT_DIR, "dist-standalone", "proto")
 
 const isWindows = process.platform === "win32"
 const TS_PROTO_PLUGIN = isWindows
@@ -34,33 +36,13 @@ const TS_PROTO_OPTIONS = [
 	"useDate=false", // Timestamp fields will not be automatically converted to Date.
 ]
 
-// List of gRPC services
-// To add a new service, simply add it to this map and run this script
-// The service handler will be automatically discovered and used by grpc-handler.ts
-const serviceNameMap = {
-	account: "cline.AccountService",
-	browser: "cline.BrowserService",
-	checkpoints: "cline.CheckpointsService",
-	file: "cline.FileService",
-	mcp: "cline.McpService",
-	state: "cline.StateService",
-	task: "cline.TaskService",
-	web: "cline.WebService",
-	models: "cline.ModelsService",
-	slash: "cline.SlashService",
-	ui: "cline.UiService",
-	// Add new services here - no other code changes needed!
-}
-const serviceDirs = Object.keys(serviceNameMap).map((serviceKey) => path.join(ROOT_DIR, "src/core/controller", serviceKey))
+// Service directories derived from imported serviceNameMap
+const serviceDirs = Object.keys(serviceNameMap).map((serviceKey) => path.join(ROOT_DIR, "src", "core", "controller", serviceKey))
 
-// List of host gRPC services (IDE API bridge)
-// These services are implemented in the IDE extension and called by the standalone Cline Core
-const hostServiceNameMap = {
-	uri: "host.UriService",
-	watch: "host.WatchService",
-	// Add new host services here
-}
-const hostServiceDirs = Object.keys(hostServiceNameMap).map((serviceKey) => path.join(ROOT_DIR, "src/hosts/vscode", serviceKey))
+// Host service directories derived from imported hostServiceNameMap
+const hostServiceDirs = Object.keys(hostServiceNameMap).map((serviceKey) =>
+	path.join(ROOT_DIR, "src", "hosts", "vscode", serviceKey),
+)
 
 async function main() {
 	console.log(chalk.bold.blue("Starting Protocol Buffer code generation..."))
@@ -176,7 +158,7 @@ export {
 	${serviceExports.join(",\n\t")}
 }`
 
-	const filePath = path.join(ROOT_DIR, "webview-ui/src/services/grpc-client.ts")
+	const filePath = path.join(ROOT_DIR, "webview-ui", "src", "services", "grpc-client.ts")
 	await writeFileWithMkdirs(filePath, content)
 	log_verbose(chalk.green(`Generated gRPC client at ${filePath}`))
 }
@@ -385,7 +367,7 @@ export interface ServiceHandlerConfig {
 export const serviceHandlers: Record<string, ServiceHandlerConfig> = {${serviceConfigs.join(",")}
 };`
 
-	const configPath = path.join(ROOT_DIR, "src/core/controller/grpc-service-config.ts")
+	const configPath = path.join(ROOT_DIR, "src", "core", "controller", "grpc-service-config.ts")
 	await writeFileWithMkdirs(configPath, content)
 	log_verbose(chalk.green(`Generated service configuration at ${configPath}`))
 }
@@ -601,13 +583,12 @@ async function cleanup() {
 	for (const file of existingFiles) {
 		await fs.unlink(path.join(TS_OUT_DIR, file))
 	}
-	await rmdir(path.join(ROOT_DIR, "src/generated"))
+	await rmdir(path.join(ROOT_DIR, "src", "generated"))
 
 	// Clean up generated files that were moved.
-	await fs.rm(path.join(ROOT_DIR, "src/standalone/services/host-grpc-client.ts"), { force: true })
-	await rmdir(path.join(ROOT_DIR, "src/standalone/services"))
-
-	await fs.rm(path.join(ROOT_DIR, "hosts/vscode"), { force: true, recursive: true })
+	await fs.rm(path.join(ROOT_DIR, "src", "standalone", "services", "host-grpc-client.ts"), { force: true })
+	await rmdir(path.join(ROOT_DIR, "src", "standalone", "services"))
+	await fs.rm(path.join(ROOT_DIR, "hosts", "vscode"), { force: true, recursive: true })
 	await rmdir(path.join(ROOT_DIR, "hosts"))
 
 	await fs.rm(path.join(ROOT_DIR, "src/standalone/server-setup.ts"), { force: true })
@@ -635,13 +616,6 @@ async function rmdir(path) {
 	}
 }
 
-function serviceNameWithoutPackage(fullServiceName) {
-	return fullServiceName.replace(/.*\./, "")
-}
-function lowercaseFirstChar(str) {
-	return str.charAt(0).toLowerCase() + str.slice(1)
-}
-
 // Check for Apple Silicon compatibility
 function checkAppleSiliconCompatibility() {
 	// Only run check on macOS

proto/file.proto

@@ -29,9 +29,6 @@ service FileService {
   // Search git commits in the workspace
   rpc searchCommits(StringRequest) returns (GitCommits);
 
-  // Select images from the file system and return as data URLs
-  rpc selectImages(EmptyRequest) returns (StringArray);
-
   // Select images and other files from the file system and returns as data URLs & paths respectively
   rpc selectFiles(BooleanRequest) returns (StringArrays);
   

proto/host/env.proto

@@ -0,0 +1,16 @@
+syntax = "proto3";
+
+package host;
+option java_package = "bot.cline.host.proto";
+option java_multiple_files = true;
+
+import "common.proto";
+
+// Provides methods for working with the user's environment.
+service EnvService {
+  // Writes text to the system clipboard.
+  rpc clipboardWriteText(cline.StringRequest) returns (cline.Empty);
+
+  // Reads text from the system clipboard.
+  rpc clipboardReadText(cline.EmptyRequest) returns (cline.String);
+}

proto/host/workspace.proto

@@ -0,0 +1,24 @@
+syntax = "proto3";
+
+package host;
+option java_package = "bot.cline.host.proto";
+option java_multiple_files = true;
+
+// Provides methods for working with workspaces/projects.
+service WorkspaceService {
+  // Returns a list of the top level directories of the workspace.
+  rpc getWorkspacePaths(GetWorkspacePathsRequest) returns (GetWorkspacePathsResponse);
+}
+
+message GetWorkspacePathsRequest {
+  // The unique ID for the workspace/project.
+  // This is currently optional in vscode. It is required in other environments where cline is running at
+  // the application level, and the user can open multiple projects.
+  optional string id = 1;
+}
+
+message GetWorkspacePathsResponse {
+  // The unique ID for the workspace/project.
+  optional string id = 1;
+  repeated string paths = 2;
+}

proto/models.proto

@@ -105,24 +105,25 @@ enum ApiProvider {
   OLLAMA = 5;
   LMSTUDIO = 6;
   GEMINI = 7;
-  OPENAI_NATIVE = 8;
-  REQUESTY = 9;
-  TOGETHER = 10;
-  DEEPSEEK = 11;
-  QWEN = 12;
-  DOUBAO = 13;
-  MISTRAL = 14;
-  VSCODE_LM = 15;
-  CLINE = 16;
-  LITELLM = 17;
-  NEBIUS = 18;
-  FIREWORKS = 19;
-  ASKSAGE = 20;
-  XAI = 21;
-  SAMBANOVA = 22;
-  CEREBRAS = 23;
-  SAPAICORE = 24;
-  CLAUDE_CODE = 25;
+  GEMINI_CLI = 8;
+  OPENAI_NATIVE = 9;
+  REQUESTY = 10;
+  TOGETHER = 11;
+  DEEPSEEK = 12;
+  QWEN = 13;
+  DOUBAO = 14;
+  MISTRAL = 15;
+  VSCODE_LM = 16;
+  CLINE = 17;
+  LITELLM = 18;
+  NEBIUS = 19;
+  FIREWORKS = 20;
+  ASKSAGE = 21;
+  XAI = 22;
+  SAMBANOVA = 23;
+  CEREBRAS = 24;
+  SAPAICORE = 25;
+  CLAUDE_CODE = 26;
 }
 
 // Model info for OpenAI-compatible models
@@ -236,4 +237,6 @@ message ModelsApiConfiguration {
   optional string sap_ai_core_token_url = 71;
   optional string sap_ai_core_base_url = 72;
   optional string claude_code_path = 73;
-}
+  optional string gemini_cli_oauth_path = 74;
+  optional string gemini_cli_project_id = 75;
+}

proto/state.proto

@@ -17,6 +17,7 @@ service StateService {
   rpc togglePlanActMode(TogglePlanActModeRequest) returns (Boolean);
   rpc updateAutoApprovalSettings(AutoApprovalSettingsRequest) returns (Empty);
   rpc updateSettings(UpdateSettingsRequest) returns (Empty);
+  rpc updateTelemetrySetting(TelemetrySettingRequest) returns (Empty);
 }
 
 message State {
@@ -88,6 +89,17 @@ message AutoApprovalSettingsRequest {
   repeated string favorites = 7;
 }
 
+enum TelemetrySettingEnum {
+  UNSET = 0;
+  ENABLED = 1;
+  DISABLED = 2;
+}
+
+message TelemetrySettingRequest {
+  Metadata metadata = 1;
+  TelemetrySettingEnum setting = 2;
+}
+
 // Message for updating settings
 message UpdateSettingsRequest {
   Metadata metadata = 1;

proto/task.proto

@@ -23,8 +23,6 @@ service TaskService {
   rpc exportTaskWithId(StringRequest) returns (Empty);
   // Toggles the favorite status of a task
   rpc toggleTaskFavorite(TaskFavoriteRequest) returns (Empty);
-  // Deletes all non-favorited tasks
-  rpc deleteNonFavoritedTasks(EmptyRequest) returns (DeleteNonFavoritedTasksResults);
   // Gets filtered task history
   rpc getTaskHistory(GetTaskHistoryRequest) returns (TaskHistoryArray);
   // Sends a response to a previous ask operation
@@ -35,6 +33,8 @@ service TaskService {
   rpc taskCompletionViewChanges(Int64Request) returns (Empty);
   // Executes a quick win task with command and title
   rpc executeQuickWin(ExecuteQuickWinRequest) returns (Empty);
+  // Deletes all task history
+  rpc deleteAllTaskHistory(EmptyRequest) returns (DeleteAllTaskHistoryCount);
 }
 
 // Request message for creating a new task
@@ -66,12 +66,6 @@ message TaskResponse {
   int32 cache_reads = 10;
 }
 
-// Results returned when deleting non-favorited tasks
-message DeleteNonFavoritedTasksResults {
-  int32 tasks_preserved = 1;
-  int32 tasks_deleted = 2;
-}
-
 // Request for getting task history with filtering
 message GetTaskHistoryRequest {
   Metadata metadata = 1;
@@ -116,3 +110,8 @@ message ExecuteQuickWinRequest {
   string command = 2;
   string title = 3;
 }
+
+// Results returned when deleting all task history
+message DeleteAllTaskHistoryCount {
+  int32 tasks_deleted = 1;
+}

proto/ui.proto

@@ -265,4 +265,7 @@ service UiService {
 
   // Returns the HTML for the webview index page. This is only used by external clients, not by the vscode webview.
   rpc getWebviewHtml(EmptyRequest) returns (String);
+  
+  // Opens a URL in the default browser
+  rpc openUrl(StringRequest) returns (Empty);
 }

scripts/get-vscode-usages.sh

@@ -13,7 +13,8 @@ grep -Ev '//.*vscode' | # remove commented out code
 sed 's|.*vscode\.|vscode.|'| # remove everything before vscode.
 sed 's/[^a-zA-Z0-9_.].*$//' | # remove everything after last identifier
 grep -E '\.[a-z][^.]+$' | # remove types (last part of identifier should be lowercase)
-sort | uniq > $SDK_DEST
+sort | uniq -c | sort -n | # Count occurrences
+cat > $SDK_DEST
 }
 echo Wrote uses of the vscode SDK to $(realpath $SDK_DEST)
 

src/api/index.ts

@@ -8,6 +8,7 @@ import { OpenAiHandler } from "./providers/openai"
 import { OllamaHandler } from "./providers/ollama"
 import { LmStudioHandler } from "./providers/lmstudio"
 import { GeminiHandler } from "./providers/gemini"
+import { GeminiCliHandler } from "./providers/gemini-cli"
 import { OpenAiNativeHandler } from "./providers/openai-native"
 import { ApiStream, ApiStreamUsageChunk } from "./transform/stream"
 import { DeepSeekHandler } from "./providers/deepseek"
@@ -38,8 +39,7 @@ export interface SingleCompletionHandler {
 	completePrompt(prompt: string): Promise<string>
 }
 
-export function buildApiHandler(configuration: ApiConfiguration): ApiHandler {
-	const { apiProvider, ...options } = configuration
+function createHandlerForProvider(apiProvider: string | undefined, options: any): ApiHandler {
 	switch (apiProvider) {
 		case "anthropic":
 			return new AnthropicHandler(options)
@@ -57,6 +57,8 @@ export function buildApiHandler(configuration: ApiConfiguration): ApiHandler {
 			return new LmStudioHandler(options)
 		case "gemini":
 			return new GeminiHandler(options)
+		case "gemini-cli":
+			return new GeminiCliHandler(options)
 		case "openai-native":
 			return new OpenAiNativeHandler(options)
 		case "deepseek":
@@ -97,3 +99,27 @@ export function buildApiHandler(configuration: ApiConfiguration): ApiHandler {
 			return new AnthropicHandler(options)
 	}
 }
+
+export function buildApiHandler(configuration: ApiConfiguration): ApiHandler {
+	const { apiProvider, ...options } = configuration
+
+	// Validate thinking budget tokens against model's maxTokens to prevent API errors
+	// wrapped in a try-catch for safety, but this should never throw
+	try {
+		if (options.thinkingBudgetTokens && options.thinkingBudgetTokens > 0) {
+			const handler = createHandlerForProvider(apiProvider, options)
+
+			const modelInfo = handler.getModel().info
+			if (modelInfo.maxTokens && options.thinkingBudgetTokens > modelInfo.maxTokens) {
+				const clippedValue = modelInfo.maxTokens - 1
+				options.thinkingBudgetTokens = clippedValue
+			} else {
+				return handler // don't rebuild unless its necessary
+			}
+		}
+	} catch (error) {
+		console.error("buildApiHandler error:", error)
+	}
+
+	return createHandlerForProvider(apiProvider, options)
+}

src/api/providers/claude-code.ts

@@ -4,7 +4,7 @@ import { type ApiHandler } from ".."
 import { ApiStreamUsageChunk, type ApiStream } from "../transform/stream"
 import { withRetry } from "../retry"
 import { runClaudeCode } from "@/integrations/claude-code/run"
-import { ClaudeCodeMessage } from "@/integrations/claude-code/types"
+import { filterMessagesForClaudeCode } from "@/integrations/claude-code/message-filter"
 
 export class ClaudeCodeHandler implements ApiHandler {
 	private options: ApiHandlerOptions
@@ -19,39 +19,16 @@ export class ClaudeCodeHandler implements ApiHandler {
 		maxDelay: 15000,
 	})
 	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
+		// Filter out image blocks since Claude Code doesn't support them
+		const filteredMessages = filterMessagesForClaudeCode(messages)
+
 		const claudeProcess = runClaudeCode({
 			systemPrompt,
-			messages,
+			messages: filteredMessages,
 			path: this.options.claudeCodePath,
 			modelId: this.getModel().id,
 		})
 
-		const dataQueue: string[] = []
-		let processError = null
-		let errorOutput = ""
-		let exitCode: number | null = null
-
-		claudeProcess.stdout.on("data", (data) => {
-			const output = data.toString()
-			const lines = output.split("\n").filter((line: string) => line.trim() !== "")
-
-			for (const line of lines) {
-				dataQueue.push(line)
-			}
-		})
-
-		claudeProcess.stderr.on("data", (data) => {
-			errorOutput += data.toString()
-		})
-
-		claudeProcess.on("close", (code) => {
-			exitCode = code
-		})
-
-		claudeProcess.on("error", (error) => {
-			processError = error
-		})
-
 		// Usage is included with assistant messages,
 		// but cost is included in the result chunk
 		let usage: ApiStreamUsageChunk = {
@@ -62,61 +39,75 @@ export class ClaudeCodeHandler implements ApiHandler {
 			cacheWriteTokens: 0,
 		}
 
-		while (exitCode !== 0 || dataQueue.length > 0) {
-			if (dataQueue.length === 0) {
-				await new Promise((resolve) => setImmediate(resolve))
-			}
+		let isPaidUsage = true
 
-			if (exitCode !== null && exitCode !== 0) {
-				throw new Error(
-					`Claude Code process exited with code ${exitCode}.${errorOutput ? ` Error output: ${errorOutput.trim()}` : ""}`,
-				)
-			}
-
-			const data = dataQueue.shift()
-			if (!data) {
-				continue
-			}
-
-			const chunk = this.attemptParseChunk(data)
-
-			if (!chunk) {
+		for await (const chunk of claudeProcess) {
+			if (typeof chunk === "string") {
 				yield {
 					type: "text",
-					text: data || "",
+					text: chunk,
 				}
 
 				continue
 			}
 
 			if (chunk.type === "system" && chunk.subtype === "init") {
+				// Based on my tests, subscription usage sets the `apiKeySource` to "none"
+				isPaidUsage = chunk.apiKeySource !== "none"
 				continue
 			}
 
 			if (chunk.type === "assistant" && "message" in chunk) {
 				const message = chunk.message
 
-				if (message.stop_reason !== null && message.stop_reason !== "tool_use") {
-					const errorMessage = message.content[0]?.text || `Claude Code stopped with reason: ${message.stop_reason}`
+				if (message.stop_reason !== null) {
+					const content = "text" in message.content[0] ? message.content[0] : undefined
 
-					if (errorMessage.includes("Invalid model name")) {
-						throw new Error(
-							errorMessage +
-								`\n\nAPI keys and subscription plans allow different models. Make sure the selected model is included in your plan.`,
-						)
+					const isError = content && content.text.startsWith(`API Error`)
+					if (isError) {
+						// Error messages are formatted as: `API Error: <<status code>> <<json>>`
+						const errorMessageStart = content.text.indexOf("{")
+						const errorMessage = content.text.slice(errorMessageStart)
+
+						const error = this.attemptParse(errorMessage)
+						if (!error) {
+							throw new Error(content.text)
+						}
+
+						if (error.error.message.includes("Invalid model name")) {
+							throw new Error(
+								content.text +
+									`\n\nAPI keys and subscription plans allow different models. Make sure the selected model is included in your plan.`,
+							)
+						}
+
+						throw new Error(errorMessage)
 					}
-
-					throw new Error(errorMessage)
 				}
 
 				for (const content of message.content) {
-					if (content.type === "text") {
-						yield {
-							type: "text",
-							text: content.text,
-						}
-					} else {
-						console.warn("Unsupported content type:", content.type)
+					switch (content.type) {
+						case "text":
+							yield {
+								type: "text",
+								text: content.text,
+							}
+							break
+						case "thinking":
+							yield {
+								type: "reasoning",
+								reasoning: content.thinking || "",
+							}
+							break
+						case "redacted_thinking":
+							yield {
+								type: "reasoning",
+								reasoning: "[Redacted thinking block]",
+							}
+							break
+						case "tool_use":
+							console.error(`tool_use is not supported yet. Received: ${JSON.stringify(content)}`)
+							break
 					}
 				}
 
@@ -129,14 +120,18 @@ export class ClaudeCodeHandler implements ApiHandler {
 			}
 
 			if (chunk.type === "result" && "result" in chunk) {
-				usage.totalCost = chunk.cost_usd || 0
+				usage.totalCost = isPaidUsage ? chunk.total_cost_usd : 0
 
 				yield usage
 			}
+		}
+	}
 
-			if (processError) {
-				throw processError
-			}
+	private attemptParse(str: string) {
+		try {
+			return JSON.parse(str)
+		} catch (err) {
+			return null
 		}
 	}
 
@@ -152,14 +147,4 @@ export class ClaudeCodeHandler implements ApiHandler {
 			info: claudeCodeModels[claudeCodeDefaultModelId],
 		}
 	}
-
-	// TOOD: Validate instead of parsing
-	private attemptParseChunk(data: string): ClaudeCodeMessage | null {
-		try {
-			return JSON.parse(data)
-		} catch (error) {
-			console.error("Error parsing chunk:", error)
-			return null
-		}
-	}
 }

src/api/providers/cline.ts

@@ -56,7 +56,21 @@ export class ClineHandler implements ApiHandler {
 				this.lastGenerationId = chunk.id
 			}
 
