diff --git a/src/core/Cline.ts b/src/core/Cline.ts index 94364bb..86a4ec5 100644 --- a/src/core/Cline.ts +++ b/src/core/Cline.ts @@ -60,6 +60,8 @@ import { BrowserSession } from "../services/browser/BrowserSession" import { OpenRouterHandler } from "../api/providers/openrouter" import { McpHub } from "../services/mcp/McpHub" import crypto from "crypto" +import { insertGroups } from "./diff/insert-groups" +import { EXPERIMENT_IDS, experiments as Experiments } from "../shared/experiments" const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) ?? path.join(os.homedir(), "Desktop") // may or may not exist but fs checking existence would immediately ask for permission which would be bad UX, need to come up with a better solution @@ -115,7 +117,7 @@ export class Cline { task?: string | undefined, images?: string[] | undefined, historyItem?: HistoryItem | undefined, - experimentalDiffStrategy: boolean = false, + experiments?: Record, ) { if (!task && !images && !historyItem) { throw new Error("Either historyItem or task/images must be provided") @@ -137,7 +139,7 @@ export class Cline { } // Initialize diffStrategy based on current state - this.updateDiffStrategy(experimentalDiffStrategy) + this.updateDiffStrategy(Experiments.isEnabled(experiments ?? {}, EXPERIMENT_IDS.DIFF_STRATEGY)) if (task || images) { this.startTask(task, images) @@ -150,9 +152,8 @@ export class Cline { async updateDiffStrategy(experimentalDiffStrategy?: boolean) { // If not provided, get from current state if (experimentalDiffStrategy === undefined) { - const { experimentalDiffStrategy: stateExperimentalDiffStrategy } = - (await this.providerRef.deref()?.getState()) ?? {} - experimentalDiffStrategy = stateExperimentalDiffStrategy ?? false + const { experiments: stateExperimental } = (await this.providerRef.deref()?.getState()) ?? {} + experimentalDiffStrategy = stateExperimental?.[EXPERIMENT_IDS.DIFF_STRATEGY] ?? false } this.diffStrategy = getDiffStrategy(this.api.getModel().id, this.fuzzyMatchThreshold, experimentalDiffStrategy) } @@ -809,7 +810,7 @@ export class Cline { }) } - const { browserViewportSize, mode, customModePrompts, preferredLanguage } = + const { browserViewportSize, mode, customModePrompts, preferredLanguage, experiments } = (await this.providerRef.deref()?.getState()) ?? {} const { customModes } = (await this.providerRef.deref()?.getState()) ?? {} const systemPrompt = await (async () => { @@ -830,6 +831,7 @@ export class Cline { this.customInstructions, preferredLanguage, this.diffEnabled, + experiments, ) })() @@ -1008,6 +1010,10 @@ export class Cline { return `[${block.name} for '${block.params.regex}'${ block.params.file_pattern ? ` in '${block.params.file_pattern}'` : "" }]` + case "insert_code_block": + return `[${block.name} for '${block.params.path}']` + case "search_and_replace": + return `[${block.name} for '${block.params.path}']` case "list_files": return `[${block.name} for '${block.params.path}']` case "list_code_definition_names": @@ -1479,6 +1485,323 @@ export class Cline { break } } + + case "insert_code_block": { + const relPath: string | undefined = block.params.path + const operations: string | undefined = block.params.operations + + const sharedMessageProps: ClineSayTool = { + tool: "appliedDiff", + path: getReadablePath(cwd, removeClosingTag("path", relPath)), + } + + try { + if (block.partial) { + const partialMessage = JSON.stringify(sharedMessageProps) + await this.ask("tool", partialMessage, block.partial).catch(() => {}) + break + } + + // Validate required parameters + if (!relPath) { + this.consecutiveMistakeCount++ + pushToolResult(await this.sayAndCreateMissingParamError("insert_code_block", "path")) + break + } + + if (!operations) { + this.consecutiveMistakeCount++ + pushToolResult( + await this.sayAndCreateMissingParamError("insert_code_block", "operations"), + ) + break + } + + const absolutePath = path.resolve(cwd, relPath) + const fileExists = await fileExistsAtPath(absolutePath) + + if (!fileExists) { + this.consecutiveMistakeCount++ + const formattedError = `File does not exist at path: ${absolutePath}\n\n\nThe specified file could not be found. Please verify the file path and try again.