-			const delta = chunk.choices[0]?.delta
+			// Check for mid-stream error via finish_reason
+			const choice = chunk.choices?.[0]
+			// OpenRouter may return finish_reason = "error" with error details
+			if ((choice?.finish_reason as string) === "error") {
+				const choiceWithError = choice as any
+				if (choiceWithError.error) {
+					const error = choiceWithError.error
+					console.error(`Cline Mid-Stream Error: ${error.code || error.type || "Unknown"} - ${error.message}`)
+					throw new Error(`Cline Mid-Stream Error: ${error.code || error.type || "Unknown"} - ${error.message}`)
+				} else {
+					throw new Error("Cline Mid-Stream Error: Stream terminated with error status but no error details provided")
+				}
+			}
+
+			const delta = choice?.delta
 			if (delta?.content) {
 				yield {
 					type: "text",
@@ -75,7 +89,7 @@ export class ClineHandler implements ApiHandler {
 
 			if (!didOutputUsage && chunk.usage) {
 				// @ts-ignore-next-line
-				let totalCost = chunk.usage.cost || 0
+				let totalCost = (chunk.usage.cost || 0) + (chunk.usage.cost_details?.upstream_inference_cost || 0)
 				const modelId = this.getModel().id
 				const provider = modelId.split("/")[0]
 
@@ -84,14 +98,26 @@ export class ClineHandler implements ApiHandler {
 					totalCost = 0
 				}
 
-				yield {
-					type: "usage",
-					cacheWriteTokens: 0,
-					cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
-					inputTokens: chunk.usage.prompt_tokens || 0,
-					outputTokens: chunk.usage.completion_tokens || 0,
-					// @ts-ignore-next-line
-					totalCost,
+				if (modelId.includes("gemini")) {
+					yield {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
+						inputTokens: (chunk.usage.prompt_tokens || 0) - (chunk.usage.prompt_tokens_details?.cached_tokens || 0),
+						outputTokens: chunk.usage.completion_tokens || 0,
+						// @ts-ignore-next-line
+						totalCost,
+					}
+				} else {
+					yield {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
+						inputTokens: chunk.usage.prompt_tokens || 0,
+						outputTokens: chunk.usage.completion_tokens || 0,
+						// @ts-ignore-next-line
+						totalCost,
+					}
 				}
 				didOutputUsage = true
 			}
@@ -117,13 +143,27 @@ export class ClineHandler implements ApiHandler {
 				})
 
 				const generation = response.data
-				return {
-					type: "usage",
-					cacheWriteTokens: 0,
-					cacheReadTokens: generation?.native_tokens_cached || 0,
-					inputTokens: generation?.native_tokens_prompt || 0,
-					outputTokens: generation?.native_tokens_completion || 0,
-					totalCost: generation?.total_cost || 0,
+				let modelId = this.options.openRouterModelId
+				if (modelId && modelId.includes("gemini")) {
+					return {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: generation?.native_tokens_cached || 0,
+						// openrouter generation endpoint fails often
+						inputTokens: (generation?.native_tokens_prompt || 0) - (generation?.native_tokens_cached || 0),
+						outputTokens: generation?.native_tokens_completion || 0,
+						totalCost: generation?.total_cost || 0,
+					}
+				} else {
+					return {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: generation?.native_tokens_cached || 0,
+						// openrouter generation endpoint fails often
+						inputTokens: generation?.native_tokens_prompt || 0,
+						outputTokens: generation?.native_tokens_completion || 0,
+						totalCost: generation?.total_cost || 0,
+					}
 				}
 			} catch (error) {
 				// ignore if fails

src/api/providers/gemini-cli.ts

@@ -0,0 +1,418 @@
+/**
+ * Gemini CLI Provider - OAuth-based API Handler
+ *
+ * This implementation provides access to Google's Gemini models through OAuth authentication,
+ * leveraging the same authentication mechanism as the official Gemini CLI tool.
+ *
+ * Attribution: This implementation is inspired by and uses concepts from the Google Gemini CLI,
+ * which is licensed under the Apache License 2.0.
+ * Original project: https://github.com/google-gemini/gemini-cli
+ *
+ * Copyright 2025 Google LLC
+ * Licensed under the Apache License, Version 2.0
+ *
+ * Key features:
+ * - OAuth2 authentication (no API keys required)
+ * - Auto-discovery of Google Cloud project IDs
+ * - Real-time streaming via Server-Sent Events
+ * - Free tier access through Google's Code Assist API
+ * - Compatible with personal Google accounts only
+ */
+
+import type { Anthropic } from "@anthropic-ai/sdk"
+import { OAuth2Client } from "google-auth-library"
+import fs from "fs/promises"
+import path from "path"
+import os from "os"
+import * as readline from "readline"
+import { Readable } from "stream"
+import { ApiHandler } from "../"
+import { ApiHandlerOptions, GeminiCliModelId, geminiCliModels, ModelInfo, geminiCliDefaultModelId } from "@shared/api"
+import { convertAnthropicMessageToGemini } from "../transform/gemini-format"
+import { ApiStream } from "../transform/stream"
+import { withRetry } from "../retry"
+
+const CODE_ASSIST_ENDPOINT = "https://cloudcode-pa.googleapis.com"
+const CODE_ASSIST_API_VERSION = "v1internal"
+
+// OAuth configuration
+const OAUTH_CLIENT_ID = "681255809395-oo8ft2oprdrnp9e3aqf6av3hmdib135j.apps.googleusercontent.com"
+// Change this line in setup.js:
+const OAUTH_CLIENT_SECRET = "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"
+
+const OAUTH_REDIRECT_URI = "http://localhost:45289"
+
+interface OAuthCredentials {
+	access_token: string
+	refresh_token: string
+	scope: string
+	token_type: string
+	expiry_date: number
+}
+
+interface GeminiCliHandlerOptions extends ApiHandlerOptions {
+	geminiCliOAuthPath?: string
+	geminiCliProjectId?: string
+}
+
+/**
+ * Handler for Google's Gemini API via OAuth (Gemini CLI style).
+ *
+ * This provider uses OAuth authentication instead of API keys, making it suitable
+ * for users who have already authenticated with the Gemini CLI tool.
+ * It automatically discovers project IDs and works with the free tier.
+ */
+export class GeminiCliHandler implements ApiHandler {
+	private options: GeminiCliHandlerOptions
+	private authClient: OAuth2Client
+	private projectId: string | null = null
+	private authInitialized: boolean = false
+
+	constructor(options: GeminiCliHandlerOptions) {
+		this.options = options
+		this.authClient = new OAuth2Client(OAUTH_CLIENT_ID, OAUTH_CLIENT_SECRET, OAUTH_REDIRECT_URI)
+	}
+
+	/**
+	 * Load OAuth credentials from the file system
+	 */
+	private async loadOAuthCredentials(): Promise<OAuthCredentials> {
+		const credPath = this.options.geminiCliOAuthPath || path.join(os.homedir(), ".gemini", "oauth_creds.json")
+		try {
+			const data = await fs.readFile(credPath, "utf8")
+			return JSON.parse(data)
+		} catch (err) {
+			throw new Error(`Failed to load OAuth credentials from ${credPath}. Please authenticate with 'gemini auth' first.`)
+		}
+	}
+
+	/**
+	 * Call a Code Assist API endpoint
+	 */
+	private async callEndpoint(method: string, body: any, retryAuth: boolean = true): Promise<any> {
+		console.log(`[GeminiCLI] Calling endpoint: ${method}`)
+		console.log(`[GeminiCLI] Request body:`, JSON.stringify(body, null, 2))
+
+		try {
+			const res = await this.authClient.request({
+				url: `${CODE_ASSIST_ENDPOINT}/${CODE_ASSIST_API_VERSION}:${method}`,
+				method: "POST",
+				headers: {
+					"Content-Type": "application/json",
+				},
+				responseType: "json",
+				body: JSON.stringify(body),
+			})
+			console.log(`[GeminiCLI] Response status:`, res.status)
+			console.log(`[GeminiCLI] Response data:`, JSON.stringify(res.data, null, 2))
+			return res.data
+		} catch (error: any) {
+			console.error(`[GeminiCLI] Error calling ${method}:`, error)
+			console.error(`[GeminiCLI] Error response:`, error.response?.data)
+			console.error(`[GeminiCLI] Error status:`, error.response?.status)
+			console.error(`[GeminiCLI] Error message:`, error.message)
+
+			// If we get a 401 and haven't retried yet, try refreshing auth
+			if (error.response?.status === 401 && retryAuth) {
+				console.log(`[GeminiCLI] Got 401, attempting to refresh authentication...`)
+				await this.initializeAuth(true) // Force refresh
+				return this.callEndpoint(method, body, false) // Retry without further auth retries
+			}
+
+			throw error
+		}
+	}
+
+	/**
+	 * Discover or retrieve the project ID
+	 */
+	private async discoverProjectId(): Promise<string> {
+		// If we already have a project ID, use it
+		if (this.options.geminiCliProjectId) {
+			return this.options.geminiCliProjectId
+		}
+
+		// If we've already discovered it, return it
+		if (this.projectId) {
+			return this.projectId
+		}
+
+		// Start with a default project ID (can be anything for personal OAuth)
+		const initialProjectId = "default"
+
+		// Prepare client metadata
+		const clientMetadata = {
+			ideType: "IDE_UNSPECIFIED",
+			platform: "PLATFORM_UNSPECIFIED",
+			pluginType: "GEMINI",
+			duetProject: initialProjectId,
+		}
+
+		try {
+			// Call loadCodeAssist to discover the actual project ID
+			const loadRequest = {
+				cloudaicompanionProject: initialProjectId,
+				metadata: clientMetadata,
+			}
+
+			const loadResponse = await this.callEndpoint("loadCodeAssist", loadRequest)
+
+			// Check if we already have a project ID from the response
+			if (loadResponse.cloudaicompanionProject) {
+				this.projectId = loadResponse.cloudaicompanionProject
+				return this.projectId as string
+			}
+
+			// If no existing project, we need to onboard
+			const defaultTier = loadResponse.allowedTiers?.find((tier: any) => tier.isDefault)
+			const tierId = defaultTier?.id || "free-tier"
+
+			const onboardRequest = {
+				tierId: tierId,
+				cloudaicompanionProject: initialProjectId,
+				metadata: clientMetadata,
+			}
+
+			let lroResponse = await this.callEndpoint("onboardUser", onboardRequest)
+
+			// Poll until operation is complete
+			while (!lroResponse.done) {
+				await new Promise((resolve) => setTimeout(resolve, 2000))
+				lroResponse = await this.callEndpoint("onboardUser", onboardRequest)
+			}
+
+			const discoveredProjectId = lroResponse.response?.cloudaicompanionProject?.id || initialProjectId
+			this.projectId = discoveredProjectId
+			return this.projectId as string
+		} catch (error: any) {
+			console.error("Failed to discover project ID:", error.response?.data || error.message)
+			throw new Error("Could not discover project ID. Make sure you're authenticated with 'gemini auth'.")
+		}
+	}
+
+	/**
+	 * Initialize the OAuth client with credentials
+	 */
+	private async initializeAuth(forceRefresh: boolean = false): Promise<void> {
+		// Check if we need to initialize or refresh
+		if (this.authInitialized && !forceRefresh) {
+			// Check if token is still valid
+			const credentials = this.authClient.credentials
+			if (credentials && credentials.expiry_date && Date.now() < credentials.expiry_date) {
+				console.log(`[GeminiCLI] Auth already initialized and token still valid`)
+				return
+			}
+		}
+
+		console.log(`[GeminiCLI] Initializing OAuth authentication...`)
+		const credentials = await this.loadOAuthCredentials()
+		const isExpired = credentials.expiry_date ? Date.now() > credentials.expiry_date : false
+
+		console.log(`[GeminiCLI] Loaded credentials:`, {
+			hasAccessToken: !!credentials.access_token,
+			hasRefreshToken: !!credentials.refresh_token,
+			tokenType: credentials.token_type,
+			expiryDate: credentials.expiry_date,
+			isExpired: isExpired,
+		})
+
+		this.authClient.setCredentials(credentials)
+
+		// If token is expired and we have a refresh token, try to refresh
+		if (isExpired && credentials.refresh_token) {
+			console.log(`[GeminiCLI] Token expired, attempting to refresh...`)
+			try {
+				const { credentials: newCredentials } = await this.authClient.refreshAccessToken()
+				console.log(`[GeminiCLI] Token refreshed successfully`)
+				// Note: In a real implementation, you'd want to save the new credentials back to the file
+				// For now, we'll just use them in memory
+			} catch (error) {
+				console.error(`[GeminiCLI] Failed to refresh token:`, error)
+				// Continue with the expired token - the API might still accept it
+			}
+		}
+
+		this.authInitialized = true
+		console.log(`[GeminiCLI] OAuth client configured`)
+	}
+
+	/**
+	 * Parse Server-Sent Events from a stream
+	 */
+	private async *parseSSEStream(stream: Readable): AsyncGenerator<any> {
+		const rl = readline.createInterface({
+			input: stream,
+			crlfDelay: Infinity,
+		})
+
+		let bufferedLines: string[] = []
+
+		for await (const line of rl) {
+			// Blank lines separate JSON objects in the stream
+			if (line === "") {
+				if (bufferedLines.length === 0) {
+					continue
+				}
+
+				try {
+					const jsonData = JSON.parse(bufferedLines.join("\n"))
+					yield jsonData
+				} catch (parseError) {
+					console.error("Error parsing JSON chunk:", parseError)
+				}
+
+				bufferedLines = []
+			} else if (line.startsWith("data: ")) {
+				bufferedLines.push(line.slice(6).trim())
+			}
+		}
+
+		// Process any remaining buffered content
+		if (bufferedLines.length > 0) {
+			try {
+				const jsonData = JSON.parse(bufferedLines.join("\n"))
+				yield jsonData
+			} catch (parseError) {
+				console.error("Error parsing final buffered content:", parseError)
+			}
+		}
+	}
+
+	/**
+	 * Create a message using the Gemini CLI OAuth API
+	 */
+	@withRetry({
+		maxRetries: 2,
+		baseDelay: 2000,
+		maxDelay: 10000,
+	})
+	async *createMessage(systemPrompt: string, messages: Anthropic.Messages.MessageParam[]): ApiStream {
+		// Initialize auth if not already done
+		await this.initializeAuth()
+		// Discover project ID if needed
+		const projectId = await this.discoverProjectId()
+
+		// Convert messages to Gemini format
+		const contents = messages.map(convertAnthropicMessageToGemini)
+
+		// Get the selected model
+		const { id: modelId, info: modelInfo } = this.getModel()
+
+		// Build the request
+		const streamRequest = {
+			model: modelId,
+			project: projectId,
+			request: {
+				contents: [
+					{
+						role: "user",
+						parts: [{ text: systemPrompt }],
+					},
+					...contents,
+				],
+				generationConfig: {
+					temperature: 0.7,
+					maxOutputTokens: modelInfo.maxTokens || 8192,
+				},
+			},
+		}
+
+		let totalContent = ""
+		let promptTokens = 0
+		let outputTokens = 0
+		let lastUsageMetadata: any = null
+
+		try {
+			// Make the streaming request
+			const response = await this.authClient.request({
+				url: `${CODE_ASSIST_ENDPOINT}/${CODE_ASSIST_API_VERSION}:streamGenerateContent`,
+				method: "POST",
+				params: { alt: "sse" },
+				headers: {
+					"Content-Type": "application/json",
+				},
+				responseType: "stream",
+				body: JSON.stringify(streamRequest),
+			})
+
+			// Process the SSE stream
+			for await (const jsonData of this.parseSSEStream(response.data as Readable)) {
+				// Extract content from the response
+				const candidate = jsonData.response?.candidates?.[0]
+				if (candidate?.content?.parts?.[0]?.text) {
+					const content = candidate.content.parts[0].text
+					totalContent += content
+
+					// Yield text chunk
+					yield {
+						type: "text",
+						text: content,
+					}
+				}
+
+				// Store usage metadata for final reporting
+				if (jsonData.response?.usageMetadata) {
+					lastUsageMetadata = jsonData.response.usageMetadata
+					promptTokens = lastUsageMetadata.promptTokenCount || promptTokens
+					outputTokens = lastUsageMetadata.candidatesTokenCount || outputTokens
+				}
+
+				// Check if this is the final chunk
+				if (candidate?.finishReason) {
+					break
+				}
+			}
+
+			// Yield usage information
+			if (lastUsageMetadata) {
+				yield {
+					type: "usage",
+					inputTokens: promptTokens,
+					outputTokens: outputTokens,
+					totalCost: 0, // Free tier
+				}
+			}
+		} catch (error) {
+			// Handle rate limit errors similar to the Gemini provider
+			if (error instanceof Error) {
+				// Check for rate limit patterns in the error message
+				const rateLimitPatterns = [
+					/got status: 429/i,
+					/429 Too Many Requests/i,
+					/rate limit exceeded/i,
+					/too many requests/i,
+					/quota exceeded/i,
+					/resource exhausted/i,
+					/code 429/i,
+				]
+
+				const isRateLimit = rateLimitPatterns.some((pattern) => pattern.test(error.message))
+
+				if (isRateLimit) {
+					const rateLimitError = Object.assign(new Error(error.message), {
+						...error,
+						status: 429,
+					})
+					throw rateLimitError
+				}
+			}
+
+			// Re-throw the original error if it's not a rate limit error
+			throw error
+		}
+	}
+
+	/**
+	 * Get the model ID and info
+	 */
+	getModel(): { id: GeminiCliModelId; info: ModelInfo } {
+		const modelId = this.options.apiModelId as GeminiCliModelId
+		if (modelId && modelId in geminiCliModels) {
+			return { id: modelId, info: geminiCliModels[modelId] }
+		}
+		return {
+			id: geminiCliDefaultModelId,
+			info: geminiCliModels[geminiCliDefaultModelId],
+		}
+	}
+}

src/api/providers/openrouter.ts

@@ -44,6 +44,7 @@ export class OpenRouterHandler implements ApiHandler {
 
 		for await (const chunk of stream) {
 			// openrouter returns an error object instead of the openai sdk throwing an error
+			// Check for error field directly on chunk
 			if ("error" in chunk) {
 				const error = chunk.error as OpenRouterErrorResponse["error"]
 				console.error(`OpenRouter API Error: ${error?.code} - ${error?.message}`)
@@ -52,6 +53,29 @@ export class OpenRouterHandler implements ApiHandler {
 				throw new Error(`OpenRouter API Error ${error.code}: ${error.message}${metadataStr}`)
 			}
 
+			// Check for error in choices[0].finish_reason
+			// OpenRouter may return errors in a non-standard way within choices
+			const choice = chunk.choices?.[0]
+			// Use type assertion since OpenRouter uses non-standard "error" finish_reason
+			if ((choice?.finish_reason as string) === "error") {
+				// Use type assertion since OpenRouter adds non-standard error property
+				const choiceWithError = choice as any
+				if (choiceWithError.error) {
+					const error = choiceWithError.error
+					console.error(
+						`OpenRouter Mid-Stream Error: ${error?.code || "Unknown"} - ${error?.message || "Unknown error"}`,
+					)
+					// Format error details
+					const errorDetails = typeof error === "object" ? JSON.stringify(error, null, 2) : String(error)
+					throw new Error(`OpenRouter Mid-Stream Error: ${errorDetails}`)
+				} else {
+					// Fallback if error details are not available
+					throw new Error(
+						`OpenRouter Mid-Stream Error: Stream terminated with error status but no error details provided`,
+					)
+				}
+			}
+
 			if (!this.lastGenerationId && chunk.id) {
 				this.lastGenerationId = chunk.id
 			}
@@ -74,14 +98,27 @@ export class OpenRouterHandler implements ApiHandler {
 			}
 
 			if (!didOutputUsage && chunk.usage) {
-				yield {
-					type: "usage",
-					cacheWriteTokens: 0,
-					cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
-					inputTokens: chunk.usage.prompt_tokens || 0,
-					outputTokens: chunk.usage.completion_tokens || 0,
-					// @ts-ignore-next-line
-					totalCost: chunk.usage.cost || 0,
+				let modelId = this.options.openRouterModelId
+				if (modelId && modelId.includes("gemini")) {
+					yield {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
+						inputTokens: (chunk.usage.prompt_tokens || 0) - (chunk.usage.prompt_tokens_details?.cached_tokens || 0),
+						outputTokens: chunk.usage.completion_tokens || 0,
+						// @ts-ignore-next-line
+						totalCost: (chunk.usage.cost || 0) + (chunk.usage.cost_details?.upstream_inference_cost || 0),
+					}
+				} else {
+					yield {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: chunk.usage.prompt_tokens_details?.cached_tokens || 0,
+						inputTokens: chunk.usage.prompt_tokens || 0,
+						outputTokens: chunk.usage.completion_tokens || 0,
+						// @ts-ignore-next-line
+						totalCost: (chunk.usage.cost || 0) + (chunk.usage.cost_details?.upstream_inference_cost || 0),
+					}
 				}
 				didOutputUsage = true
 			}
@@ -103,14 +140,27 @@ export class OpenRouterHandler implements ApiHandler {
 				const generationIterator = this.fetchGenerationDetails(this.lastGenerationId)
 				const generation = (await generationIterator.next()).value
 				// console.log("OpenRouter generation details:", generation)
-				return {
-					type: "usage",
-					cacheWriteTokens: 0,
-					cacheReadTokens: generation?.native_tokens_cached || 0,
-					// openrouter generation endpoint fails often
-					inputTokens: generation?.native_tokens_prompt || 0,
-					outputTokens: generation?.native_tokens_completion || 0,
-					totalCost: generation?.total_cost || 0,
+				let modelId = this.options.openRouterModelId
+				if (modelId && modelId.includes("gemini")) {
+					return {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: generation?.native_tokens_cached || 0,
+						// openrouter generation endpoint fails often
+						inputTokens: (generation?.native_tokens_prompt || 0) - (generation?.native_tokens_cached || 0),
+						outputTokens: generation?.native_tokens_completion || 0,
+						totalCost: generation?.total_cost || 0,
+					}
+				} else {
+					return {
+						type: "usage",
+						cacheWriteTokens: 0,
+						cacheReadTokens: generation?.native_tokens_cached || 0,
+						// openrouter generation endpoint fails often
+						inputTokens: generation?.native_tokens_prompt || 0,
+						outputTokens: generation?.native_tokens_completion || 0,
+						totalCost: generation?.total_cost || 0,
+					}
 				}
 			} catch (error) {
 				// ignore if fails

src/api/providers/sapaicore.ts

@@ -122,6 +122,7 @@ export class SapAiCoreHandler implements ApiHandler {
 		const deploymentId = await this.getDeploymentForModel(model.id)
 
 		const anthropicModels = [
+			"anthropic--claude-4-sonnet",
 			"anthropic--claude-3.7-sonnet",
 			"anthropic--claude-3.5-sonnet",
 			"anthropic--claude-3-sonnet",
@@ -136,7 +137,7 @@ export class SapAiCoreHandler implements ApiHandler {
 		if (anthropicModels.includes(model.id)) {
 			url = `${this.options.sapAiCoreBaseUrl}/v2/inference/deployments/${deploymentId}/invoke-with-response-stream`
 
-			if (model.id === "anthropic--claude-3.7-sonnet") {
+			if (model.id === "anthropic--claude-3.7-sonnet" || model.id === "anthropic--claude-4-sonnet") {
 				url = `${this.options.sapAiCoreBaseUrl}/v2/inference/deployments/${deploymentId}/converse-stream`
 				payload = {
 					inferenceConfig: {
@@ -221,7 +222,7 @@ export class SapAiCoreHandler implements ApiHandler {
 				}
 			} else if (openAIModels.includes(model.id)) {
 				yield* this.streamCompletionGPT(response.data, model)
-			} else if (model.id === "anthropic--claude-3.7-sonnet") {
+			} else if (model.id === "anthropic--claude-3.7-sonnet" || model.id === "anthropic--claude-4-sonnet") {
 				yield* this.streamCompletionSonnet37(response.data, model)
 			} else {
 				yield* this.streamCompletion(response.data, model)

src/core/assistant-message/diff.ts

@@ -8,11 +8,13 @@ const LEGACY_SEARCH_BLOCK_CHAR = "<"
 const LEGACY_REPLACE_BLOCK_CHAR = ">"
 
 // Replace the exact string constants with flexible regex patterns
-const SEARCH_BLOCK_START_REGEX = /^[-]{3,} SEARCH$/
+const SEARCH_BLOCK_START_REGEX = /^[-]{3,} SEARCH>?$/
+const LEGACY_SEARCH_BLOCK_START_REGEX = /^[<]{3,} SEARCH>?$/
+
 const SEARCH_BLOCK_END_REGEX = /^[=]{3,}$/
-const REPLACE_BLOCK_END_REGEX = /^[+]{3,} REPLACE$/
-const LEGACY_SEARCH_BLOCK_START_REGEX = /^[<]{3,} SEARCH$/
-const LEGACY_REPLACE_BLOCK_END_REGEX = /^[>]{3,} REPLACE$/
+
+const REPLACE_BLOCK_END_REGEX = /^[+]{3,} REPLACE>?$/
+const LEGACY_REPLACE_BLOCK_END_REGEX = /^[>]{3,} REPLACE>?$/
 
 // Helper functions to check if a line matches the flexible patterns
 function isSearchBlockStart(line: string): boolean {

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

@@ -6,6 +6,9 @@ import * as path from "path"
 import { FileContextTracker } from "./FileContextTracker"
 import * as diskModule from "@core/storage/disk"
 import type { TaskMetadata, FileMetadataEntry } from "./ContextTrackerTypes"
+import type { WebviewProviderCreator } from "@/hosts/host-providers"
+import * as hostProviders from "@hosts/host-providers"
+import { vscodeHostBridgeClient } from "@/hosts/vscode/client/host-grpc-client"
 
 describe("FileContextTracker", () => {
 	let sandbox: sinon.SinonSandbox
@@ -37,7 +40,6 @@ describe("FileContextTracker", () => {
 		}
 
 		// Use a function replacement instead of a direct stub
-		const originalCreateFileSystemWatcher = vscode.workspace.createFileSystemWatcher
 		vscode.workspace.createFileSystemWatcher = function () {
 			return mockFileSystemWatcher
 		}
@@ -51,6 +53,7 @@ describe("FileContextTracker", () => {
 		mockTaskMetadata = { files_in_context: [], model_usage: [] }
 		getTaskMetadataStub = sandbox.stub(diskModule, "getTaskMetadata").resolves(mockTaskMetadata)
 		saveTaskMetadataStub = sandbox.stub(diskModule, "saveTaskMetadata").resolves()
+		hostProviders.initializeHostProviders(((_) => {}) as WebviewProviderCreator, vscodeHostBridgeClient)
 
 		// Create tracker instance
 		taskId = "test-task-id"

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

@@ -5,6 +5,8 @@ import { getWorkspaceState, updateWorkspaceState } from "@core/storage/state"
 import { getGlobalState } from "@core/storage/state"
 import type { FileMetadataEntry } from "./ContextTrackerTypes"
 import type { ClineMessage } from "@shared/ExtensionMessage"
+import { getHostBridgeProvider } from "@/hosts/host-providers"
+import { getCwd } from "@/utils/path"
 
 // 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.
@@ -37,8 +39,8 @@ export class FileContextTracker {
 	/**
 	 * 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)
+	private async getCwd(): Promise<string | undefined> {
+		const cwd = await getCwd(undefined)
 		if (!cwd) {
 			console.info("No workspace folder available - cannot determine current working directory")
 		}
@@ -54,7 +56,7 @@ export class FileContextTracker {
 			return
 		}
 
-		const cwd = this.getCwd()
+		const cwd = await this.getCwd()
 		if (!cwd) {
 			return
 		}
@@ -85,7 +87,7 @@ export class FileContextTracker {
 	 */
 	async trackFileContext(filePath: string, operation: "read_tool" | "user_edited" | "cline_edited" | "file_mentioned") {
 		try {
-			const cwd = this.getCwd()
+			const cwd = await this.getCwd()
 			if (!cwd) {
 				return
 			}

src/core/controller/file/copyToClipboard.ts

@@ -1,6 +1,7 @@
 import * as vscode from "vscode"
 import { Controller } from ".."
 import { Empty, StringRequest } from "../../../shared/proto/common"
+import { writeTextToClipboard } from "@/utils/env"
 
 /**
  * Copies text to the system clipboard
@@ -11,7 +12,7 @@ import { Empty, StringRequest } from "../../../shared/proto/common"
 export async function copyToClipboard(controller: Controller, request: StringRequest): Promise<Empty> {
 	try {
 		if (request.value) {
-			await vscode.env.clipboard.writeText(request.value)
+			await writeTextToClipboard(request.value)
 		}
 	} catch (error) {
 		console.error("Error copying to clipboard:", error)

src/core/controller/file/searchCommits.ts

@@ -13,7 +13,7 @@ import { convertGitCommitsToProtoGitCommits } from "@shared/proto-conversions/fi
  * @returns GitCommits containing the matching commits
  */
 export const searchCommits: FileMethodHandler = async (controller: Controller, request: StringRequest): Promise<GitCommits> => {
-	const cwd = getWorkspacePath()
+	const cwd = await getWorkspacePath()
 	if (!cwd) {
 		return GitCommits.create({ commits: [] })
 	}

src/core/controller/file/searchFiles.ts

@@ -12,10 +12,10 @@ import { convertSearchResultsToProtoFileInfos } from "@shared/proto-conversions/
  * @returns Results containing matching files/folders
  */
 export const searchFiles: FileMethodHandler = async (
-	controller: Controller,
+	_controller: Controller,
 	request: FileSearchRequest,
 ): Promise<FileSearchResults> => {
-	const workspacePath = getWorkspacePath()
+	const workspacePath = await getWorkspacePath()
 
 	if (!workspacePath) {
 		// Handle case where workspace path is not available

src/core/controller/file/selectImages.ts

@@ -1,21 +0,0 @@
-import { Controller } from ".."
-import { EmptyRequest, StringArray } from "@shared/proto/common"
-import { selectImages as selectImagesIntegration } from "@integrations/misc/process-images"
-import { FileMethodHandler } from "./index"
-
-/**
- * Prompts the user to select images from the file system and returns them as data URLs
- * @param controller The controller instance
- * @param request Empty request, no parameters needed
- * @returns Array of image data URLs
- */
-export const selectImages: FileMethodHandler = async (controller: Controller, request: EmptyRequest): Promise<StringArray> => {
-	try {
-		const images = await selectImagesIntegration()
-		return StringArray.create({ values: images })
-	} catch (error) {
-		console.error("Error selecting images:", error)
-		// Return empty array on error
-		return StringArray.create({ values: [] })
-	}
-}

src/core/controller/grpc-service.ts

@@ -37,6 +37,7 @@ export class ServiceRegistry {
 	 * @param serviceName The name of the service (used for logging)
 	 */
 	constructor(serviceName: string) {
+		console.log(`Registering Protobus service: ${serviceName}...`)
 		this.serviceName = serviceName
 	}
 
@@ -56,7 +57,6 @@ export class ServiceRegistry {
 		}
 
 		this.methodMetadata[methodName] = { isStreaming, ...metadata }
-		console.log(`Registered ${this.serviceName} method: ${methodName}${isStreaming ? " (streaming)" : ""}`)
 	}
 
 	/**

src/core/controller/index.ts

@@ -16,7 +16,7 @@ import { McpHub } from "@services/mcp/McpHub"
 import { telemetryService } from "@/services/posthog/telemetry/TelemetryService"
 import { ApiProvider, ModelInfo } from "@shared/api"
 import { ChatContent } from "@shared/ChatContent"
-import { ChatSettings } from "@shared/ChatSettings"
+import { ChatSettings, StoredChatSettings } from "@shared/ChatSettings"
 import { ExtensionMessage, ExtensionState, Platform } from "@shared/ExtensionMessage"
 import { HistoryItem } from "@shared/HistoryItem"
 import { McpMarketplaceCatalog } from "@shared/mcp"
@@ -44,6 +44,11 @@ import { sendAddToInputEvent } from "./ui/subscribeToAddToInput"
 import { sendAuthCallbackEvent } from "./account/subscribeToAuthCallback"
 import { sendMcpMarketplaceCatalogEvent } from "./mcp/subscribeToMcpMarketplaceCatalog"
 import { sendRelinquishControlEvent } from "./ui/subscribeToRelinquishControl"
+import { handleTaskServiceRequest } from "./task"
+import { BooleanRequest } from "@shared/proto/common"
+import { getHostBridgeProvider } from "@/hosts/host-providers"
+import { GetWorkspacePathsRequest } from "@/shared/proto/index.host"
+import { getCwd } from "@/utils/path"
 
 /*
 https://github.com/microsoft/vscode-webview-ui-toolkit-samples/blob/main/default/weather-webview/src/providers/WeatherViewProvider.ts
@@ -56,11 +61,12 @@ export class Controller {
 	private postMessage: (message: ExtensionMessage) => Thenable<boolean> | undefined
 
 	private disposables: vscode.Disposable[] = []
+	private mode: "plan" | "act" = "plan" // In-memory plan/act mode state
 	task?: Task
 	workspaceTracker: WorkspaceTracker
 	mcpHub: McpHub
 	accountService: ClineAccountService
-	latestAnnouncementId = "may-22-2025_16:11:00" // update to some unique identifier when we add a new announcement
+	latestAnnouncementId = "june-25-2025_16:11:00" // update to some unique identifier when we add a new announcement
 
 	constructor(
 		readonly context: vscode.ExtensionContext,
@@ -135,7 +141,7 @@ export class Controller {
 			apiConfiguration,
 			autoApprovalSettings,
 			browserSettings,
-			chatSettings,
+			chatSettings: storedChatSettings,
 			shellIntegrationTimeout,
 			terminalReuseEnabled,
 			terminalOutputLineLimit,
@@ -145,6 +151,12 @@ export class Controller {
 			taskHistory,
 		} = await getAllExtensionState(this.context)
 
+		// Reconstruct ChatSettings with in-memory mode and stored preferences
+		const chatSettings: ChatSettings = {
+			...storedChatSettings, // Spread stored preferences (preferredLanguage, openAIReasoningEffort)
+			mode: this.mode, // Use in-memory mode (override any stored mode)
+		}
+
 		const NEW_USER_TASK_COUNT_THRESHOLD = 10
 
 		// Check if the user has completed enough tasks to no longer be considered a "new user"
@@ -209,35 +221,6 @@ export class Controller {
 				await this.fetchMcpMarketplace(message.bool)
 				break
 			}
-
-			// telemetry
-			case "telemetrySetting": {
-				if (message.telemetrySetting) {
-					await this.updateTelemetrySetting(message.telemetrySetting)
-				}
-				await this.postStateToWebview()
-				break
-			}
-
-			case "clearAllTaskHistory": {
-				const answer = await vscode.window.showWarningMessage(
-					"What would you like to delete?",
-					{ modal: true },
-					"Delete All Except Favorites",
-					"Delete Everything",
-					"Cancel",
-				)
-
-				if (answer === "Delete All Except Favorites") {
-					await this.deleteNonFavoriteTaskHistory()
-					await this.postStateToWebview()
-				} else if (answer === "Delete Everything") {
-					await this.deleteAllTaskHistory()
-					await this.postStateToWebview()
-				}
-				sendRelinquishControlEvent()
-				break
-			}
 			case "grpc_request": {
 				if (message.grpc_request) {
 					await handleGrpcRequest(this, message.grpc_request)
@@ -265,6 +248,9 @@ export class Controller {
 	async togglePlanActModeWithChatSettings(chatSettings: ChatSettings, chatContent?: ChatContent): Promise<boolean> {
 		const didSwitchToActMode = chatSettings.mode === "act"
 
+		// Store mode in-memory only
+		this.mode = chatSettings.mode
+
 		// Capture mode switch telemetry | Capture regardless of if we know the taskId
 		telemetryService.captureModeSwitch(this.task?.taskId ?? "0", chatSettings.mode)
 
@@ -434,7 +420,9 @@ export class Controller {
 			}
 		}
 
-		await updateWorkspaceState(this.context, "chatSettings", chatSettings)
+		// Save only non-mode properties to workspace storage
+		const { mode, ...persistentChatSettings }: { mode: string } & StoredChatSettings = chatSettings
+		await updateWorkspaceState(this.context, "chatSettings", persistentChatSettings)
 		await this.postStateToWebview()
 
 		if (this.task) {
@@ -698,8 +686,8 @@ export class Controller {
 
 	// Context menus and code actions
 
-	getFileMentionFromPath(filePath: string) {
-		const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0)
+	async getFileMentionFromPath(filePath: string) {
+		const cwd = await getCwd()
 		if (!cwd) {
 			return "@/" + filePath
 		}
@@ -830,110 +818,6 @@ export class Controller {
 		await downloadTask(historyItem.ts, apiConversationHistory)
 	}
 
-	async deleteAllTaskHistory() {
-		await this.clearTask()
-		await updateGlobalState(this.context, "taskHistory", undefined)
-		try {
-			// Remove all contents of tasks directory
-			const taskDirPath = path.join(this.context.globalStorageUri.fsPath, "tasks")
-			if (await fileExistsAtPath(taskDirPath)) {
-				await fs.rm(taskDirPath, { recursive: true, force: true })
-			}
-			// Remove checkpoints directory contents
-			const checkpointsDirPath = path.join(this.context.globalStorageUri.fsPath, "checkpoints")
-			if (await fileExistsAtPath(checkpointsDirPath)) {
-				await fs.rm(checkpointsDirPath, { recursive: true, force: true })
-			}
-		} catch (error) {
-			vscode.window.showErrorMessage(
-				`Encountered error while deleting task history, there may be some files left behind. Error: ${error instanceof Error ? error.message : String(error)}`,
-			)
-		}
-		// await this.postStateToWebview()
-	}
-
-	async deleteNonFavoriteTaskHistory() {
-		await this.clearTask()
-
-		const taskHistory = ((await getGlobalState(this.context, "taskHistory")) as HistoryItem[]) || []
-		const favoritedTasks = taskHistory.filter((task) => task.isFavorited === true)
-
-		// If user has no favorited tasks, show a warning message
-		if (favoritedTasks.length === 0) {
-			vscode.window.showWarningMessage("No favorited tasks found. Please favorite tasks before using this option.")
-			await this.postStateToWebview()
-			return
-		}
-
-		await updateGlobalState(this.context, "taskHistory", favoritedTasks)
-
-		// Delete non-favorited task directories
-		try {
-			const preserveTaskIds = favoritedTasks.map((task) => task.id)
-			const taskDirPath = path.join(this.context.globalStorageUri.fsPath, "tasks")
-
-			if (await fileExistsAtPath(taskDirPath)) {
-				const taskDirs = await fs.readdir(taskDirPath)
-				for (const taskDir of taskDirs) {
-					if (!preserveTaskIds.includes(taskDir)) {
-						await fs.rm(path.join(taskDirPath, taskDir), { recursive: true, force: true })
-					}
-				}
-			}
-		} catch (error) {
-			vscode.window.showErrorMessage(
-				`Error deleting task history: ${error instanceof Error ? error.message : String(error)}`,
-			)
-		}
-
-		await this.postStateToWebview()
-	}
-
-	async deleteTaskWithId(id: string) {
-		console.info("deleteTaskWithId: ", id)
-
-		try {
-			if (id === this.task?.taskId) {
-				await this.clearTask()
-				console.debug("cleared task")
-			}
-
-			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
-			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
-
-			if (updatedTaskHistory.length === 0) {
-				await this.deleteAllTaskHistory()
-			}
-		} catch (error) {
-			console.debug(`Error deleting task:`, error)
-		}
-
-		await this.postStateToWebview()
-	}
-
 	async deleteTaskFromState(id: string) {
 		// Remove the task from history
 		const taskHistory = ((await getGlobalState(this.context, "taskHistory")) as HistoryItem[] | undefined) || []
@@ -948,7 +832,7 @@ export class Controller {
 
 	async postStateToWebview() {
 		const state = await this.getStateToPostToWebview()
-		await sendStateUpdate(state)
+		await sendStateUpdate(this.id, state)
 	}
 
 	async getStateToPostToWebview(): Promise<ExtensionState> {
@@ -958,7 +842,7 @@ export class Controller {
 			taskHistory,
 			autoApprovalSettings,
 			browserSettings,
-			chatSettings,
+			chatSettings: storedChatSettings,
 			userInfo,
 			mcpMarketplaceEnabled,
 			mcpRichDisplayEnabled,
@@ -975,6 +859,12 @@ export class Controller {
 			terminalOutputLineLimit,
 		} = await getAllExtensionState(this.context)
 