\n` + await this.say("error", formattedError) + pushToolResult(formattedError) + break + } + + let parsedOperations: Array<{ + start_line: number + content: string + }> + + try { + parsedOperations = JSON.parse(operations) + if (!Array.isArray(parsedOperations)) { + throw new Error("Operations must be an array") + } + } catch (error) { + this.consecutiveMistakeCount++ + await this.say("error", `Failed to parse operations JSON: ${error.message}`) + pushToolResult(formatResponse.toolError("Invalid operations JSON format")) + break + } + + this.consecutiveMistakeCount = 0 + + // Read the file + const fileContent = await fs.readFile(absolutePath, "utf8") + this.diffViewProvider.editType = "modify" + this.diffViewProvider.originalContent = fileContent + const lines = fileContent.split("\n") + + const updatedContent = insertGroups( + lines, + parsedOperations.map((elem) => { + return { + index: elem.start_line - 1, + elements: elem.content.split("\n"), + } + }), + ).join("\n") + + // Show changes in diff view + if (!this.diffViewProvider.isEditing) { + await this.ask("tool", JSON.stringify(sharedMessageProps), true).catch(() => {}) + // First open with original content + await this.diffViewProvider.open(relPath) + await this.diffViewProvider.update(fileContent, false) + this.diffViewProvider.scrollToFirstDiff() + await delay(200) + } + + const diff = formatResponse.createPrettyPatch(relPath, fileContent, updatedContent) + + if (!diff) { + pushToolResult(`No changes needed for '${relPath}'`) + break + } + + await this.diffViewProvider.update(updatedContent, true) + + const completeMessage = JSON.stringify({ + ...sharedMessageProps, + diff, + } satisfies ClineSayTool) + + const didApprove = await this.ask("tool", completeMessage, false).then( + (response) => response.response === "yesButtonClicked", + ) + + if (!didApprove) { + await this.diffViewProvider.revertChanges() + pushToolResult("Changes were rejected by the user.") + break + } + + const { newProblemsMessage, userEdits, finalContent } = + await this.diffViewProvider.saveChanges() + this.didEditFile = true + + if (!userEdits) { + pushToolResult( + `The code block was successfully inserted in ${relPath.toPosix()}.${newProblemsMessage}`, + ) + await this.diffViewProvider.reset() + break + } + + const userFeedbackDiff = JSON.stringify({ + tool: "appliedDiff", + path: getReadablePath(cwd, relPath), + diff: userEdits, + } satisfies ClineSayTool) + + console.debug("[DEBUG] User made edits, sending feedback diff:", userFeedbackDiff) + await this.say("user_feedback_diff", userFeedbackDiff) + pushToolResult( + `The user made the following updates to your content:\n\n${userEdits}\n\n` + + `The updated content, which includes both your original modifications and the user's edits, has been successfully saved to ${relPath.toPosix()}. Here is the full, updated content of the file:\n\n` + + `\n${finalContent}\n\n\n` + + `Please note:\n` + + `1. You do not need to re-write the file with these changes, as they have already been applied.\n` + + `2. Proceed with the task using this updated file content as the new baseline.\n` + + `3. If the user's edits have addressed part of the task or changed the requirements, adjust your approach accordingly.` + + `${newProblemsMessage}`, + ) + await this.diffViewProvider.reset() + } catch (error) { + handleError("insert block", error) + await this.diffViewProvider.reset() + } + break + } + + case "search_and_replace": { + const relPath: string | undefined = block.params.path + const operations: string | undefined = block.params.operations + + const sharedMessageProps: ClineSayTool = { + tool: "appliedDiff", + path: getReadablePath(cwd, removeClosingTag("path", relPath)), + } + + try { + if (block.partial) { + const partialMessage = JSON.stringify({ + path: removeClosingTag("path", relPath), + operations: removeClosingTag("operations", operations), + }) + await this.ask("tool", partialMessage, block.partial).catch(() => {}) + break + } else { + if (!relPath) { + this.consecutiveMistakeCount++ + pushToolResult( + await this.sayAndCreateMissingParamError("search_and_replace", "path"), + ) + break + } + if (!operations) { + this.consecutiveMistakeCount++ + pushToolResult( + await this.sayAndCreateMissingParamError("search_and_replace", "operations"), + ) + break + } + + const absolutePath = path.resolve(cwd, relPath) + const fileExists = await fileExistsAtPath(absolutePath) + + if (!fileExists) { + this.consecutiveMistakeCount++ + const formattedError = `File does not exist at path: ${absolutePath}\n\n\nThe specified file could not be found. Please verify the file path and try again.\n` + await this.say("error", formattedError) + pushToolResult(formattedError) + break + } + + let parsedOperations: Array<{ + search: string + replace: string + start_line?: number + end_line?: number + use_regex?: boolean + ignore_case?: boolean + regex_flags?: string + }> + + try { + parsedOperations = JSON.parse(operations) + if (!Array.isArray(parsedOperations)) { + throw new Error("Operations must be an array") + } + } catch (error) { + this.consecutiveMistakeCount++ + await this.say("error", `Failed to parse operations JSON: ${error.message}`) + pushToolResult(formatResponse.toolError("Invalid operations JSON format")) + break + } + + // Read the original file content + const fileContent = await fs.readFile(absolutePath, "utf-8") + this.diffViewProvider.editType = "modify" + this.diffViewProvider.originalContent = fileContent + let lines = fileContent.split("\n") + + for (const op of parsedOperations) { + const flags = op.regex_flags ?? (op.ignore_case ? "gi" : "g") + const multilineFlags = flags.includes("m") ? flags : flags + "m" + + const searchPattern = op.use_regex + ? new RegExp(op.search, multilineFlags) + : new RegExp(escapeRegExp(op.search), multilineFlags) + + if (op.start_line || op.end_line) { + const startLine = Math.max((op.start_line ?? 1) - 1, 0) + const endLine = Math.min((op.end_line ?? lines.length) - 1, lines.length - 1) + + // Get the content before and after the target section + const beforeLines = lines.slice(0, startLine) + const afterLines = lines.slice(endLine + 1) + + // Get the target section and perform replacement + const targetContent = lines.slice(startLine, endLine + 1).join("\n") + const modifiedContent = targetContent.replace(searchPattern, op.replace) + const modifiedLines = modifiedContent.split("\n") + + // Reconstruct the full content with the modified section + lines = [...beforeLines, ...modifiedLines, ...afterLines] + } else { + // Global replacement + const fullContent = lines.join("\n") + const modifiedContent = fullContent.replace(searchPattern, op.replace) + lines = modifiedContent.split("\n") + } + } + + const newContent = lines.join("\n") + + this.consecutiveMistakeCount = 0 + + // Show diff preview + const diff = formatResponse.createPrettyPatch(relPath, fileContent, newContent) + + if (!diff) { + pushToolResult(`No changes needed for '${relPath}'`) + break + } + + await this.diffViewProvider.open(relPath) + await this.diffViewProvider.update(newContent, true) + this.diffViewProvider.scrollToFirstDiff() + + const completeMessage = JSON.stringify({ + ...sharedMessageProps, + diff: diff, + } satisfies ClineSayTool) + + const didApprove = await askApproval("tool", completeMessage) + if (!didApprove) { + await this.diffViewProvider.revertChanges() // This likely handles closing the diff view + break + } + + const { newProblemsMessage, userEdits, finalContent } = + await this.diffViewProvider.saveChanges() + this.didEditFile = true // used to determine if we should wait for busy terminal to update before sending api request + if (userEdits) { + await this.say( + "user_feedback_diff", + JSON.stringify({ + tool: fileExists ? "editedExistingFile" : "newFileCreated", + path: getReadablePath(cwd, relPath), + diff: userEdits, + } satisfies ClineSayTool), + ) + pushToolResult( + `The user made the following updates to your content:\n\n${userEdits}\n\n` + + `The updated content, which includes both your original modifications and the user's edits, has been successfully saved to ${relPath.toPosix()}. Here is the full, updated content of the file, including line numbers:\n\n` + + `\n${addLineNumbers(finalContent || "")}\n\n\n` + + `Please note:\n` + + `1. You do not need to re-write the file with these changes, as they have already been applied.\n` + + `2. Proceed with the task using this updated file content as the new baseline.\n` + + `3. If the user's edits have addressed part of the task or changed the requirements, adjust your approach accordingly.` + + `${newProblemsMessage}`, + ) + } else { + pushToolResult( + `Changes successfully applied to ${relPath.toPosix()}:\n\n${newProblemsMessage}`, + ) + } + await this.diffViewProvider.reset() + break + } + } catch (error) { + await handleError("applying search and replace", error) + await this.diffViewProvider.reset() + break + } + } + case "read_file": { const relPath: string | undefined = block.params.path const sharedMessageProps: ClineSayTool = { @@ -2751,3 +3074,7 @@ export class Cline { return `\n${details.trim()}\n` } } + +function escapeRegExp(string: string): string { + return string.replace(/[.*+?^${}()|[\]\\]/g, "\\$&") +} diff --git a/src/core/assistant-message/index.ts b/src/core/assistant-message/index.ts index 56e5e66..092dd06 100644 --- a/src/core/assistant-message/index.ts +++ b/src/core/assistant-message/index.ts @@ -13,6 +13,8 @@ export const toolUseNames = [ "read_file", "write_to_file", "apply_diff", + "insert_code_block", + "search_and_replace", "search_files", "list_files", "list_code_definition_names", @@ -50,6 +52,7 @@ export const toolParamNames = [ "end_line", "mode_slug", "reason", + "operations", ] as const export type ToolParamName = (typeof toolParamNames)[number] @@ -78,6 +81,11 @@ export interface WriteToFileToolUse extends ToolUse { params: Partial, "path" | "content" | "line_count">> } +export interface InsertCodeBlockToolUse extends ToolUse { + name: "insert_code_block" + params: Partial, "path" | "operations">> +} + export interface SearchFilesToolUse extends ToolUse { name: "search_files" params: Partial, "path" | "regex" | "file_pattern">> diff --git a/src/core/diff/insert-groups.ts b/src/core/diff/insert-groups.ts new file mode 100644 index 0000000..805f658 --- /dev/null +++ b/src/core/diff/insert-groups.ts @@ -0,0 +1,31 @@ +/** + * Inserts multiple groups of elements at specified indices in an array + * @param original Array to insert into, split by lines + * @param insertGroups Array of groups to insert, each with an index and elements to insert + * @returns New array with all insertions applied + */ +export interface InsertGroup { + index: number + elements: string[] +} + +export function insertGroups(original: string[], insertGroups: InsertGroup[]): string[] { + // Sort groups by index to maintain order + insertGroups.sort((a, b) => a.index - b.index) + + let result: string[] = [] + let lastIndex = 0 + + insertGroups.forEach(({ index, elements }) => { + // Add elements from original array up to insertion point + result.push(...original.slice(lastIndex, index)) + // Add the group of elements + result.push(...elements) + lastIndex = index + }) + + // Add remaining elements from original array + result.push(...original.slice(lastIndex)) + + return result +} diff --git a/src/core/diff/strategies/search-replace.ts b/src/core/diff/strategies/search-replace.ts index 1ede3c3..959f949 100644 --- a/src/core/diff/strategies/search-replace.ts +++ b/src/core/diff/strategies/search-replace.ts @@ -273,7 +273,7 @@ Your search/replace content here : "" return { success: false, - error: `No sufficiently similar match found${lineRange} (${Math.floor(bestMatchScore * 100)}% similar, needs ${Math.floor(this.fuzzyThreshold * 100)}%)\n\nDebug Info:\n- Similarity Score: ${Math.floor(bestMatchScore * 100)}%\n- Required Threshold: ${Math.floor(this.fuzzyThreshold * 100)}%\n- Search Range: ${startLine && endLine ? `lines ${startLine}-${endLine}` : "start to end"}\n\nSearch Content:\n${searchChunk}${bestMatchSection}${originalContentSection}`, + error: `No sufficiently similar match found${lineRange} (${Math.floor(bestMatchScore * 100)}% similar, needs ${Math.floor(this.fuzzyThreshold * 100)}%)\n\nDebug Info:\n- Similarity Score: ${Math.floor(bestMatchScore * 100)}%\n- Required Threshold: ${Math.floor(this.fuzzyThreshold * 100)}%\n- Search Range: ${startLine && endLine ? `lines ${startLine}-${endLine}` : "start to end"}\n- Tip: Use read_file to get the latest content of the file before attempting the diff again, as the file content may have changed\n\nSearch Content:\n${searchChunk}${bestMatchSection}${originalContentSection}`, } } diff --git a/src/core/prompts/__tests__/__snapshots__/system.test.ts.snap b/src/core/prompts/__tests__/__snapshots__/system.test.ts.snap index 8801e45..926a032 100644 --- a/src/core/prompts/__tests__/__snapshots__/system.test.ts.snap +++ b/src/core/prompts/__tests__/__snapshots__/system.test.ts.snap @@ -248,11 +248,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -264,7 +263,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -553,11 +551,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -569,7 +566,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -858,11 +854,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -874,7 +869,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -1211,11 +1205,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -1228,7 +1221,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. Then if you want to test your work, you might use browser_action to launch the site, wait for the user's response confirming the site was launched along with a screenshot, then perhaps e.g., click a button to test functionality if needed, wait for the user's response confirming the button was clicked along with a screenshot of the new state, before finally closing the browser. @@ -1930,11 +1922,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -1946,7 +1937,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -2283,11 +2273,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -2300,7 +2289,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. Then if you want to test your work, you might use browser_action to launch the site, wait for the user's response confirming the site was launched along with a screenshot, then perhaps e.g., click a button to test functionality if needed, wait for the user's response confirming the button was clicked along with a screenshot of the new state, before finally closing the browser. @@ -2649,11 +2637,12 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- You should use apply_diff instead of write_to_file when making changes to existing files since it is much faster and easier to apply a diff than to write the entire file again. Only use write_to_file to edit files when apply_diff has failed repeatedly to apply the diff. +- For editing files, you have access to these tools: write_to_file (for creating new files or complete file rewrites), apply_diff (for replacing lines in existing files). +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. +- You should always prefer using other editing tools over write_to_file when making changes to existing files since write_to_file is much slower and cannot handle large files. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -2665,7 +2654,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -2954,11 +2942,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -2970,7 +2957,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -3302,11 +3288,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -3318,7 +3303,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. @@ -3593,11 +3577,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '/test/path', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '/test/path'). For example, if you needed to run \`npm install\` in a project outside of '/test/path', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool. +- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -3609,7 +3592,6 @@ RULES - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc. diff --git a/src/core/prompts/__tests__/system.test.ts b/src/core/prompts/__tests__/system.test.ts index 4138d11..276a06e 100644 --- a/src/core/prompts/__tests__/system.test.ts +++ b/src/core/prompts/__tests__/system.test.ts @@ -11,6 +11,7 @@ import { defaultModeSlug, modes } from "../../../shared/modes" import "../../../utils/path" import { addCustomInstructions } from "../sections/custom-instructions" import * as modesSection from "../sections/modes" +import { EXPERIMENT_IDS } from "../../../shared/experiments" // Mock the sections jest.mock("../sections/modes", () => ({ @@ -121,6 +122,7 @@ const createMockMcpHub = (): McpHub => describe("SYSTEM_PROMPT", () => { let mockMcpHub: McpHub + let experiments: Record beforeAll(() => { // Ensure fs mock is properly initialized @@ -140,6 +142,10 @@ describe("SYSTEM_PROMPT", () => { "/mock/mcp/path", ] dirs.forEach((dir) => mockFs._mockDirectories.add(dir)) + experiments = { + [EXPERIMENT_IDS.SEARCH_AND_REPLACE]: false, + [EXPERIMENT_IDS.INSERT_BLOCK]: false, + } }) beforeEach(() => { @@ -164,6 +170,10 @@ describe("SYSTEM_PROMPT", () => { defaultModeSlug, // mode undefined, // customModePrompts undefined, // customModes + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, ) expect(prompt).toMatchSnapshot() @@ -179,7 +189,11 @@ describe("SYSTEM_PROMPT", () => { "1280x800", // browserViewportSize defaultModeSlug, // mode undefined, // customModePrompts - undefined, // customModes + undefined, // customModes, + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, ) expect(prompt).toMatchSnapshot() @@ -197,7 +211,11 @@ describe("SYSTEM_PROMPT", () => { undefined, // browserViewportSize defaultModeSlug, // mode undefined, // customModePrompts - undefined, // customModes + undefined, // customModes, + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, ) expect(prompt).toMatchSnapshot() @@ -213,7 +231,11 @@ describe("SYSTEM_PROMPT", () => { undefined, // browserViewportSize defaultModeSlug, // mode undefined, // customModePrompts - undefined, // customModes + undefined, // customModes, + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, ) expect(prompt).toMatchSnapshot() @@ -229,7 +251,11 @@ describe("SYSTEM_PROMPT", () => { "900x600", // different viewport size defaultModeSlug, // mode undefined, // customModePrompts - undefined, // customModes + undefined, // customModes, + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, ) expect(prompt).toMatchSnapshot() @@ -249,6 +275,7 @@ describe("SYSTEM_PROMPT", () => { undefined, // globalCustomInstructions undefined, // preferredLanguage true, // diffEnabled + experiments, ) expect(prompt).toContain("apply_diff") @@ -269,6 +296,7 @@ describe("SYSTEM_PROMPT", () => { undefined, // globalCustomInstructions undefined, // preferredLanguage false, // diffEnabled + experiments, ) expect(prompt).not.toContain("apply_diff") @@ -289,6 +317,7 @@ describe("SYSTEM_PROMPT", () => { undefined, // globalCustomInstructions undefined, // preferredLanguage undefined, // diffEnabled + experiments, ) expect(prompt).not.toContain("apply_diff") @@ -308,6 +337,8 @@ describe("SYSTEM_PROMPT", () => { undefined, // customModes undefined, // globalCustomInstructions "Spanish", // preferredLanguage + undefined, // diffEnabled + experiments, ) expect(prompt).toContain("Language Preference:") @@ -337,6 +368,9 @@ describe("SYSTEM_PROMPT", () => { undefined, // customModePrompts customModes, // customModes "Global instructions", // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, ) // Role