+		// Reconstruct ChatSettings with in-memory mode and stored preferences
+		const chatSettings: ChatSettings = {
+			...storedChatSettings, // Spread stored preferences (preferredLanguage, openAIReasoningEffort)
+			mode: this.mode, // Use in-memory mode (override any stored mode)
+		}
+
 		const localClineRulesToggles =
 			((await getWorkspaceState(this.context, "localClineRulesToggles")) as ClineRulesToggles) || {}
 
@@ -1101,7 +991,7 @@ export class Controller {
 	async generateGitCommitMessage() {
 		try {
 			// Check if there's a workspace folder open
-			const cwd = vscode.workspace.workspaceFolders?.[0]?.uri.fsPath
+			const cwd = await getCwd()
 			if (!cwd) {
 				vscode.window.showErrorMessage("No workspace folder open")
 				return

src/core/controller/state/subscribeToState.ts

@@ -3,8 +3,8 @@ import { Controller } from "../index"
 import { EmptyRequest } from "../../../shared/proto/common"
 import { StreamingResponseHandler, getRequestRegistry } from "../grpc-handler"
 
-// Keep track of active state subscriptions
-const activeStateSubscriptions = new Set<StreamingResponseHandler>()
+// Keep track of active state subscriptions by controller ID
+const activeStateSubscriptions = new Map<string, StreamingResponseHandler>()
 
 /**
  * Subscribe to state updates
@@ -19,23 +19,25 @@ export async function subscribeToState(
 	responseStream: StreamingResponseHandler,
 	requestId?: string,
 ): Promise<void> {
+	const controllerId = controller.id
+
 	// Send the initial state
 	const initialState = await controller.getStateToPostToWebview()
 	const initialStateJson = JSON.stringify(initialState)
 
-	console.log("[DEBUG] set up state subscription")
+	console.log(`[DEBUG] set up state subscription for controller ${controllerId}`)
 
 	await responseStream({
 		stateJson: initialStateJson,
 	})
 
-	// Add this subscription to the active subscriptions
-	activeStateSubscriptions.add(responseStream)
+	// Add this subscription to the active subscriptions with the controller ID
+	activeStateSubscriptions.set(controllerId, responseStream)
 
 	// Register cleanup when the connection is closed
 	const cleanup = () => {
-		activeStateSubscriptions.delete(responseStream)
-		console.log("[DEBUG] Cleaned up state subscription")
+		activeStateSubscriptions.delete(controllerId)
+		console.log(`[DEBUG] Cleaned up state subscription for controller ${controllerId}`)
 	}
 
 	// Register the cleanup function with the request registry if we have a requestId
@@ -45,30 +47,31 @@ export async function subscribeToState(
 }
 
 /**
- * Send a state update to all active subscribers
+ * Send a state update to a specific controller's subscription
+ * @param controllerId The ID of the controller to send the state to
  * @param state The state to send
  */
-export async function sendStateUpdate(state: any): Promise<void> {
-	const stateJson = JSON.stringify(state)
+export async function sendStateUpdate(controllerId: string, state: any): Promise<void> {
+	// Get the subscription for this specific controller
+	const responseStream = activeStateSubscriptions.get(controllerId)
 
-	// Send the update to all active subscribers
-	const promises = Array.from(activeStateSubscriptions).map(async (responseStream) => {
-		try {
-			// The issue might be that we're not properly formatting the response
-			// Let's ensure we're sending a properly formatted State message
-			await responseStream(
-				{
-					stateJson,
-				},
-				false, // Not the last message
-			)
-			console.log("[DEBUG] sending followup state", stateJson.length, "chars")
-		} catch (error) {
-			console.error("Error sending state update:", error)
-			// Remove the subscription if there was an error
-			activeStateSubscriptions.delete(responseStream)
-		}
-	})
+	if (!responseStream) {
+		console.log(`[DEBUG] No active state subscription for controller ${controllerId}`)
+		return
+	}
 
-	await Promise.all(promises)
+	try {
+		const stateJson = JSON.stringify(state)
+		await responseStream(
+			{
+				stateJson,
+			},
+			false, // Not the last message
+		)
+		console.log(`[DEBUG] sending followup state to controller ${controllerId}`, stateJson.length, "chars")
+	} catch (error) {
+		console.error(`Error sending state update to controller ${controllerId}:`, error)
+		// Remove the subscription if there was an error
+		activeStateSubscriptions.delete(controllerId)
+	}
 }

src/core/controller/state/updateAutoApprovalSettings.ts

@@ -22,7 +22,7 @@ export async function updateAutoApprovalSettings(controller: Controller, request
 		await updateGlobalState(controller.context, "autoApprovalSettings", settings)
 
 		if (controller.task) {
-			controller.task.autoApprovalSettings = settings
+			controller.task.updateAutoApprovalSettings(settings)
 		}
 
 		await controller.postStateToWebview()

src/core/controller/state/updateTelemetrySetting.ts

@@ -0,0 +1,16 @@
+import { Controller } from ".."
+import { Empty } from "../../../shared/proto/common"
+import { TelemetrySettingRequest } from "../../../shared/proto/state"
+import { convertProtoTelemetrySettingToDomain } from "../../../shared/proto-conversions/state/telemetry-setting-conversion"
+
+/**
+ * Updates the telemetry setting
+ * @param controller The controller instance
+ * @param request The telemetry setting request
+ * @returns Empty response
+ */
+export async function updateTelemetrySetting(controller: Controller, request: TelemetrySettingRequest): Promise<Empty> {
+	const telemetrySetting = convertProtoTelemetrySettingToDomain(request.setting)
+	await controller.updateTelemetrySetting(telemetrySetting)
+	return Empty.create()
+}

src/core/controller/task/deleteAllTaskHistory.ts

@@ -0,0 +1,138 @@
+import path from "path"
+import fs from "fs/promises"
+import { Controller } from ".."
+import { DeleteAllTaskHistoryCount } from "../../../shared/proto/task"
+import { getGlobalState, updateGlobalState } from "../../storage/state"
+import { fileExistsAtPath } from "../../../utils/fs"
+import vscode from "vscode"
+
+/**
+ * Deletes all task history, with an option to preserve favorites
+ * @param controller The controller instance
+ * @param request Request with option to preserve favorites
+ * @returns Results with count of deleted tasks
+ */
+export async function deleteAllTaskHistory(controller: Controller): Promise<DeleteAllTaskHistoryCount> {
+	try {
+		// Clear current task first
+		await controller.clearTask()
+
+		// Get existing task history
+		const taskHistory = ((await getGlobalState(controller.context, "taskHistory")) as any[]) || []
+		const totalTasks = taskHistory.length
+
+		const userChoice = await vscode.window.showWarningMessage(
+			"What would you like to delete?",
+			{ modal: true },
+			"Delete All Except Favorites",
+			"Delete Everything",
+		)
+
+		// Default VS Code Cancel button returns `undefined` - don't delete anything
+		if (userChoice === undefined) {
+			return DeleteAllTaskHistoryCount.create({
+				tasksDeleted: 0,
+			})
+		}
+
+		// If preserving favorites, filter out non-favorites
+		if (userChoice === "Delete All Except Favorites") {
+			const favoritedTasks = taskHistory.filter((task) => task.isFavorited === true)
+
+			// If there are favorited tasks, update state
+			if (favoritedTasks.length > 0) {
+				await updateGlobalState(controller.context, "taskHistory", favoritedTasks)
+
+				// Delete non-favorited task directories
+				const preserveTaskIds = favoritedTasks.map((task) => task.id)
+				await cleanupTaskFiles(controller, preserveTaskIds)
+
+				// Update webview
+				try {
+					await controller.postStateToWebview()
+				} catch (webviewErr) {
+					console.error("Error posting to webview:", webviewErr)
+				}
+
+				return DeleteAllTaskHistoryCount.create({
+					tasksDeleted: totalTasks - favoritedTasks.length,
+				})
+			} else {
+				// No favorited tasks found - show warning and ask user what to do
+				const answer = await vscode.window.showWarningMessage(
+					"No favorited tasks found. Would you like to delete all tasks anyway?",
+					{ modal: true },
+					"Delete All Tasks",
+				)
+
+				// User cancelled - don't delete anything
+				if (answer === undefined) {
+					return DeleteAllTaskHistoryCount.create({
+						tasksDeleted: 0,
+					})
+				}
+				// If user chose "Delete All Tasks", fall through to the `delete everything` section below
+			}
+		}
+
+		// Delete everything (not preserving favorites)
+		await updateGlobalState(controller.context, "taskHistory", undefined)
+
+		try {
+			// Remove all contents of tasks directory
+			const taskDirPath = path.join(controller.context.globalStorageUri.fsPath, "tasks")
+			if (await fileExistsAtPath(taskDirPath)) {
+				await fs.rm(taskDirPath, { recursive: true, force: true })
+			}
+
+			// Remove checkpoints directory contents
+			const checkpointsDirPath = path.join(controller.context.globalStorageUri.fsPath, "checkpoints")
+			if (await fileExistsAtPath(checkpointsDirPath)) {
+				await fs.rm(checkpointsDirPath, { recursive: true, force: true })
+			}
+		} catch (error) {
+			vscode.window.showErrorMessage(
+				`Encountered error while deleting task history, there may be some files left behind. Error: ${error instanceof Error ? error.message : String(error)}`,
+			)
+		}
+
+		// Update webview
+		try {
+			await controller.postStateToWebview()
+		} catch (webviewErr) {
+			console.error("Error posting to webview:", webviewErr)
+		}
+
+		return DeleteAllTaskHistoryCount.create({
+			tasksDeleted: totalTasks,
+		})
+	} catch (error) {
+		console.error("Error in deleteAllTaskHistory:", error)
+		throw error
+	}
+}
+
+/**
+ * Helper function to cleanup task files while preserving specified tasks
+ */
+async function cleanupTaskFiles(controller: Controller, preserveTaskIds: string[]) {
+	const taskDirPath = path.join(controller.context.globalStorageUri.fsPath, "tasks")
+
+	try {
+		if (await fileExistsAtPath(taskDirPath)) {
+			const taskDirs = await fs.readdir(taskDirPath)
+			console.debug(`[cleanupTaskFiles] Found ${taskDirs.length} task directories`)
+
+			// Delete only non-preserved task directories
+			for (const dir of taskDirs) {
+				if (!preserveTaskIds.includes(dir)) {
+					await fs.rm(path.join(taskDirPath, dir), { recursive: true, force: true })
+				}
+			}
+		}
+	} catch (error) {
+		console.error("Error cleaning up task files:", error)
+	}
+
+	return true
+}

src/core/controller/task/deleteNonFavoritedTasks.ts

@@ -1,88 +0,0 @@
-import path from "path"
-import fs from "fs/promises"
-import { Controller } from ".."
-import { EmptyRequest } from "../../../shared/proto/common"
-import { DeleteNonFavoritedTasksResults } from "../../../shared/proto/task"
-import { getGlobalState, updateGlobalState } from "../../storage/state"
-import { fileExistsAtPath } from "../../../utils/fs"
-
-/**
- * Deletes all non-favorited tasks, preserving only favorited ones
- * @param controller The controller instance
- * @param request Empty request
- * @returns DeleteNonFavoritedTasksResults with counts of preserved and deleted tasks
- */
-export async function deleteNonFavoritedTasks(
-	controller: Controller,
-	_request: EmptyRequest,
-): Promise<DeleteNonFavoritedTasksResults> {
-	try {
-		// Clear current task first
-		await controller.clearTask()
-
-		// Get existing task history
-		const taskHistory = ((await getGlobalState(controller.context, "taskHistory")) as any[]) || []
-
-		// Filter out non-favorited tasks
-		const favoritedTasks = taskHistory.filter((task) => task.isFavorited === true)
-		const deletedCount = taskHistory.length - favoritedTasks.length
-
-		console.log(`[deleteNonFavoritedTasks] Found ${favoritedTasks.length} favorited tasks to preserve`)
-
-		// Update global state
-		if (favoritedTasks.length > 0) {
-			await updateGlobalState(controller.context, "taskHistory", favoritedTasks)
-		} else {
-			await updateGlobalState(controller.context, "taskHistory", undefined)
-		}
-
-		// Handle file system cleanup for deleted tasks
-		const preserveTaskIds = favoritedTasks.map((task) => task.id)
-		await cleanupTaskFiles(controller, preserveTaskIds)
-
-		// Update webview
-		try {
-			await controller.postStateToWebview()
-		} catch (webviewErr) {
-			console.error("Error posting to webview:", webviewErr)
-		}
-
-		return DeleteNonFavoritedTasksResults.create({
-			tasksPreserved: favoritedTasks.length,
-			tasksDeleted: deletedCount,
-		})
-	} catch (error) {
-		console.error("Error in deleteNonFavoritedTasks:", error)
-		throw error
-	}
-}
-
-/**
- * Helper function to cleanup task files while preserving specified tasks
- */
-async function cleanupTaskFiles(controller: Controller, preserveTaskIds: string[]) {
-	const taskDirPath = path.join(controller.context.globalStorageUri.fsPath, "tasks")
-
-	try {
-		if (await fileExistsAtPath(taskDirPath)) {
-			if (preserveTaskIds.length > 0) {
-				const taskDirs = await fs.readdir(taskDirPath)
-				console.debug(`[cleanupTaskFiles] Found ${taskDirs.length} task directories`)
-
-				// Delete only non-preserved task directories
-				for (const dir of taskDirs) {
-					if (!preserveTaskIds.includes(dir)) {
-						await fs.rm(path.join(taskDirPath, dir), { recursive: true, force: true })
-					}
-				}
-			} else {
-				// No tasks to preserve, delete everything
-				await fs.rm(taskDirPath, { recursive: true, force: true })
-			}
-		}
-	} catch (error) {
-		console.error("Error cleaning up task files:", error)
-	}
-
-	return true
-}

src/core/controller/task/deleteTasksWithIds.ts

@@ -1,6 +1,9 @@
+import path from "path"
+import fs from "fs/promises"
 import { Controller } from ".."
-import { Empty, StringArrayRequest } from "../../../shared/proto/common"
+import { Empty, StringArrayRequest, BooleanRequest } from "../../../shared/proto/common"
 import { TaskMethodHandler } from "./index"
+import { fileExistsAtPath } from "../../../utils/fs"
 
 /**
  * Deletes tasks with the specified IDs
@@ -17,7 +20,72 @@ export const deleteTasksWithIds: TaskMethodHandler = async (
 		throw new Error("Missing task IDs")
 	}
 
-	await Promise.all(request.value.map((value) => controller.deleteTaskWithId(value)))
+	for (const id of request.value) {
+		await deleteTaskWithId(controller, id)
+	}
 
 	return Empty.create()
 }
+
+/**
+ * Deletes a single task with the specified ID
+ * @param controller The controller instance
+ * @param id The task ID to delete
+ */
+async function deleteTaskWithId(controller: Controller, id: string): Promise<void> {
+	console.info("deleteTaskWithId: ", id)
+
+	try {
+		// Clear current task if it matches the ID being deleted
+		if (id === controller.task?.taskId) {
+			await controller.clearTask()
+			console.debug("cleared task")
+		}
+
+		// Get task file paths
+		const { taskDirPath, apiConversationHistoryFilePath, uiMessagesFilePath, contextHistoryFilePath, taskMetadataFilePath } =
+			await controller.getTaskWithId(id)
+
+		// Remove task from state
+		const updatedTaskHistory = await controller.deleteTaskFromState(id)
+
+		// Delete the task files
+		for (const filePath of [
+			apiConversationHistoryFilePath,
+			uiMessagesFilePath,
+			contextHistoryFilePath,
+			taskMetadataFilePath,
+		]) {
+			const fileExists = await fileExistsAtPath(filePath)
+			if (fileExists) {
+				await fs.unlink(filePath)
+			}
+		}
+
+		// Remove empty task directory
+		try {
+			await fs.rmdir(taskDirPath) // succeeds if the dir is empty
+		} catch (error) {
+			console.debug("Could not remove task directory (may not be empty):", error)
+		}
+
+		// If no tasks remain, clean up everything
+		if (updatedTaskHistory.length === 0) {
+			const taskDirPath = path.join(controller.context.globalStorageUri.fsPath, "tasks")
+			const checkpointsDirPath = path.join(controller.context.globalStorageUri.fsPath, "checkpoints")
+
+			if (await fileExistsAtPath(taskDirPath)) {
+				await fs.rm(taskDirPath, { recursive: true, force: true })
+			}
+			if (await fileExistsAtPath(checkpointsDirPath)) {
+				await fs.rm(checkpointsDirPath, { recursive: true, force: true })
+			}
+		}
+	} catch (error) {
+		console.debug(`Error deleting task ${id}:`, error)
+		throw error // Re-throw to let caller handle the error
+	}
+
+	// Update webview state
+	await controller.postStateToWebview()
+}

src/core/controller/task/getTaskHistory.ts

@@ -15,7 +15,7 @@ export async function getTaskHistory(controller: Controller, request: GetTaskHis
 
 		// Get task history from global state
 		const taskHistory = ((await getGlobalState(controller.context, "taskHistory")) as any[]) || []
-		const workspacePath = getWorkspacePath()
+		const workspacePath = await getWorkspacePath()
 
 		// Apply filters
 		let filteredTasks = taskHistory.filter((item) => {

src/core/controller/ui/openUrl.ts

@@ -0,0 +1,20 @@
+import type { Controller } from "../index"
+import type { StringRequest } from "../../../shared/proto/common"
+import { Empty } from "../../../shared/proto/common"
+import { openUrlInBrowser } from "../../../utils/github-url-utils"
+
+/**
+ * Opens a URL in the default browser
+ * @param controller The controller instance
+ * @param request The URL to open
+ * @returns Empty response
+ */
+export async function openUrl(controller: Controller, request: StringRequest): Promise<Empty> {
+	try {
+		await openUrlInBrowser(request.value)
+		return Empty.create({})
+	} catch (error) {
+		console.error(`Failed to open URL: ${error}`)
+		throw error
+	}
+}

src/core/prompts/model_prompts/claude4.ts

@@ -268,7 +268,9 @@ Usage:
 </new_task>
 
 ## plan_mode_respond
-Description: Respond to the user's inquiry in an effort to plan a solution to the user's task. This tool should be used when you need to provide a response to a question or statement from the user about how you plan to accomplish the task. This tool is only available in PLAN MODE. The environment_details will specify the current mode, if it is not PLAN MODE then you should not use this tool. Depending on the user's message, you may ask questions to get clarification about the user's request, architect a solution to the task, and to brainstorm ideas with the user. For example, if the user's task is to create a website, you may start by asking some clarifying questions, then present a detailed plan for how you will accomplish the task given the context, and perhaps engage in a back and forth to finalize the details before the user switches you to ACT MODE to implement the solution. IMPORTANT NOTE: You should NOT ask for permission to read files or explore the repo. Just do that proactively. This tool should only be used when you've already gathered enough information to make a plan, or if you have a question for the user.
+Description: Respond to the user's inquiry in an effort to plan a solution to the user's task. This tool should ONLY be used when you have already explored the relevant files and are ready to present a concrete plan. DO NOT use this tool to announce what files you're going to read - just read them first. This tool is only available in PLAN MODE. The environment_details will specify the current mode, if it is not PLAN_MODE then you should not use this tool. For example, if the user's task is to create a website, you may start by asking some clarifying questions with the ask_followup_question tool if their message was vague, explore the codebase, read files, then present a detailed plan for how you will accomplish the task given the context, and perhaps engage in a back and forth to finalize the details before the user switches you to ACT_MODE to implement the solution. 
+CRITICAL: You must complete your information gathering (reading files, exploring the codebase) BEFORE using this tool. The user expects to see a well thought-out plan based on actual analysis, not intentions.
+
 Parameters:
 - response: (required) The response to provide to the user. Do not try to use tools in this parameter, this is simply a chat response. (You MUST use the response parameter, do not simply place the response text directly within <plan_mode_respond> tags.)
 Usage:
@@ -572,8 +574,8 @@ In each user message, the environment_details will specify the current mode. The
 ## What is PLAN MODE?
 