definition should be at the top @@ -368,6 +402,10 @@ describe("SYSTEM_PROMPT", () => { defaultModeSlug, customModePrompts, undefined, + undefined, + undefined, + undefined, + experiments, ) // Role definition from promptComponent should be at the top @@ -394,18 +432,160 @@ describe("SYSTEM_PROMPT", () => { defaultModeSlug, customModePrompts, undefined, + undefined, + undefined, + undefined, + experiments, ) // Should use the default mode's role definition expect(prompt.indexOf(modes[0].roleDefinition)).toBeLessThan(prompt.indexOf("TOOL USE")) }) + describe("experimental tools", () => { + it("should disable experimental tools by default", async () => { + const prompt = await SYSTEM_PROMPT( + mockContext, + "/test/path", + false, // supportsComputerUse + undefined, // mcpHub + undefined, // diffStrategy + undefined, // browserViewportSize + defaultModeSlug, // mode + undefined, // customModePrompts + undefined, // customModes + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, // experiments - undefined should disable all experimental tools + ) + + // Verify experimental tools are not included in the prompt + expect(prompt).not.toContain(EXPERIMENT_IDS.SEARCH_AND_REPLACE) + expect(prompt).not.toContain(EXPERIMENT_IDS.INSERT_BLOCK) + }) + + it("should enable experimental tools when explicitly enabled", async () => { + const experiments = { + [EXPERIMENT_IDS.SEARCH_AND_REPLACE]: true, + [EXPERIMENT_IDS.INSERT_BLOCK]: true, + } + + const prompt = await SYSTEM_PROMPT( + mockContext, + "/test/path", + false, // supportsComputerUse + undefined, // mcpHub + undefined, // diffStrategy + undefined, // browserViewportSize + defaultModeSlug, // mode + undefined, // customModePrompts + undefined, // customModes + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, + ) + + // Verify experimental tools are included in the prompt when enabled + expect(prompt).toContain(EXPERIMENT_IDS.SEARCH_AND_REPLACE) + expect(prompt).toContain(EXPERIMENT_IDS.INSERT_BLOCK) + }) + + it("should selectively enable experimental tools", async () => { + const experiments = { + [EXPERIMENT_IDS.SEARCH_AND_REPLACE]: true, + [EXPERIMENT_IDS.INSERT_BLOCK]: false, + } + + const prompt = await SYSTEM_PROMPT( + mockContext, + "/test/path", + false, // supportsComputerUse + undefined, // mcpHub + undefined, // diffStrategy + undefined, // browserViewportSize + defaultModeSlug, // mode + undefined, // customModePrompts + undefined, // customModes + undefined, // globalCustomInstructions + undefined, // preferredLanguage + undefined, // diffEnabled + experiments, + ) + + // Verify only enabled experimental tools are included + expect(prompt).toContain(EXPERIMENT_IDS.SEARCH_AND_REPLACE) + expect(prompt).not.toContain(EXPERIMENT_IDS.INSERT_BLOCK) + }) + + it("should list all available editing tools in base instruction", async () => { + const experiments = { + [EXPERIMENT_IDS.SEARCH_AND_REPLACE]: true, + [EXPERIMENT_IDS.INSERT_BLOCK]: true, + } + + const prompt = await SYSTEM_PROMPT( + mockContext, + "/test/path", + false, + undefined, + new SearchReplaceDiffStrategy(), + undefined, + defaultModeSlug, + undefined, + undefined, + undefined, + undefined, + true, // diffEnabled + experiments, + ) + + // Verify base instruction lists all available tools + expect(prompt).toContain("apply_diff (for replacing lines in existing files)") + expect(prompt).toContain("write_to_file (for creating new files or complete file rewrites)") + expect(prompt).toContain("insert_code_block (for adding sections to existing files)") + expect(prompt).toContain("search_and_replace (for finding and replacing individual pieces of text)") + }) + + it("should provide detailed instructions for each enabled tool", async () => { + const experiments = { + [EXPERIMENT_IDS.SEARCH_AND_REPLACE]: true, + [EXPERIMENT_IDS.INSERT_BLOCK]: true, + } + + const prompt = await SYSTEM_PROMPT( + mockContext, + "/test/path", + false, + undefined, + new