 - While you are usually in ACT MODE, the user may switch to PLAN MODE in order to have a back and forth with you to plan how to best accomplish the task. 
-- When starting in PLAN MODE, depending on the user's request, you may need to do some information gathering e.g. using read_file or search_files to get more context about the task. You may also ask the user clarifying questions to get a better understanding of the task. You may return mermaid diagrams to visually display your understanding.
-- Once you've gained more context about the user's request, you should architect a detailed plan for how you will accomplish the task. Returning mermaid diagrams may be helpful here as well.
+- When starting in PLAN MODE, depending on the user's request, you may need to do some information gathering e.g. using read_file or search_files to get more context about the task. You may also ask the user clarifying questions with ask_followup_question to get a better understanding of the task. You may return mermaid diagrams to visually display your understanding.
+- Once you've gained more context about the user's request, you should architect a detailed plan for how you will accomplish the task. Returning mermaid diagrams may be helpful here as well. Present the plan to the user using the plan_mode_respond tool.
 - Then you might ask the user if they are pleased with this plan, or if they would like to make any changes. Think of this as a brainstorming session where you can discuss the task and plan the best way to accomplish it.
 - If at any point a mermaid diagram would make your plan clearer to help the user quickly see the structure, you are encouraged to include a Mermaid code block in the response. (Note: if you use colors in your mermaid diagrams, be sure to use high contrast colors so the text is readable.)
 - Finally once it seems like you've reached a good plan, ask the user to switch you back to ACT MODE to implement the solution.
@@ -621,6 +623,7 @@ RULES
 - When you want to modify a file, use the replace_in_file or write_to_file tool directly with the desired changes. You do not need to display the changes before using the tool.
 - Do not ask for more information than necessary. Use the tools provided to accomplish the user's request efficiently and effectively. When you've completed your task, you must use the attempt_completion tool to present the result to the user. The user may provide feedback, which you can use to make improvements and try again.
 - You are only allowed to ask the user questions using the ask_followup_question tool. Use this tool only when you need additional details to complete a task, and be sure to use a clear and concise question that will help you move forward with the task. However if you can use the available tools to avoid having to ask the user questions, you should do so. For example, if the user mentions a file that may be in an outside directory like the Desktop, you should use the list_files tool to list the files in the Desktop and check if the file they are talking about is there, rather than asking the user to provide the file path themselves.
+- When the user is being vague, you should be proactive about asking clarifying questions using the ask_followup_question tool to ensure you understand their request. However, if you can infer the user's intent based on the context and available tools, you should proceed without asking unnecessary questions
 - When executing commands, if you don't see the expected output, assume the terminal executed the command successfully and proceed with the task. The user's terminal may be unable to stream the output back properly. If you absolutely need to see the actual terminal output, use the ask_followup_question tool to request the user to copy and paste it back to you.
 - The user may provide a file's contents directly in their message, in which case you shouldn't use the read_file tool to get the file contents again since you already have it.
 - Your goal is to try to accomplish the user's task, NOT engage in a back and forth conversation.${

src/core/prompts/responses.ts

@@ -35,11 +35,6 @@ Otherwise, if you have not completed the task and do not need additional informa
 	tooManyMistakes: (feedback?: string) =>
 		`You seem to be having trouble proceeding. The user has provided the following feedback to help guide you:\n<feedback>\n${feedback}\n</feedback>`,
 
-	autoApprovalFeedback: (feedback?: string) =>
-		`The user has provided the following feedback to help guide you:\n<feedback>\n${feedback}\n</feedback>`,
-
-	autoApprovalRejection: () => `The user has rejected continuing with auto-approved requests and chose to stop the task.`,
-
 	missingToolParameterError: (paramName: string) =>
 		`Missing value for required parameter '${paramName}'. Please retry with complete response.\n\n${toolUseInstructionsReminder}`,
 

src/core/prompts/system.ts

@@ -12,14 +12,14 @@ export const SYSTEM_PROMPT = async (
 	supportsBrowserUse: boolean,
 	mcpHub: McpHub,
 	browserSettings: BrowserSettings,
-	isClaude4ModelFamily: boolean = false,
+	isNextGenModel: boolean = false,
 ) => {
 
-	if (isClaude4ModelFamily && USE_EXPERIMENTAL_CLAUDE4_FEATURES) {
+	if (isNextGenModel && USE_EXPERIMENTAL_CLAUDE4_FEATURES) {
 		return SYSTEM_PROMPT_CLAUDE4_EXPERIMENTAL(cwd, supportsBrowserUse, mcpHub, browserSettings)
 	}
 
-  if (isClaude4ModelFamily) {
+  if (isNextGenModel) {
     return SYSTEM_PROMPT_CLAUDE4(cwd, supportsBrowserUse, mcpHub, browserSettings)
   }
 

src/core/storage/state-migrations.ts

@@ -0,0 +1,147 @@
+import * as vscode from "vscode"
+import { ensureRulesDirectoryExists } from "./disk"
+import fs from "fs/promises"
+import path from "path"
+import { getGlobalState, getWorkspaceState, updateGlobalState, updateWorkspaceState } from "./state"
+import { GlobalStateKey } from "./state-keys"
+
+export async function migratePlanActGlobalToWorkspaceStorage(context: vscode.ExtensionContext) {
+	// Keys that were migrated from global storage to workspace storage
+	const keysToMigrate = [
+		// Core settings
+		"apiProvider",
+		"apiModelId",
+		"thinkingBudgetTokens",
+		"reasoningEffort",
+		"chatSettings",
+		"vsCodeLmModelSelector",
+
+		// Provider-specific model keys
+		"awsBedrockCustomSelected",
+		"awsBedrockCustomModelBaseId",
+		"openRouterModelId",
+		"openRouterModelInfo",
+		"openAiModelId",
+		"openAiModelInfo",
+		"ollamaModelId",
+		"lmStudioModelId",
+		"liteLlmModelId",
+		"liteLlmModelInfo",
+		"requestyModelId",
+		"requestyModelInfo",
+		"togetherModelId",
+		"fireworksModelId",
+
+		// Previous mode settings
+		"previousModeApiProvider",
+		"previousModeModelId",
+		"previousModeModelInfo",
+		"previousModeVsCodeLmModelSelector",
+		"previousModeThinkingBudgetTokens",
+		"previousModeReasoningEffort",
+		"previousModeAwsBedrockCustomSelected",
+		"previousModeAwsBedrockCustomModelBaseId",
+	]
+
+	for (const key of keysToMigrate) {
+		const globalValue = await getGlobalState(context, key as GlobalStateKey)
+		if (globalValue !== undefined) {
+			const workspaceValue = await getWorkspaceState(context, key)
+			if (workspaceValue === undefined) {
+				await updateWorkspaceState(context, key, globalValue)
+			}
+			// Delete from global storage regardless of whether we copied it
+			await updateGlobalState(context, key as GlobalStateKey, undefined)
+		}
+	}
+}
+
+export async function migrateMcpMarketplaceEnableSetting(mcpMarketplaceEnabledRaw: boolean | undefined): Promise<boolean> {
+	const config = vscode.workspace.getConfiguration("cline")
+	const mcpMarketplaceEnabled = config.get<boolean>("mcpMarketplace.enabled")
+	if (mcpMarketplaceEnabled !== undefined) {
+		// Remove from VSCode configuration
+		await config.update("mcpMarketplace.enabled", undefined, true)
+
+		return !mcpMarketplaceEnabled
+	}
+	return mcpMarketplaceEnabledRaw ?? true
+}
+
+export async function migrateEnableCheckpointsSetting(enableCheckpointsSettingRaw: boolean | undefined): Promise<boolean> {
+	const config = vscode.workspace.getConfiguration("cline")
+	const enableCheckpoints = config.get<boolean>("enableCheckpoints")
+	if (enableCheckpoints !== undefined) {
+		// Remove from VSCode configuration
+		await config.update("enableCheckpoints", undefined, true)
+		return enableCheckpoints
+	}
+	return enableCheckpointsSettingRaw ?? true
+}
+
+export async function migrateCustomInstructionsToGlobalRules(context: vscode.ExtensionContext) {
+	try {
+		const customInstructions = (await context.globalState.get("customInstructions")) as string | undefined
+
+		if (customInstructions?.trim()) {
+			console.log("Migrating custom instructions to global Cline rules...")
+
+			// Create global .clinerules directory if it doesn't exist
+			const globalRulesDir = await ensureRulesDirectoryExists()
+
+			// Use a fixed filename for custom instructions
+			const migrationFileName = "custom_instructions.md"
+			const migrationFilePath = path.join(globalRulesDir, migrationFileName)
+
+			try {
+				// Check if file already exists to determine if we should append
+				let existingContent = ""
+				try {
+					existingContent = await fs.readFile(migrationFilePath, "utf8")
+				} catch (readError) {
+					// File doesn't exist, which is fine
+				}
+
+				// Append or create the file with custom instructions
+				const contentToWrite = existingContent
+					? `${existingContent}\n\n---\n\n${customInstructions.trim()}`
+					: customInstructions.trim()
+
+				await fs.writeFile(migrationFilePath, contentToWrite)
+				console.log(`Successfully ${existingContent ? "appended to" : "created"} migration file: ${migrationFilePath}`)
+			} catch (fileError) {
+				console.error("Failed to write migration file:", fileError)
+				return
+			}
+
+			// Remove customInstructions from global state only after successful file creation
+			await context.globalState.update("customInstructions", undefined)
+			console.log("Successfully migrated custom instructions to global Cline rules")
+		}
+	} catch (error) {
+		console.error("Failed to migrate custom instructions to global rules:", error)
+		// Continue execution - migration failure shouldn't break extension startup
+	}
+}
+
+export async function migrateModeFromWorkspaceStorageToControllerState(context: vscode.ExtensionContext) {
+	try {
+		// Get current chatSettings from workspace storage
+		const chatSettings = (await getWorkspaceState(context, "chatSettings")) as any
+
+		if (chatSettings && typeof chatSettings === "object" && "mode" in chatSettings) {
+			console.log("Cleaning up mode from workspace storage...")
+
+			// Remove mode property from chatSettings
+			const { mode, ...cleanedChatSettings } = chatSettings
+
+			// Save cleaned chatSettings back to workspace storage
+			await updateWorkspaceState(context, "chatSettings", cleanedChatSettings)
+
+			console.log("Successfully removed mode from workspace storage chatSettings")
+		}
+	} catch (error) {
+		console.error("Failed to cleanup mode from workspace storage:", error)
+		// Continue execution - migration failure shouldn't break extension startup
+	}
+}

src/core/storage/state.ts

@@ -1,5 +1,5 @@
 import * as vscode from "vscode"
-import { DEFAULT_CHAT_SETTINGS, OpenAIReasoningEffort } from "@shared/ChatSettings"
+import { DEFAULT_CHAT_SETTINGS } from "@shared/ChatSettings"
 import { DEFAULT_BROWSER_SETTINGS } from "@shared/BrowserSettings"
 import { DEFAULT_AUTO_APPROVAL_SETTINGS } from "@shared/AutoApprovalSettings"
 import { GlobalStateKey, SecretKey } from "./state-keys"
@@ -7,13 +7,11 @@ import { ApiConfiguration, ApiProvider, BedrockModelId, ModelInfo } from "@share
 import { HistoryItem } from "@shared/HistoryItem"
 import { AutoApprovalSettings } from "@shared/AutoApprovalSettings"
 import { BrowserSettings } from "@shared/BrowserSettings"
-import { ChatSettings } from "@shared/ChatSettings"
+import { StoredChatSettings } from "@shared/ChatSettings"
 import { TelemetrySetting } from "@shared/TelemetrySetting"
 import { UserInfo } from "@shared/UserInfo"
 import { ClineRulesToggles } from "@shared/cline-rules"
-import { ensureRulesDirectoryExists } from "./disk"
-import fs from "fs/promises"
-import path from "path"
+import { migrateEnableCheckpointsSetting, migrateMcpMarketplaceEnableSetting } from "./state-migrations"
 /*
 	Storage
 	https://dev.to/kompotkot/how-to-use-secretstorage-in-your-vscode-extensions-2hco
@@ -54,125 +52,6 @@ export async function getWorkspaceState(context: vscode.ExtensionContext, key: s
 	return await context.workspaceState.get(key)
 }
 
-export async function migratePlanActGlobalToWorkspaceStorage(context: vscode.ExtensionContext) {
-	// Keys that were migrated from global storage to workspace storage
-	const keysToMigrate = [
-		// Core settings
-		"apiProvider",
-		"apiModelId",
-		"thinkingBudgetTokens",
-		"reasoningEffort",
-		"chatSettings",
-		"vsCodeLmModelSelector",
-
-		// Provider-specific model keys
-		"awsBedrockCustomSelected",
-		"awsBedrockCustomModelBaseId",
-		"openRouterModelId",
-		"openRouterModelInfo",
-		"openAiModelId",
-		"openAiModelInfo",
-		"ollamaModelId",
-		"lmStudioModelId",
-		"liteLlmModelId",
-		"liteLlmModelInfo",
-		"requestyModelId",
-		"requestyModelInfo",
-		"togetherModelId",
-		"fireworksModelId",
-
-		// Previous mode settings
-		"previousModeApiProvider",
-		"previousModeModelId",
-		"previousModeModelInfo",
-		"previousModeVsCodeLmModelSelector",
-		"previousModeThinkingBudgetTokens",
-		"previousModeReasoningEffort",
-		"previousModeAwsBedrockCustomSelected",
-		"previousModeAwsBedrockCustomModelBaseId",
-	]
-
-	for (const key of keysToMigrate) {
-		const globalValue = await getGlobalState(context, key as GlobalStateKey)
-		if (globalValue !== undefined) {
-			const workspaceValue = await getWorkspaceState(context, key)
-			if (workspaceValue === undefined) {
-				await updateWorkspaceState(context, key, globalValue)
-			}
-			// Delete from global storage regardless of whether we copied it
-			await updateGlobalState(context, key as GlobalStateKey, undefined)
-		}
-	}
-}
-
-async function migrateMcpMarketplaceEnableSetting(mcpMarketplaceEnabledRaw: boolean | undefined): Promise<boolean> {
-	const config = vscode.workspace.getConfiguration("cline")
-	const mcpMarketplaceEnabled = config.get<boolean>("mcpMarketplace.enabled")
-	if (mcpMarketplaceEnabled !== undefined) {
-		// Remove from VSCode configuration
-		await config.update("mcpMarketplace.enabled", undefined, true)
-
-		return !mcpMarketplaceEnabled
-	}
-	return mcpMarketplaceEnabledRaw ?? true
-}
-
-async function migrateEnableCheckpointsSetting(enableCheckpointsSettingRaw: boolean | undefined): Promise<boolean> {
-	const config = vscode.workspace.getConfiguration("cline")
-	const enableCheckpoints = config.get<boolean>("enableCheckpoints")
-	if (enableCheckpoints !== undefined) {
-		// Remove from VSCode configuration
-		await config.update("enableCheckpoints", undefined, true)
-		return enableCheckpoints
-	}
-	return enableCheckpointsSettingRaw ?? true
-}
-
-export async function migrateCustomInstructionsToGlobalRules(context: vscode.ExtensionContext) {
-	try {
-		const customInstructions = (await context.globalState.get("customInstructions")) as string | undefined
-
-		if (customInstructions?.trim()) {
-			console.log("Migrating custom instructions to global Cline rules...")
-
-			// Create global .clinerules directory if it doesn't exist
-			const globalRulesDir = await ensureRulesDirectoryExists()
-
-			// Use a fixed filename for custom instructions
-			const migrationFileName = "custom_instructions.md"
-			const migrationFilePath = path.join(globalRulesDir, migrationFileName)
-
-			try {
-				// Check if file already exists to determine if we should append
-				let existingContent = ""
-				try {
-					existingContent = await fs.readFile(migrationFilePath, "utf8")
-				} catch (readError) {
-					// File doesn't exist, which is fine
-				}
-
-				// Append or create the file with custom instructions
-				const contentToWrite = existingContent
-					? `${existingContent}\n\n---\n\n${customInstructions.trim()}`
-					: customInstructions.trim()
-
-				await fs.writeFile(migrationFilePath, contentToWrite)
-				console.log(`Successfully ${existingContent ? "appended to" : "created"} migration file: ${migrationFilePath}`)
-			} catch (fileError) {
-				console.error("Failed to write migration file:", fileError)
-				return
-			}
-
-			// Remove customInstructions from global state only after successful file creation
-			await context.globalState.update("customInstructions", undefined)
-			console.log("Successfully migrated custom instructions to global Cline rules")
-		}
-	} catch (error) {
-		console.error("Failed to migrate custom instructions to global rules:", error)
-		// Continue execution - migration failure shouldn't break extension startup
-	}
-}
-
 export async function getAllExtensionState(context: vscode.ExtensionContext) {
 	const [
 		isNewUser,
@@ -360,7 +239,7 @@ export async function getAllExtensionState(context: vscode.ExtensionContext) {
 		previousModeSapAiCoreResourceGroup,
 		previousModeSapAiCoreModelId,
 	] = await Promise.all([
-		getWorkspaceState(context, "chatSettings") as Promise<ChatSettings | undefined>,
+		getWorkspaceState(context, "chatSettings") as Promise<StoredChatSettings | undefined>,
 		getWorkspaceState(context, "apiProvider") as Promise<ApiProvider | undefined>,
 		getWorkspaceState(context, "apiModelId") as Promise<string | undefined>,
 		getWorkspaceState(context, "thinkingBudgetTokens") as Promise<number | undefined>,

src/core/task/ToolExecutor.ts

@@ -11,6 +11,7 @@ import { ClineIgnoreController } from "@core/ignore/ClineIgnoreController"
 import { AutoApprovalSettings } from "@shared/AutoApprovalSettings"
 import { ChatSettings } from "@shared/ChatSettings"
 import { ToolParamName, ToolUse, ToolUseName } from "../assistant-message"
+import { AutoApprove } from "./tools/autoApprove"
 import {
 	BrowserAction,
 	BrowserActionResult,
@@ -30,7 +31,7 @@ import { ClineAskResponse } from "@shared/WebviewMessage"
 import { fixModelHtmlEscaping, removeInvalidChars } from "@utils/string"
 import { fileExistsAtPath } from "@utils/fs"
 import { formatResponse } from "../prompts/responses"
-import { isClaude4ModelFamily } from "@utils/model-utils"
+import { isClaude4ModelFamily, isGemini2dot5ModelFamily } from "@utils/model-utils"
 import { ToolResponse, USE_EXPERIMENTAL_CLAUDE4_FEATURES } from "."
 import { serializeError } from "serialize-error"
 import * as path from "path"
@@ -57,6 +58,17 @@ import { setTimeout as setTimeoutPromise } from "node:timers/promises"
 import { ChangeLocation, StreamingJsonReplacer } from "../assistant-message/diff-json"
 
 export class ToolExecutor {
+	private autoApprover: AutoApprove
+
+	// Auto-approval methods using the AutoApprove class
+	private shouldAutoApproveTool(toolName: ToolUseName): boolean | [boolean, boolean] {
+		return this.autoApprover.shouldAutoApproveTool(toolName)
+	}
+
+	private shouldAutoApproveToolWithPath(blockname: ToolUseName, autoApproveActionpath: string | undefined): boolean {
+		return this.autoApprover.shouldAutoApproveToolWithPath(blockname, autoApproveActionpath)
+	}
+
 	constructor(
 		// Core Services & Managers
 		private context: vscode.ExtensionContext,
@@ -96,21 +108,28 @@ export class ToolExecutor {
 		private saveCheckpoint: (isAttemptCompletionMessage?: boolean) => Promise<void>,
 		private reinitExistingTaskFromId: (taskId: string) => Promise<void>,
 		private cancelTask: () => Promise<void>,
-		private shouldAutoApproveTool: (toolName: ToolUseName) => boolean | [boolean, boolean],
-		private shouldAutoApproveToolWithPath: (blockname: ToolUseName, autoApproveActionpath: string | undefined) => boolean,
 		private sayAndCreateMissingParamError: (toolName: ToolUseName, paramName: string, relPath?: string) => Promise<any>,
 		private removeLastPartialMessageIfExistsWithType: (type: "ask" | "say", askOrSay: ClineAsk | ClineSay) => Promise<void>,
 		private executeCommandTool: (command: string) => Promise<[boolean, any]>,
 		private doesLatestTaskCompletionHaveNewChanges: () => Promise<boolean>,
-	) {}
+	) {
+		this.autoApprover = new AutoApprove(autoApprovalSettings)
+	}
+
+	/**
+	 * Updates the auto approval settings
+	 */
+	public updateAutoApprovalSettings(settings: AutoApprovalSettings): void {
+		this.autoApprover.updateSettings(settings)
+	}
 
 	private pushToolResult = (content: ToolResponse, block: ToolUse) => {
-		const isClaude4Model = isClaude4ModelFamily(this.api)
+		const isNextGenModel = isClaude4ModelFamily(this.api) || isGemini2dot5ModelFamily(this.api)
 
 		if (typeof content === "string") {
 			const resultText = content || "(tool did not return anything)"
 
-			if (isClaude4Model && USE_EXPERIMENTAL_CLAUDE4_FEATURES) {
+			if (isNextGenModel && USE_EXPERIMENTAL_CLAUDE4_FEATURES) {
 				// Claude 4 family: Use function_results format
 				this.taskState.userMessageContent.push({
 					type: "text",
@@ -472,9 +491,9 @@ export class ToolExecutor {
 
 						const currentFullJson = block.params.diff
 						// Check if we should use streaming (e.g., for specific models)
-						const isClaude4Model = isClaude4ModelFamily(this.api)
+						const isNextGenModel = isClaude4ModelFamily(this.api) || isGemini2dot5ModelFamily(this.api)
 						// Going through claude family of models
-						if (isClaude4Model && USE_EXPERIMENTAL_CLAUDE4_FEATURES && currentFullJson) {
+						if (isNextGenModel && USE_EXPERIMENTAL_CLAUDE4_FEATURES && currentFullJson) {
 							const streamingResult = await this.handleStreamingJsonReplacement(block, relPath, currentFullJson)
 
 							if (streamingResult.error) {
@@ -555,7 +574,7 @@ export class ToolExecutor {
 						tool: fileExists ? "editedExistingFile" : "newFileCreated",
 						path: getReadablePath(this.cwd, this.removeClosingTag(block, "path", relPath)),
 						content: diff || content,
-						operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
+						operationIsLocatedInWorkspace: await isLocatedInWorkspace(relPath),
 					}
 
 					if (block.partial) {
@@ -626,7 +645,7 @@ export class ToolExecutor {
 						const completeMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: diff || content,
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(relPath),
 							// ? formatResponse.createPrettyPatch(
 							// 		relPath,
 							// 		this.diffViewProvider.originalContent,
@@ -769,7 +788,7 @@ export class ToolExecutor {
 						const partialMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: undefined,
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(relPath),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -800,7 +819,7 @@ export class ToolExecutor {
 						const completeMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: absolutePath,
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(relPath),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(relPath),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -839,7 +858,6 @@ export class ToolExecutor {
 				}
 			}
 			case "list_files": {
-				const isClaude4Model = isClaude4ModelFamily(this.api)
 				const relDirPath: string | undefined = block.params.path
 				const recursiveRaw: string | undefined = block.params.recursive
 				const recursive = recursiveRaw?.toLowerCase() === "true"
@@ -852,7 +870,7 @@ export class ToolExecutor {
 						const partialMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: "",
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(block.params.path),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -884,7 +902,7 @@ export class ToolExecutor {
 						const completeMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: result,
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(block.params.path),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -927,7 +945,7 @@ export class ToolExecutor {
 						const partialMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: "",
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(block.params.path),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -956,7 +974,7 @@ export class ToolExecutor {
 						const completeMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: result,
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(block.params.path),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -989,7 +1007,6 @@ export class ToolExecutor {
 				}
 			}
 			case "search_files": {
-				const isClaude4Model = isClaude4ModelFamily(this.api)
 				const relDirPath: string | undefined = block.params.path
 				const regex: string | undefined = block.params.regex
 				const filePattern: string | undefined = block.params.file_pattern
@@ -1004,7 +1021,7 @@ export class ToolExecutor {
 						const partialMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: "",
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(block.params.path),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")
@@ -1041,7 +1058,7 @@ export class ToolExecutor {
 						const completeMessage = JSON.stringify({
 							...sharedMessageProps,
 							content: results,
-							operationIsLocatedInWorkspace: isLocatedInWorkspace(block.params.path),
+							operationIsLocatedInWorkspace: await isLocatedInWorkspace(block.params.path),
 						} satisfies ClineSayTool)
 						if (this.shouldAutoApproveToolWithPath(block.name, block.params.path)) {
 							this.removeLastPartialMessageIfExistsWithType("ask", "tool")

src/core/task/index.ts

@@ -90,7 +90,7 @@ import WorkspaceTracker from "@integrations/workspace/WorkspaceTracker"
 import { McpHub } from "@services/mcp/McpHub"
 import { isInTestMode } from "../../services/test/TestMode"
 import { processFilesIntoText } from "@integrations/misc/extract-text"
-import { isClaude4ModelFamily } from "@utils/model-utils"
+import { isClaude4ModelFamily, isGemini2dot5ModelFamily } from "@utils/model-utils"
 import { MessageStateHandler } from "./message-state"
 import { formatErrorWithStatusCode, updateApiReqMsg } from "./utils"
 import { TaskState } from "./TaskState"
@@ -314,8 +314,6 @@ export class Task {
 			this.saveCheckpoint.bind(this),
 			this.reinitExistingTaskFromId.bind(this),
 			this.cancelTask.bind(this),
-			this.shouldAutoApproveTool.bind(this),
-			this.shouldAutoApproveToolWithPath.bind(this),
 			this.sayAndCreateMissingParamError.bind(this),
 			this.removeLastPartialMessageIfExistsWithType.bind(this),
 			this.executeCommandTool.bind(this),
@@ -333,6 +331,13 @@ export class Task {
 		return context
 	}
 
+	/**
+	 * Updates the auto approval settings for this task
+	 */
+	public updateAutoApprovalSettings(settings: AutoApprovalSettings): void {
+		this.toolExecutor.updateAutoApprovalSettings(settings)
+	}
+
 	async restoreCheckpoint(messageTs: number, restoreType: ClineCheckpointRestore, offset?: number) {
 		const clineMessages = this.messageStateHandler.getClineMessages()
 		const messageIndex = clineMessages.findIndex((m) => m.ts === messageTs) - (offset || 0)
@@ -1554,69 +1559,6 @@ export class Task {
 		}
 	}
 
-	// Check if the tool should be auto-approved based on the settings
-	// Returns bool for most tools, and tuple for tools with nested settings
-	shouldAutoApproveTool(toolName: ToolUseName): boolean | [boolean, boolean] {
-		if (this.autoApprovalSettings.enabled) {
-			switch (toolName) {
-				case "read_file":
-				case "list_files":
-				case "list_code_definition_names":
-				case "search_files":
-					return [
-						this.autoApprovalSettings.actions.readFiles,
-						this.autoApprovalSettings.actions.readFilesExternally ?? false,
-					]
-				case "new_rule":
-				case "write_to_file":
-				case "replace_in_file":
-					return [
-						this.autoApprovalSettings.actions.editFiles,
-						this.autoApprovalSettings.actions.editFilesExternally ?? false,
-					]
-				case "execute_command":
-					return [
-						this.autoApprovalSettings.actions.executeSafeCommands ?? false,
-						this.autoApprovalSettings.actions.executeAllCommands ?? false,
-					]
-				case "browser_action":
-					return this.autoApprovalSettings.actions.useBrowser
-				case "web_fetch":
-					return this.autoApprovalSettings.actions.useBrowser
-				case "access_mcp_resource":
-				case "use_mcp_tool":
-					return this.autoApprovalSettings.actions.useMcp
-			}
-		}
-		return false
-	}
-
-	// Check if the tool should be auto-approved based on the settings
-	// and the path of the action. Returns true if the tool should be auto-approved
-	// based on the user's settings and the path of the action.
-	shouldAutoApproveToolWithPath(blockname: ToolUseName, autoApproveActionpath: string | undefined): boolean {
-		let isLocalRead: boolean = false
-		if (autoApproveActionpath) {
-			const absolutePath = path.resolve(cwd, autoApproveActionpath)
-			isLocalRead = absolutePath.startsWith(cwd)
-		} else {
-			// If we do not get a path for some reason, default to a (safer) false return
-			isLocalRead = false
-		}
-
-		// Get auto-approve settings for local and external edits
-		const autoApproveResult = this.shouldAutoApproveTool(blockname)
-		const [autoApproveLocal, autoApproveExternal] = Array.isArray(autoApproveResult)
-			? autoApproveResult
-			: [autoApproveResult, false]
-
-		if ((isLocalRead && autoApproveLocal) || (!isLocalRead && autoApproveLocal && autoApproveExternal)) {
-			return true
-		} else {
-			return false
-		}
-	}
-
 	/**
 	 * Migrates the disableBrowserTool setting from VSCode configuration to browserSettings
 	 */
@@ -1654,8 +1596,8 @@ export class Task {
 
 		const supportsBrowserUse = modelSupportsBrowserUse && !disableBrowserTool // only enable browser use if the model supports it and the user hasn't disabled it
 
-		const isClaude4Model = isClaude4ModelFamily(this.api)
-		let systemPrompt = await SYSTEM_PROMPT(cwd, supportsBrowserUse, this.mcpHub, this.browserSettings, isClaude4Model)
+		const isNextGenModel = isClaude4ModelFamily(this.api) || isGemini2dot5ModelFamily(this.api)
+		let systemPrompt = await SYSTEM_PROMPT(cwd, supportsBrowserUse, this.mcpHub, this.browserSettings, isNextGenModel)
 
 		await this.migratePreferredLanguageToolSetting()
 		const preferredLanguage = getLanguageKey(this.chatSettings.preferredLanguage as LanguageDisplay)
@@ -1965,7 +1907,7 @@ export class Task {
 				"mistake_limit_reached",
 				this.api.getModel().id.includes("claude")
 					? `This may indicate a failure in his thought process or inability to use a tool properly, which can be mitigated with some user guidance (e.g. "Try breaking down the task into smaller steps").`
-					: "Cline uses complex prompts and iterative task execution that may be challenging for less capable models. For best results, it's recommended to use Claude 3.7 Sonnet for its advanced agentic coding capabilities.",
+					: "Cline uses complex prompts and iterative task execution that may be challenging for less capable models. For best results, it's recommended to use Claude 4 Sonnet for its advanced agentic coding capabilities.",
 			)
 			if (response === "messageResponse") {
 				// This userContent is for the *next* API call.
@@ -2005,80 +1947,12 @@ export class Task {
 					message: `Cline has auto-approved ${this.autoApprovalSettings.maxRequests.toString()} API requests.`,
 				})
 			}
-			const { response, text, images, files } = await this.ask(
+			await this.ask(
 				"auto_approval_max_req_reached",
 				`Cline has auto-approved ${this.autoApprovalSettings.maxRequests.toString()} API requests. Would you like to reset the count and proceed with the task?`,
 			)
-
-			if (response !== "yesButtonClicked") {
-				// This userContent is for the *next* API call.
-
-				const feedbackUserContent: UserContent = []
-				feedbackUserContent.push({
-					type: "text",
-					text: formatResponse.autoApprovalRejection(),
-				})
-
-				if (text) {
-					await this.say("user_feedback", text, images, files) // Save feedback to history
-					feedbackUserContent.push({
-						type: "text",
-						text: formatResponse.autoApprovalFeedback(text),
-					})
-				}
-
-				if (images && images.length > 0) {
-					feedbackUserContent.push(...formatResponse.imageBlocks(images))
-				}
-
-				let fileContentString = ""
-				if (files && files.length > 0) {
-					fileContentString = await processFilesIntoText(files)
-				}
-
-				if (fileContentString) {
-					feedbackUserContent.push({
-						type: "text",
-						text: fileContentString,
-					})
-				}
-
-				userContent = feedbackUserContent
-			} else {
-				// User hit the approve button, and may have provided feedback
-				const feedbackUserContent: UserContent = []
-
-				if (text) {
-					await this.say("user_feedback", text, images, files)
-					feedbackUserContent.push({
-						type: "text",
-						text: formatResponse.autoApprovalFeedback(text),
-					})
-				}
-
-				if (images && images.length > 0) {
-					feedbackUserContent.push(...formatResponse.imageBlocks(images))
-				}
-
-				let fileContentString = ""
-				if (files && files.length > 0) {
-					fileContentString = await processFilesIntoText(files)
-				}
-
-				if (fileContentString) {
-					feedbackUserContent.push({
-						type: "text",
-						text: fileContentString,
-					})
-				}
-
-				userContent = feedbackUserContent
-			}
-
+			// if we get past the promise it means the user approved and did not start a new task
 			this.taskState.consecutiveAutoApprovedRequestsCount = 0
-			if (userContent.length > 0) {
-				await this.saveCheckpoint()
-			}
 		}
 
 		// get previous api req's index to check token usage and determine if we need to truncate conversation history
@@ -2291,8 +2165,8 @@ export class Task {
 							assistantMessage += chunk.text
 							// parse raw assistant message into content blocks
 							const prevLength = this.taskState.assistantMessageContent.length
-							const isClaude4Model = isClaude4ModelFamily(this.api)
-							if (isClaude4Model && USE_EXPERIMENTAL_CLAUDE4_FEATURES) {
+							const isNextGenModel = isClaude4ModelFamily(this.api) || isGemini2dot5ModelFamily(this.api)
+							if (isNextGenModel && USE_EXPERIMENTAL_CLAUDE4_FEATURES) {
 								this.taskState.assistantMessageContent = parseAssistantMessageV3(assistantMessage)
 							} else {
 								this.taskState.assistantMessageContent = parseAssistantMessageV2(assistantMessage)

src/core/task/tools/autoApprove.ts

@@ -0,0 +1,83 @@
+import { AutoApprovalSettings } from "@shared/AutoApprovalSettings"
+import { ToolUseName } from "@core/assistant-message"
+import * as path from "path"
+import * as vscode from "vscode"
+import os from "os"
+
+export const cwd =
+	vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) ?? path.join(os.homedir(), "Desktop")
+
+export class AutoApprove {
+	autoApprovalSettings: AutoApprovalSettings
+
+	constructor(autoApprovalSettings: AutoApprovalSettings) {
+		this.autoApprovalSettings = autoApprovalSettings
+	}
+
+	// Check if the tool should be auto-approved based on the settings
+	// Returns bool for most tools, and tuple for tools with nested settings
+	shouldAutoApproveTool(toolName: ToolUseName): boolean | [boolean, boolean] {
+		if (this.autoApprovalSettings.enabled) {
+			switch (toolName) {
+				case "read_file":
+				case "list_files":
+				case "list_code_definition_names":
+				case "search_files":
+					return [
+						this.autoApprovalSettings.actions.readFiles,
+						this.autoApprovalSettings.actions.readFilesExternally ?? false,
+					]
+				case "new_rule":
+				case "write_to_file":
+				case "replace_in_file":
+					return [
+						this.autoApprovalSettings.actions.editFiles,
+						this.autoApprovalSettings.actions.editFilesExternally ?? false,
+					]
+				case "execute_command":
+					return [
+						this.autoApprovalSettings.actions.executeSafeCommands ?? false,
+						this.autoApprovalSettings.actions.executeAllCommands ?? false,
+					]
+				case "browser_action":
+					return this.autoApprovalSettings.actions.useBrowser
+				case "web_fetch":
+					return this.autoApprovalSettings.actions.useBrowser
+				case "access_mcp_resource":
+				case "use_mcp_tool":
+					return this.autoApprovalSettings.actions.useMcp
+			}
+		}
+		return false
+	}
+
+	// Check if the tool should be auto-approved based on the settings
+	// and the path of the action. Returns true if the tool should be auto-approved
+	// based on the user's settings and the path of the action.
+	shouldAutoApproveToolWithPath(blockname: ToolUseName, autoApproveActionpath: string | undefined): boolean {
+		let isLocalRead: boolean = false
+		if (autoApproveActionpath) {
+			const absolutePath = path.resolve(cwd, autoApproveActionpath)
+			isLocalRead = absolutePath.startsWith(cwd)
+		} else {
+			// If we do not get a path for some reason, default to a (safer) false return
+			isLocalRead = false
+		}
+
+		// Get auto-approve settings for local and external edits
+		const autoApproveResult = this.shouldAutoApproveTool(blockname)
+		const [autoApproveLocal, autoApproveExternal] = Array.isArray(autoApproveResult)
+			? autoApproveResult
+			: [autoApproveResult, false]
+
+		if ((isLocalRead && autoApproveLocal) || (!isLocalRead && autoApproveLocal && autoApproveExternal)) {
+			return true
+		} else {
+			return false
+		}
+	}
+
+	updateSettings(settings: AutoApprovalSettings): void {
+		this.autoApprovalSettings = settings
+	}
+}

src/core/webview/index.ts

@@ -200,7 +200,7 @@ export abstract class WebviewProvider {
 				<link href="${codiconsUri}" rel="stylesheet" />
 				<link href="${katexCssUri}" rel="stylesheet" />
 				<meta http-equiv="Content-Security-Policy" content="default-src 'none';
-					connect-src https://*.posthog.com https://*.firebaseauth.com https://*.firebaseio.com https://*.googleapis.com https://*.firebase.com; 
+					connect-src https://*.posthog.com https://*.cline.bot https://*.firebaseauth.com https://*.firebaseio.com https://*.googleapis.com https://*.firebase.com; 
 					font-src ${this.getCspSource()} data:; 
 					style-src ${this.getCspSource()} 'unsafe-inline'; 
 					img-src ${this.getCspSource()} https: data:; 

src/extension.ts

@@ -22,7 +22,11 @@ import { WebviewProviderType as WebviewProviderTypeEnum } from "@shared/proto/ui
 import { WebviewProviderType } from "./shared/webview/types"
 import { sendHistoryButtonClickedEvent } from "./core/controller/ui/subscribeToHistoryButtonClicked"
 import { sendAccountButtonClickedEvent } from "./core/controller/ui/subscribeToAccountButtonClicked"
-import { migratePlanActGlobalToWorkspaceStorage, migrateCustomInstructionsToGlobalRules } from "./core/storage/state"
+import {
+	migratePlanActGlobalToWorkspaceStorage,
+	migrateCustomInstructionsToGlobalRules,
+	migrateModeFromWorkspaceStorageToControllerState,
+} from "./core/storage/state-migrations"
 
 import { sendFocusChatInputEvent } from "./core/controller/ui/subscribeToFocusChatInput"
 import { FileContextTracker } from "./core/context/context-tracking/FileContextTracker"
@@ -30,6 +34,7 @@ import * as hostProviders from "@hosts/host-providers"
 import { vscodeHostBridgeClient } from "@/hosts/vscode/client/host-grpc-client"
 import { VscodeWebviewProvider } from "./core/webview/VscodeWebviewProvider"
 import { ExtensionContext } from "vscode"
+import { writeTextToClipboard, readTextFromClipboard } from "@/utils/env"
 
 /*
 Built using https://github.com/microsoft/vscode-webview-ui-toolkit
@@ -60,6 +65,9 @@ export async function activate(context: vscode.ExtensionContext) {
 	// Migrate custom instructions to global Cline rules (one-time cleanup)
 	await migrateCustomInstructionsToGlobalRules(context)
 
+	// Migrate mode from workspace storage to controller state (one-time cleanup)
+	await migrateModeFromWorkspaceStorageToControllerState(context)
+
 	// Clean up orphaned file context warnings (startup cleanup)
 	await FileContextTracker.cleanupOrphanedWarnings(context)
 
@@ -362,17 +370,17 @@ export async function activate(context: vscode.ExtensionContext) {
 			}
 
 			// Save current clipboard content
-			const tempCopyBuffer = await vscode.env.clipboard.readText()
+			const tempCopyBuffer = await readTextFromClipboard()
 
 			try {
 				// Copy the *existing* terminal selection (without selecting all)
 				await vscode.commands.executeCommand("workbench.action.terminal.copySelection")
 
 				// Get copied content
-				let terminalContents = (await vscode.env.clipboard.readText()).trim()
+				let terminalContents = (await readTextFromClipboard()).trim()
 
 				// Restore original clipboard content
-				await vscode.env.clipboard.writeText(tempCopyBuffer)
+				await writeTextToClipboard(tempCopyBuffer)
 
 				if (!terminalContents) {
 					// No terminal content was copied (either nothing selected or some error)
@@ -398,7 +406,7 @@ export async function activate(context: vscode.ExtensionContext) {
 				await visibleWebview?.controller.addSelectedTerminalOutputToChat(terminalContents, terminal.name)
 			} catch (error) {
 				// Ensure clipboard is restored even if an error occurs
-				await vscode.env.clipboard.writeText(tempCopyBuffer)
+				await writeTextToClipboard(tempCopyBuffer)
 				console.error("Error getting terminal contents:", error)
 				vscode.window.showErrorMessage("Failed to get terminal contents")
 			}

src/hosts/host-provider-types.ts

@@ -1,4 +1,9 @@
-import { UriServiceClientInterface, WatchServiceClientInterface } from "@generated/hosts/host-bridge-client-types"
+import {
+	UriServiceClientInterface,
+	WatchServiceClientInterface,
+	WorkspaceServiceClientInterface,
+	EnvServiceClientInterface,
+} from "@generated/hosts/host-bridge-client-types"
 
 /**
  * Interface for host bridge client providers
@@ -6,6 +11,8 @@ import { UriServiceClientInterface, WatchServiceClientInterface } from "@generat
 export interface HostBridgeClientProvider {
 	uriServiceClient: UriServiceClientInterface
 	watchServiceClient: WatchServiceClientInterface
+	workspaceClient: WorkspaceServiceClientInterface
+	envClient: EnvServiceClientInterface
 }
 
 /**

src/hosts/vscode/client/host-grpc-client.ts

@@ -5,4 +5,6 @@ import * as host from "@shared/proto/index.host"
 export const vscodeHostBridgeClient: HostBridgeClientProvider = {
 	uriServiceClient: createGrpcClient(host.UriServiceDefinition),
 	watchServiceClient: createGrpcClient(host.WatchServiceDefinition),
+	workspaceClient: createGrpcClient(host.WorkspaceServiceDefinition),
+	envClient: createGrpcClient(host.EnvServiceDefinition),
 }

src/hosts/vscode/env/clipboardReadText.ts

@@ -0,0 +1,7 @@
+import { EmptyRequest, String } from "@/shared/proto/common"
+import * as vscode from "vscode"
+
+export async function clipboardReadText(_: EmptyRequest): Promise<String> {
+	const text = await vscode.env.clipboard.readText()
+	return String.create({ value: text })
+}

src/hosts/vscode/env/clipboardWriteText.ts

@@ -0,0 +1,7 @@
+import { StringRequest, Empty } from "@/shared/proto/common"
+import * as vscode from "vscode"
+
+export async function clipboardWriteText(request: StringRequest): Promise<Empty> {
+	await vscode.env.clipboard.writeText(request.value)
+	return Empty.create({})
+}

src/hosts/vscode/workspace/getWorkspacePaths.ts

@@ -0,0 +1,6 @@
+import { GetWorkspacePathsRequest, GetWorkspacePathsResponse } from "@/shared/proto/index.host"
+import * as vscode from "vscode"
+export async function getWorkspacePaths(_: GetWorkspacePathsRequest): Promise<GetWorkspacePathsResponse> {
+	const paths = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath) ?? []
+	return GetWorkspacePathsResponse.create({ paths: paths })
+}

src/integrations/claude-code/message-filter.ts

@@ -0,0 +1,33 @@
+import type { Anthropic } from "@anthropic-ai/sdk"
+
+/**
+ * Filters out image blocks from messages since Claude Code doesn't support images.
+ * Replaces image blocks with text placeholders similar to how VSCode LM provider handles it.
+ */
+export function filterMessagesForClaudeCode(messages: Anthropic.Messages.MessageParam[]): Anthropic.Messages.MessageParam[] {
+	return messages.map((message) => {
+		// Handle simple string messages
+		if (typeof message.content === "string") {
+			return message
+		}
+
+		// Handle complex message structures
+		const filteredContent = message.content.map((block) => {
+			if (block.type === "image") {
+				// Replace image blocks with text placeholders
+				const sourceType = block.source?.type || "unknown"
+				const mediaType = block.source?.media_type || "unknown"
+				return {
+					type: "text" as const,
+					text: `[Image (${sourceType}): ${mediaType} not supported by Claude Code]`,
+				}
+			}
+			return block
+		})
+
+		return {
+			...message,
+			content: filteredContent,
+		}
+	})
+}

src/integrations/claude-code/run.ts

@@ -1,31 +1,124 @@
 import * as vscode from "vscode"
-import Anthropic from "@anthropic-ai/sdk"
+import type Anthropic from "@anthropic-ai/sdk"
 import { execa } from "execa"
+import { ClaudeCodeMessage } from "./types"
+import readline from "readline"
 
 const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0)
 
-export function runClaudeCode({
-	systemPrompt,
-	messages,
-	path,
-	modelId,
-}: {
+type ClaudeCodeOptions = {
 	systemPrompt: string
 	messages: Anthropic.Messages.MessageParam[]
 	path?: string
 	modelId?: string
-}) {
+}
+
+type ProcessState = {
+	partialData: string | null
+	error: Error | null
+	stderrLogs: string
+	exitCode: number | null
+}
+
+export async function* runClaudeCode(options: ClaudeCodeOptions): AsyncGenerator<ClaudeCodeMessage | string> {
+	const process = runProcess(options)
+
+	const rl = readline.createInterface({
+		input: process.stdout,
+	})
+
+	try {
+		const processState: ProcessState = {
+			error: null,
+			stderrLogs: "",
+			exitCode: null,
+			partialData: null,
+		}
+
+		process.stderr.on("data", (data) => {
+			processState.stderrLogs += data.toString()
+		})
+
+		process.on("close", (code) => {
+			processState.exitCode = code
+		})
+
+		process.on("error", (err) => {
+			processState.error = err
+		})
+
+		for await (const line of rl) {
+			if (processState.error) {
+				throw processState.error
+			}
+
+			if (line.trim()) {
+				const chunk = parseChunk(line, processState)
+
+				if (!chunk) {
+					continue
+				}
+
+				yield chunk
+			}
+		}
+
+		// We rely on the assistant message. If the output was truncated, it's better having a poorly formatted message
+		// from which to extract something, than throwing an error/showing the model didn't return any messages.
+		if (processState.partialData && processState.partialData.startsWith(`{"type":"assistant"`)) {
+			yield processState.partialData
+		}
+
+		const { exitCode } = await process
+		if (exitCode !== null && exitCode !== 0) {
+			const errorOutput = processState.error?.message || processState.stderrLogs?.trim()
+			throw new Error(
+				`Claude Code process exited with code ${exitCode}.${errorOutput ? ` Error output: ${errorOutput}` : ""}`,
+			)
+		}
+	} finally {
+		rl.close()
+		if (!process.killed) {
+			process.kill()
+		}
+	}
+}
+
+// We want the model to use our custom tool format instead of built-in tools.
+// Disabling built-in tools prevents tool-only responses and ensures text output.
+const claudeCodeTools = [
+	"Task",
+	"Bash",
+	"Glob",
+	"Grep",
+	"LS",
+	"exit_plan_mode",
+	"Read",
+	"Edit",
+	"MultiEdit",
+	"Write",
+	"NotebookRead",
+	"NotebookEdit",
+	"WebFetch",
+	"TodoRead",
+	"TodoWrite",
+	"WebSearch",
+].join(",")
+
+const CLAUDE_CODE_TIMEOUT = 600000 // 10 minutes
+
+function runProcess({ systemPrompt, messages, path, modelId }: ClaudeCodeOptions) {
 	const claudePath = path || "claude"
 
-	// TODO: Is it worh using sessions? Where do we store the session ID?
 	const args = [
 		"-p",
-		JSON.stringify(messages),
 		"--system-prompt",
 		systemPrompt,
 		"--verbose",
 		"--output-format",
 		"stream-json",
+		"--disallowedTools",
+		claudeCodeTools,
 		// Cline will handle recursive calls
 		"--max-turns",
 		"1",
@@ -35,11 +128,54 @@ export function runClaudeCode({
 		args.push("--model", modelId)
 	}
 
-	return execa(claudePath, args, {
-		stdin: "ignore",
+	const claudeCodeProcess = execa(claudePath, args, {
+		stdin: "pipe",
 		stdout: "pipe",
 		stderr: "pipe",
-		env: process.env,
+		env: {
+			...process.env,
+			// The default is 32000. However, I've gotten larger responses, so we increase it unless the user specified it.
+			CLAUDE_CODE_MAX_OUTPUT_TOKENS: process.env.CLAUDE_CODE_MAX_OUTPUT_TOKENS || "64000",
+		},
 		cwd,
+		maxBuffer: 1024 * 1024 * 1000,
+		timeout: CLAUDE_CODE_TIMEOUT,
 	})
+
+	claudeCodeProcess.stdin.write(JSON.stringify(messages))
+	claudeCodeProcess.stdin.end()
+
+	return claudeCodeProcess
+}
+
+function parseChunk(data: string, processState: ProcessState) {
+	if (processState.partialData) {
+		processState.partialData += data
+
+		const chunk = attemptParseChunk(processState.partialData)
+
+		if (!chunk) {
+			return null
+		}
+
+		processState.partialData = null
+		return chunk
+	}
+
+	const chunk = attemptParseChunk(data)
+
+	if (!chunk) {
+		processState.partialData = data
+	}
+
+	return chunk
+}
+
+function attemptParseChunk(data: string): ClaudeCodeMessage | null {
+	try {
+		return JSON.parse(data)
+	} catch (error) {
+		console.error("Error parsing chunk:", error, data.length)
+		return null
+	}
 }

src/integrations/claude-code/types.ts

@@ -1,34 +1,17 @@
+import type { Anthropic } from "@anthropic-ai/sdk"
+
 type InitMessage = {
 	type: "system"
 	subtype: "init"
 	session_id: string
 	tools: string[]
 	mcp_servers: string[]
-}
-
-type ClaudeCodeContent = {
-	type: "text"
-	text: string
+	apiKeySource: "none" | "/login managed key" | string
 }
 
 type AssistantMessage = {
 	type: "assistant"
-	message: {
-		id: string
-		type: "message"
-		role: "assistant"
-		model: string
-		content: ClaudeCodeContent[]
-		stop_reason: null
-		stop_sequence: null
-		usage: {
-			input_tokens: number
-			cache_creation_input_tokens?: number
-			cache_read_input_tokens?: number
-			output_tokens: number
-			service_tier: "standard"
-		}
-	}
+	message: Anthropic.Messages.Message
 	session_id: string
 }
 
@@ -39,13 +22,12 @@ type ErrorMessage = {
 type ResultMessage = {
 	type: "result"
 	subtype: "success"
-	cost_usd: number
+	total_cost_usd: number
 	is_error: boolean
 	duration_ms: number
 	duration_api_ms: number
 	num_turns: number
 	result: string
-	total_cost: number
 	session_id: string
 }
 

src/integrations/git/commit-message-generator.ts

@@ -1,5 +1,6 @@
 import * as vscode from "vscode"
 import { getWorkingState } from "@utils/git"
+import { writeTextToClipboard } from "@utils/env"
 
 /**
  * Formats the git diff into a prompt for the AI
@@ -57,7 +58,7 @@ export function extractCommitMessage(aiResponse: string): string {
  * @param message The commit message to copy
  */
 export async function copyCommitMessageToClipboard(message: string): Promise<void> {
-	await vscode.env.clipboard.writeText(message)
+	await writeTextToClipboard(message)
 	vscode.window.showInformationMessage("Commit message copied to clipboard")
 }
 

src/integrations/misc/process-images.ts

@@ -1,66 +0,0 @@
-import * as vscode from "vscode"
-import fs from "fs/promises"
-import * as path from "path"
-import sizeOf from "image-size"
-
-export async function selectImages(): Promise<string[]> {
-	const options: vscode.OpenDialogOptions = {
-		canSelectMany: true,
-		openLabel: "Select",
-		filters: {
-			Images: ["png", "jpg", "jpeg", "webp"], // supported by anthropic and openrouter
-		},
-	}
-
-	const fileUris = await vscode.window.showOpenDialog(options)
-
-	if (!fileUris || fileUris.length === 0) {
-		return []
-	}
-
-	const processedImagePromises = fileUris.map(async (uri) => {
-		const imagePath = uri.fsPath
-		let buffer: Buffer
-		try {
-			// Read the file into a buffer first
-			buffer = await fs.readFile(imagePath)
-			// Convert Node.js Buffer to Uint8Array
-			const uint8Array = new Uint8Array(buffer.buffer, buffer.byteOffset, buffer.byteLength)
-			const dimensions = sizeOf(uint8Array) // Get dimensions from Uint8Array
-			if (dimensions.width! > 7500 || dimensions.height! > 7500) {
-				console.warn(`Image dimensions exceed 7500px, skipping: ${imagePath}`)
-				vscode.window.showErrorMessage(
-					`Image too large: ${path.basename(imagePath)} was skipped (dimensions exceed 7500px).`,
-				)
-				return null
-			}
-		} catch (error) {
-			console.error(`Error reading file or getting dimensions for ${imagePath}:`, error)
-			vscode.window.showErrorMessage(`Could not read dimensions for ${path.basename(imagePath)}, skipping.`)
-			return null
-		}
-
-		// If dimensions are valid, proceed to convert the existing buffer to base64
-		const base64 = buffer.toString("base64")
-		const mimeType = getMimeType(imagePath)
-		return `data:${mimeType};base64,${base64}`
-	})
-
-	const dataUrlsWithNulls = await Promise.all(processedImagePromises)
-	return dataUrlsWithNulls.filter((url) => url !== null) as string[] // Filter out skipped images
-}
-
-function getMimeType(filePath: string): string {
-	const ext = path.extname(filePath).toLowerCase()
-	switch (ext) {
-		case ".png":
-			return "image/png"
-		case ".jpeg":
-		case ".jpg":
-			return "image/jpeg"
-		case ".webp":
-			return "image/webp"
-		default:
-			throw new Error(`Unsupported file type: ${ext}`)
-	}
-}

src/integrations/terminal/get-latest-output.ts

@@ -1,4 +1,5 @@
 import * as vscode from "vscode"
+import { readTextFromClipboard, writeTextToClipboard } from "@utils/env"
 
 /**
  * Gets the contents of the active terminal
@@ -6,7 +7,7 @@ import * as vscode from "vscode"
  */
 export async function getLatestTerminalOutput(): Promise<string> {
 	// Store original clipboard content to restore later
-	const originalClipboard = await vscode.env.clipboard.readText()
+	const originalClipboard = await readTextFromClipboard()
 
 	try {
 		// Select terminal content
@@ -19,7 +20,7 @@ export async function getLatestTerminalOutput(): Promise<string> {
 		await vscode.commands.executeCommand("workbench.action.terminal.clearSelection")
 
 		// Get terminal contents from clipboard
-		let terminalContents = (await vscode.env.clipboard.readText()).trim()
+		let terminalContents = (await readTextFromClipboard()).trim()
 
 		// Check if there's actually a terminal open
 		if (terminalContents === originalClipboard) {
@@ -40,6 +41,6 @@ export async function getLatestTerminalOutput(): Promise<string> {
 		return terminalContents
 	} finally {
 		// Restore original clipboard content
-		await vscode.env.clipboard.writeText(originalClipboard)
+		await writeTextToClipboard(originalClipboard)
 	}
 }

src/integrations/workspace/WorkspaceTracker.ts

@@ -2,13 +2,22 @@ import * as vscode from "vscode"
 import * as path from "path"
 import { listFiles } from "@services/glob/list-files"
 import { sendWorkspaceUpdateEvent } from "@core/controller/file/subscribeToWorkspaceUpdates"
-
-const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0)
+import { getCwd } from "@/utils/path"
 
 // Note: this is not a drop-in replacement for listFiles at the start of tasks, since that will be done for Desktops when there is no workspace selected
 class WorkspaceTracker {
 	private disposables: vscode.Disposable[] = []
 	private filePaths: Set<string> = new Set()
+	private cwd: string = ""
+
+	constructor() {
+		this.initializeCwd()
+		this.registerListeners()
+	}
+
+	private async initializeCwd() {
+		this.cwd = await getCwd()
+	}
 
 	private get activeFiles() {
 		return new Set(
@@ -18,16 +27,12 @@ class WorkspaceTracker {
 		)
 	}
 
-	constructor() {
-		this.registerListeners()
-	}
-
 	async populateFilePaths() {
 		// should not auto get filepaths for desktop since it would immediately show permission popup before cline ever creates a file
-		if (!cwd) {
+		if (!this.cwd) {
 			return
 		}
-		const [files, _] = await listFiles(cwd, true, 1_000)
+		const [files, _] = await listFiles(this.cwd, true, 1_000)
 		files.forEach((file) => this.filePaths.add(this.normalizeFilePath(file)))
 		this.workspaceDidUpdate()
 	}
@@ -91,18 +96,18 @@ class WorkspaceTracker {
 	}
 
 	private async workspaceDidUpdate() {
-		if (!cwd) {
+		if (!this.cwd) {
 			return
 		}
 		const filePaths = Array.from(new Set([...this.activeFiles, ...this.filePaths])).map((file) => {
-			const relativePath = path.relative(cwd, file).toPosix()
+			const relativePath = path.relative(this.cwd, file).toPosix()
 			return file.endsWith("/") ? relativePath + "/" : relativePath
 		})
 		await sendWorkspaceUpdateEvent(filePaths)
 	}
 
 	private normalizeFilePath(filePath: string): string {
-		const resolvedPath = cwd ? path.resolve(cwd, filePath) : path.resolve(filePath)
+		const resolvedPath = this.cwd ? path.resolve(this.cwd, filePath) : path.resolve(filePath)
 		return filePath.endsWith("/") ? resolvedPath + "/" : resolvedPath
 	}
 

src/shared/ChatSettings.ts

@@ -8,6 +8,9 @@ export interface ChatSettings {
 
 export type PartialChatSettings = Partial<ChatSettings>
 
+// Type for chat settings stored in workspace (excludes in-memory mode)
+export type StoredChatSettings = Omit<ChatSettings, "mode">
+
 export const DEFAULT_CHAT_SETTINGS: ChatSettings = {
 	mode: "act",
 	preferredLanguage: "English",

src/shared/WebviewMessage.ts

@@ -12,7 +12,6 @@ export interface WebviewMessage {
 		| "fetchMcpMarketplace"
 		| "searchCommits"
 		| "telemetrySetting"
-		| "clearAllTaskHistory"
 		| "grpc_request"
 		| "grpc_request_cancel"
 

src/shared/api.ts

@@ -10,6 +10,7 @@ export type ApiProvider =
 	| "ollama"
 	| "lmstudio"
 	| "gemini"
+	| "gemini-cli"
 	| "openai-native"
 	| "requesty"
 	| "together"
@@ -69,6 +70,8 @@ export interface ApiHandlerOptions {
 	lmStudioBaseUrl?: string
 	geminiApiKey?: string
 	geminiBaseUrl?: string
+	geminiCliOAuthPath?: string
+	geminiCliProjectId?: string
 	openAiNativeApiKey?: string
 	deepSeekApiKey?: string
 	requestyApiKey?: string
@@ -230,11 +233,31 @@ export const anthropicModels = {
 export type ClaudeCodeModelId = keyof typeof claudeCodeModels
 export const claudeCodeDefaultModelId: ClaudeCodeModelId = "claude-sonnet-4-20250514"
 export const claudeCodeModels = {
-	"claude-sonnet-4-20250514": anthropicModels["claude-sonnet-4-20250514"],
-	"claude-opus-4-20250514": anthropicModels["claude-opus-4-20250514"],
-	"claude-3-7-sonnet-20250219": anthropicModels["claude-3-7-sonnet-20250219"],
-	"claude-3-5-sonnet-20241022": anthropicModels["claude-3-5-sonnet-20241022"],
-	"claude-3-5-haiku-20241022": anthropicModels["claude-3-5-haiku-20241022"],
+	"claude-sonnet-4-20250514": {
+		...anthropicModels["claude-sonnet-4-20250514"],
+		supportsImages: false,
+		supportsPromptCache: false,
+	},
+	"claude-opus-4-20250514": {
+		...anthropicModels["claude-opus-4-20250514"],
+		supportsImages: false,
+		supportsPromptCache: false,
+	},
+	"claude-3-7-sonnet-20250219": {
+		...anthropicModels["claude-3-7-sonnet-20250219"],
+		supportsImages: false,
+		supportsPromptCache: false,
+	},
+	"claude-3-5-sonnet-20241022": {
+		...anthropicModels["claude-3-5-sonnet-20241022"],
+		supportsImages: false,
+		supportsPromptCache: false,
+	},
+	"claude-3-5-haiku-20241022": {
+		...anthropicModels["claude-3-5-haiku-20241022"],
+		supportsImages: false,
+		supportsPromptCache: false,
+	},
 } as const satisfies Record<string, ModelInfo>
 
 // AWS Bedrock
@@ -380,7 +403,7 @@ export const bedrockModels = {
 
 // OpenRouter
 // https://openrouter.ai/models?order=newest&supported_parameters=tools
-export const openRouterDefaultModelId = "anthropic/claude-3.7-sonnet" // will always exist in openRouterModels
+export const openRouterDefaultModelId = "anthropic/claude-sonnet-4" // will always exist in openRouterModels
 export const openRouterDefaultModelInfo: ModelInfo = {
 	maxTokens: 8192,
 	contextWindow: 200_000,
@@ -392,7 +415,7 @@ export const openRouterDefaultModelInfo: ModelInfo = {
 	cacheWritesPrice: 3.75,
 	cacheReadsPrice: 0.3,
 	description:
-		"Claude 3.7 Sonnet is an advanced large language model with improved reasoning, coding, and problem-solving capabilities. It introduces a hybrid reasoning approach, allowing users to choose between rapid responses and extended, step-by-step processing for complex tasks. The model demonstrates notable improvements in coding, particularly in front-end development and full-stack updates, and excels in agentic workflows, where it can autonomously navigate multi-step processes. \n\nClaude 3.7 Sonnet maintains performance parity with its predecessor in standard mode while offering an extended reasoning mode for enhanced accuracy in math, coding, and instruction-following tasks.\n\nRead more at the [blog post here](https://www.anthropic.com/news/claude-3-7-sonnet)",
+		"Claude Sonnet 4 delivers superior intelligence across coding, agentic search, and AI agent capabilities. It's a powerful choice for agentic coding, and can complete tasks across the entire software development lifecycle—from initial planning to bug fixes, maintenance to large refactors. It offers strong performance in both planning and solving for complex coding tasks, making it an ideal choice to power end-to-end software development processes.\n\nRead more in the [blog post here](https://www.anthropic.com/claude/sonnet)",
 }
 // Vertex AI
 // https://cloud.google.com/vertex-ai/generative-ai/docs/partner-models/use-claude
@@ -701,7 +724,7 @@ export const openAiModelInfoSaneDefaults: OpenAiCompatibleModelInfo = {
 // Gemini
 // https://ai.google.dev/gemini-api/docs/models/gemini
 export type GeminiModelId = keyof typeof geminiModels
-export const geminiDefaultModelId: GeminiModelId = "gemini-2.0-flash-001"
+export const geminiDefaultModelId: GeminiModelId = "gemini-2.5-pro"
 export const geminiModels = {
 	"gemini-2.5-pro": {
 		maxTokens: 65536,
@@ -855,6 +878,138 @@ export const geminiModels = {
 	},
 } as const satisfies Record<string, ModelInfo>
 
+// Gemini CLI (OAuth-based)
+export type GeminiCliModelId = keyof typeof geminiCliModels
+export const geminiCliDefaultModelId: GeminiCliModelId = "gemini-2.5-flash"
+export const geminiCliModels = {
+	"gemini-2.5-pro": {
+		maxTokens: 65536,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0, // Free tier via OAuth
+		outputPrice: 0, // Free tier via OAuth
+		description: "Google's Gemini 2.5 Pro model via OAuth (free tier)",
+	},
+	"gemini-2.5-flash": {
+		maxTokens: 65536,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0, // Free tier via OAuth
+		outputPrice: 0, // Free tier via OAuth
+		description: "Google's Gemini 2.5 Flash model via OAuth (free tier)",
+	},
+	"gemini-2.0-flash-001": {
+		maxTokens: 8192,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0, // Free tier via OAuth
+		outputPrice: 0, // Free tier via OAuth
+		description: "Google's Gemini 2.0 Flash model via OAuth (free tier)",
+	},
+	"gemini-2.0-flash-lite-preview-02-05": {
+		maxTokens: 8192,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 2.0 Flash Lite Preview model via OAuth",
+	},
+	"gemini-2.0-pro-exp-02-05": {
+		maxTokens: 8192,
+		contextWindow: 2_097_152,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 2.0 Pro Experimental model via OAuth",
+	},
+	"gemini-2.0-flash-thinking-exp-01-21": {
+		maxTokens: 65_536,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 2.0 Flash Thinking Experimental model via OAuth",
+	},
+	"gemini-2.0-flash-thinking-exp-1219": {
+		maxTokens: 8192,
+		contextWindow: 32_767,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 2.0 Flash Thinking Experimental (1219) model via OAuth",
+	},
+	"gemini-2.0-flash-exp": {
+		maxTokens: 8192,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 2.0 Flash Experimental model via OAuth",
+	},
+	"gemini-1.5-flash-002": {
+		maxTokens: 8192,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0, // Free tier via OAuth
+		outputPrice: 0, // Free tier via OAuth
+		description: "Google's Gemini 1.5 Flash 002 model via OAuth (free tier)",
+	},
+	"gemini-1.5-flash-exp-0827": {
+		maxTokens: 8192,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 1.5 Flash Experimental (0827) model via OAuth",
+	},
+	"gemini-1.5-flash-8b-exp-0827": {
+		maxTokens: 8192,
+		contextWindow: 1_048_576,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 1.5 Flash 8B Experimental model via OAuth",
+	},
+	"gemini-1.5-pro-002": {
+		maxTokens: 8192,
+		contextWindow: 2_097_152,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 1.5 Pro 002 model via OAuth",
+	},
+	"gemini-1.5-pro-exp-0827": {
+		maxTokens: 8192,
+		contextWindow: 2_097_152,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini 1.5 Pro Experimental model via OAuth",
+	},
+	"gemini-exp-1206": {
+		maxTokens: 8192,
+		contextWindow: 2_097_152,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 0,
+		outputPrice: 0,
+		description: "Google's Gemini Experimental (1206) model via OAuth",
+	},
+} as const satisfies Record<string, ModelInfo>
+
 // OpenAI Native
 // https://openai.com/api/pricing/
 export type OpenAiNativeModelId = keyof typeof openAiNativeModels
@@ -1856,7 +2011,7 @@ export const liteLlmModelInfoSaneDefaults: LiteLLMModelInfo = {
 // AskSage Models
 // https://docs.asksage.ai/
 export type AskSageModelId = keyof typeof askSageModels
-export const askSageDefaultModelId: AskSageModelId = "claude-35-sonnet"
+export const askSageDefaultModelId: AskSageModelId = "claude-4-sonnet"
 export const askSageDefaultURL: string = "https://api.asksage.ai/server"
 export const askSageModels = {
 	"gpt-4o": {
@@ -2354,6 +2509,14 @@ export const requestyDefaultModelInfo: ModelInfo = {
 export type SapAiCoreModelId = keyof typeof sapAiCoreModels
 export const sapAiCoreDefaultModelId: SapAiCoreModelId = "anthropic--claude-3.5-sonnet"
 export const sapAiCoreModels = {
+	"anthropic--claude-4-sonnet": {
+		maxTokens: 8192,
+		contextWindow: 200_000,
+		supportsImages: true,
+		supportsPromptCache: false,
+		inputPrice: 3.0,
+		outputPrice: 15.0,
+	},
 	"anthropic--claude-3.7-sonnet": {
 		maxTokens: 64_000,
 		contextWindow: 200_000,

src/shared/proto-conversions/models/api-configuration-conversion.ts

@@ -202,6 +202,8 @@ function convertApiProviderToProto(provider: string | undefined): ProtoApiProvid
 			return ProtoApiProvider.LMSTUDIO
 		case "gemini":
 			return ProtoApiProvider.GEMINI
+		case "gemini-cli":
+			return ProtoApiProvider.GEMINI_CLI
 		case "openai-native":
 			return ProtoApiProvider.OPENAI_NATIVE
 		case "requesty":
@@ -262,6 +264,8 @@ function convertProtoToApiProvider(provider: ProtoApiProvider): ApiProvider {
 			return "lmstudio"
 		case ProtoApiProvider.GEMINI:
 			return "gemini"
+		case ProtoApiProvider.GEMINI_CLI:
+			return "gemini-cli"
 		case ProtoApiProvider.OPENAI_NATIVE:
 			return "openai-native"
 		case ProtoApiProvider.REQUESTY:
@@ -379,6 +383,8 @@ export function convertApiConfigurationToProto(config: ApiConfiguration): ProtoA
 		sapAiCoreTokenUrl: config.sapAiCoreTokenUrl,
 		sapAiCoreBaseUrl: config.sapAiCoreBaseUrl,
 		claudeCodePath: config.claudeCodePath,
+		geminiCliOauthPath: config.geminiCliOAuthPath,
+		geminiCliProjectId: config.geminiCliProjectId,
 	}
 }
 
@@ -458,5 +464,7 @@ export function convertProtoToApiConfiguration(protoConfig: ProtoApiConfiguratio
 		sapAiCoreTokenUrl: protoConfig.sapAiCoreTokenUrl,
 		sapAiCoreBaseUrl: protoConfig.sapAiCoreBaseUrl,
 		claudeCodePath: protoConfig.claudeCodePath,
+		geminiCliOAuthPath: protoConfig.geminiCliOauthPath,
+		geminiCliProjectId: protoConfig.geminiCliProjectId,
 	}
 }

src/shared/proto-conversions/state/telemetry-setting-conversion.ts

@@ -0,0 +1,34 @@
+import { TelemetrySettingEnum } from "../../proto/state"
+import { TelemetrySetting } from "../../TelemetrySetting"
+
+/**
+ * Converts a domain TelemetrySetting string to a proto TelemetrySettingEnum
+ */
+export function convertDomainTelemetrySettingToProto(setting: TelemetrySetting): TelemetrySettingEnum {
+	switch (setting) {
+		case "unset":
+			return TelemetrySettingEnum.UNSET
+		case "enabled":
+			return TelemetrySettingEnum.ENABLED
+		case "disabled":
+			return TelemetrySettingEnum.DISABLED
+		default:
+			return TelemetrySettingEnum.UNSET
+	}
+}
+
+/**
+ * Converts a proto TelemetrySettingEnum to a domain TelemetrySetting string
+ */
+export function convertProtoTelemetrySettingToDomain(setting: TelemetrySettingEnum): TelemetrySetting {
+	switch (setting) {
+		case TelemetrySettingEnum.UNSET:
+			return "unset"
+		case TelemetrySettingEnum.ENABLED:
+			return "enabled"
+		case TelemetrySettingEnum.DISABLED:
+			return "disabled"
+		default:
+			return "unset"
+	}
+}

src/standalone/ExternalWebviewProvider.ts

@@ -10,7 +10,9 @@ https://github.com/KumarVariable/vscode-extension-sidebar-html/blob/master/src/c
 */
 
 export class ExternalWebviewProvider extends WebviewProvider {
-	private RESOURCE_AUTHORITY: string = "file.resources"
+	// This hostname cannot be changed without updating the external webview handler.
+	private RESOURCE_HOSTNAME: string = "internal.resources"
+
 	constructor(context: vscode.ExtensionContext, outputChannel: vscode.OutputChannel, providerType: WebviewProviderType) {
 		super(context, outputChannel, providerType)
 	}
@@ -19,10 +21,10 @@ export class ExternalWebviewProvider extends WebviewProvider {
 		if (uri.scheme !== "file") {
 			return uri
 		}
-		return URI.from({ scheme: "https", authority: this.RESOURCE_AUTHORITY, path: uri.fsPath })
+		return URI.from({ scheme: "https", authority: this.RESOURCE_HOSTNAME, path: uri.fsPath })
 	}
 	override getCspSource() {
-		return "csp-source"
+		return `'self' https://${this.RESOURCE_HOSTNAME}`
 	}
 	override postMessageToWebview(message: ExtensionMessage) {
 		console.log(`postMessageToWebview: ${message}`)

src/standalone/host-bridge-client-manager.ts

@@ -1,16 +1,28 @@
 import { Channel, createChannel } from "nice-grpc"
-import { UriServiceClientImpl, WatchServiceClientImpl } from "@generated/standalone/host-bridge-clients"
-import { UriServiceClientInterface, WatchServiceClientInterface } from "@generated/hosts/host-bridge-client-types"
+import {
+	UriServiceClientImpl,
+	WatchServiceClientImpl,
+	WorkspaceServiceClientImpl,
+	EnvServiceClientImpl,
+} from "@generated/standalone/host-bridge-clients"
+import {
+	UriServiceClientInterface,
+	WatchServiceClientInterface,
+	WorkspaceServiceClientInterface,
+	EnvServiceClientInterface,
+} from "@generated/hosts/host-bridge-client-types"
 import { HostBridgeClientProvider } from "@/hosts/host-provider-types"
 
 /**
- * Singleton class to hold the gRPC clients for the host bridge. The clients should be re-used to avoid
+ * Manager to hold the gRPC clients for the host bridge. The clients should be re-used to avoid
  * creating a new TCP connection every time a rpc is made.
  */
 export class ExternalHostBridgeClientManager implements HostBridgeClientProvider {
 	private channel: Channel
 	uriServiceClient: UriServiceClientInterface
 	watchServiceClient: WatchServiceClientInterface
+	workspaceClient: WorkspaceServiceClientInterface
+	envClient: EnvServiceClientInterface
 
 	constructor() {
 		const address = process.env.HOST_BRIDGE_ADDRESS || "localhost:50052"
@@ -18,6 +30,8 @@ export class ExternalHostBridgeClientManager implements HostBridgeClientProvider
 
 		this.uriServiceClient = new UriServiceClientImpl(this.channel)
 		this.watchServiceClient = new WatchServiceClientImpl(this.channel)
+		this.workspaceClient = new WorkspaceServiceClientImpl(this.channel)
+		this.envClient = new EnvServiceClientImpl(this.channel)
 	}
 
 	public close(): void {

src/standalone/standalone.ts

@@ -20,6 +20,10 @@ async function main() {
 	hostProviders.initializeHostProviders(createWebview, new ExternalHostBridgeClientManager())
 	activate(extensionContext)
 	const controller = new Controller(extensionContext, outputChannel, postMessage, uuidv4())
+	startProtobusService(controller)
+}
+
+function startProtobusService(controller: Controller) {
 	const server = new grpc.Server()
 
 	// Set up health check.
@@ -29,12 +33,15 @@ async function main() {
 	// Add all the handlers for the ProtoBus services to the server.
 	addProtobusServices(server, controller, wrapHandler, wrapStreamingResponseHandler)
 
-	// Set up reflection.
-	const reflection = new ReflectionService(getPackageDefinition())
+	// Create reflection service with protobus service names
+	const packageDefinition = getPackageDefinition()
+	const reflection = new ReflectionService(packageDefinition, {
+		services: getProtobusServiceNames(packageDefinition),
+	})
 	reflection.addToServer(server)
 
 	// Start the server.
-	const host = "127.0.0.1:50051"
+	const host = process.env.PROTOBUS_ADDRESS || "127.0.0.1:50051"
 	server.bindAsync(host, grpc.ServerCredentials.createInsecure(), (err) => {
 		if (err) {
 			log(`Error: Failed to bind to ${host}, port may be unavailable. ${err.message}`)
@@ -45,6 +52,14 @@ async function main() {
 	})
 }
 
+function getProtobusServiceNames(packageDefinition: { [x: string]: any }): string[] {
+	// Filter service names to only include cline services
+	const protobusServiceNames = Object.keys(packageDefinition).filter(
+		(name) => name.startsWith("cline.") || name.startsWith("grpc.health"),
+	)
+	return protobusServiceNames
+}
+
 const createWebview = () => {
 	return new ExternalWebviewProvider(extensionContext, outputChannel, WebviewProviderType.SIDEBAR)
 }

src/standalone/utils.ts

@@ -11,9 +11,9 @@ const log = (...args: unknown[]) => {
 function getPackageDefinition() {
 	// Load service definitions.
 	const descriptorSet = fs.readFileSync("proto/descriptor_set.pb")
-	const clineDef = protoLoader.loadFileDescriptorSetFromBuffer(descriptorSet)
+	const descriptorDefs = protoLoader.loadFileDescriptorSetFromBuffer(descriptorSet)
 	const healthDef = protoLoader.loadSync(health.protoPath)
-	const packageDefinition = { ...clineDef, ...healthDef }
+	const packageDefinition = { ...descriptorDefs, ...healthDef }
 	return packageDefinition
 }
 

src/utils/env.ts

@@ -0,0 +1,32 @@
+import { getHostBridgeProvider } from "@/hosts/host-providers"
+import { StringRequest, EmptyRequest } from "@/shared/proto/common"
+
+/**
+ * Writes text to the system clipboard
+ * @param text The text to write to the clipboard
+ * @returns Promise that resolves when the operation is complete
+ * @throws Error if the operation fails
+ */
+export async function writeTextToClipboard(text: string): Promise<void> {
+	try {
+		await getHostBridgeProvider().envClient.clipboardWriteText(StringRequest.create({ value: text }))
+	} catch (error) {
+		const errorMessage = error instanceof Error ? error.message : String(error)
+		throw new Error(`Failed to write to clipboard: ${errorMessage}`)
+	}
+}
+
+/**
+ * Reads text from the system clipboard
+ * @returns Promise that resolves to the clipboard text
+ * @throws Error if the operation fails
+ */
+export async function readTextFromClipboard(): Promise<string> {
+	try {
+		const response = await getHostBridgeProvider().envClient.clipboardReadText(EmptyRequest.create({}))
+		return response.value
+	} catch (error) {
+		const errorMessage = error instanceof Error ? error.message : String(error)
+		throw new Error(`Failed to read from clipboard: ${errorMessage}`)
+	}
+}

src/utils/github-url-utils.ts

@@ -17,6 +17,7 @@ import * as vscode from "vscode"
 import * as cp from "child_process"
 import * as os from "os"
 import * as util from "util"
+import { writeTextToClipboard } from "@/utils/env"
 
 /**
  * Creates a properly encoded GitHub issue URL.
@@ -81,7 +82,7 @@ export async function openUrlInBrowser(url: string): Promise<void> {
 
 	// Always copy to clipboard as a fallback
 	try {
-		await vscode.env.clipboard.writeText(url)
+		await writeTextToClipboard(url)
 		console.log("URL copied to clipboard as backup")
 	} catch (error) {
 		console.error(`Failed to copy URL to clipboard: ${error}`)
@@ -159,7 +160,7 @@ export async function openUrlInBrowser(url: string): Promise<void> {
 				)
 				.then((selection) => {
 					if (selection === "Copy URL Again") {
-						vscode.env.clipboard.writeText(url)
+						writeTextToClipboard(url)
 					}
 				})
 		}

src/utils/model-utils.ts

@@ -3,5 +3,13 @@ import { ApiHandler } from "@api/index"
 export function isClaude4ModelFamily(api: ApiHandler): boolean {
 	const model = api.getModel()
 	const modelId = model.id
-	return modelId.includes("sonnet-4") || modelId.includes("opus-4")
+	return (
+		modelId.includes("sonnet-4") || modelId.includes("opus-4") || modelId.includes("4-sonnet") || modelId.includes("4-opus")
+	)
+}
+
+export function isGemini2dot5ModelFamily(api: ApiHandler): boolean {
+	const model = api.getModel()
+	const modelId = model.id
+	return modelId.includes("gemini-2.5")
 }

src/utils/path.ts

@@ -1,6 +1,7 @@
 import * as path from "path"
 import os from "os"
 import * as vscode from "vscode"
+import { getHostBridgeProvider } from "@/hosts/host-providers"
 
 /*
 The Node.js 'path' module resolves and normalizes paths differently depending on the platform:
@@ -101,9 +102,17 @@ export function getReadablePath(cwd: string, relPath?: string): string {
 	}
 }
 
-export const getWorkspacePath = (defaultCwdPath = "") => {
-	const cwdPath = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) || defaultCwdPath
+// Returns the path of the first workspace directory, or the defaultCwdPath if there is no workspace open.
+export const getCwd = async (defaultCwdPath = ""): Promise<string> => {
+	const workspaceFolders = await getHostBridgeProvider().workspaceClient.getWorkspacePaths({})
+	return workspaceFolders.paths.shift() || defaultCwdPath
+}
+
+// Returns the workspace path of the file in the current editor.
+// If there is no path, it returns the top level workspace directory.
+export const getWorkspacePath = async (defaultCwdPath = "") => {
 	const currentFileUri = vscode.window.activeTextEditor?.document.uri
+	const cwdPath = await getCwd(defaultCwdPath)
 	if (currentFileUri) {
 		const workspaceFolder = vscode.workspace.getWorkspaceFolder(currentFileUri)
 		return workspaceFolder?.uri.fsPath || cwdPath
@@ -111,8 +120,8 @@ export const getWorkspacePath = (defaultCwdPath = "") => {
 	return cwdPath
 }
 
-export const isLocatedInWorkspace = (pathToCheck: string = ""): boolean => {
-	const workspacePath = getWorkspacePath()
+export const isLocatedInWorkspace = async (pathToCheck: string = ""): Promise<boolean> => {
+	const workspacePath = await getWorkspacePath()
 
 	// Handle long paths in Windows
 	if (pathToCheck.startsWith("\\\\?\\") || workspacePath.startsWith("\\\\?\\")) {