SearchReplaceDiffStrategy(), + undefined, + defaultModeSlug, + undefined, + undefined, + undefined, + undefined, + true, + experiments, + ) + + // Verify detailed instructions for each tool + expect(prompt).toContain( + "You should always prefer using other editing tools over write_to_file when making changes to existing files since write_to_file is much slower and cannot handle large files.", + ) + expect(prompt).toContain("The insert_code_block tool adds code snippets or content blocks to files") + expect(prompt).toContain("The search_and_replace tool finds and replaces text or regex in files") + }) + }) + afterAll(() => { jest.restoreAllMocks() }) }) describe("addCustomInstructions", () => { + let experiments: Record beforeAll(() => { // Ensure fs mock is properly initialized const mockFs = jest.requireMock("fs/promises") @@ -417,6 +597,11 @@ describe("addCustomInstructions", () => { } throw new Error(`ENOENT: no such file or directory, mkdir '${path}'`) }) + + experiments = { + [EXPERIMENT_IDS.SEARCH_AND_REPLACE]: false, + [EXPERIMENT_IDS.INSERT_BLOCK]: false, + } }) beforeEach(() => { @@ -434,6 +619,10 @@ describe("addCustomInstructions", () => { "architect", // mode undefined, // customModePrompts undefined, // customModes + undefined, + undefined, + undefined, + experiments, ) expect(prompt).toMatchSnapshot() @@ -450,6 +639,10 @@ describe("addCustomInstructions", () => { "ask", // mode undefined, // customModePrompts undefined, // customModes + undefined, + undefined, + undefined, + experiments, ) expect(prompt).toMatchSnapshot() diff --git a/src/core/prompts/sections/rules.ts b/src/core/prompts/sections/rules.ts index cfb8a24..cb944a1 100644 --- a/src/core/prompts/sections/rules.ts +++ b/src/core/prompts/sections/rules.ts @@ -3,14 +3,58 @@ import { modes, ModeConfig } from "../../../shared/modes" import * as vscode from "vscode" import * as path from "path" +function getEditingInstructions(diffStrategy?: DiffStrategy, experiments?: Record): string { + const instructions: string[] = [] + const availableTools: string[] = ["write_to_file (for creating new files or complete file rewrites)"] + + // Collect available editing tools + if (diffStrategy) { + availableTools.push("apply_diff (for replacing lines in existing files)") + } + if (experiments?.["insert_code_block"]) { + availableTools.push("insert_code_block (for adding sections to existing files)") + } + if (experiments?.["search_and_replace"]) { + availableTools.push("search_and_replace (for finding and replacing individual pieces of text)") + } + + // Base editing instruction mentioning all available tools + if (availableTools.length > 1) { + instructions.push(`- For editing files, you have access to these tools: ${availableTools.join(", ")}.`) + } + + // Additional details for experimental features + if (experiments?.["insert_code_block"]) { + instructions.push( + "- The insert_code_block tool adds code snippets or content blocks to files, such as adding a new function to a JavaScript file or inserting a new route in a Python file. This tool will insert it at the specified line location. It can support multiple operations at once.", + ) + } + + if (experiments?.["search_and_replace"]) { + instructions.push( + "- The search_and_replace tool finds and replaces text or regex in files. This tool allows you to search for a specific regex pattern or text and replace it with another value. Be cautious when using this tool to ensure you are replacing the correct text. It can support multiple operations at once.", + ) + } + + instructions.push( + "- When using the write_to_file tool to modify a file, use the tool directly with the desired content. You do not need to display the content before using the tool. ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project.", + ) + + if (availableTools.length > 1) { + instructions.push( + "- You should always prefer using other editing tools over write_to_file when making changes to existing files since write_to_file is much slower and cannot handle large files.", + ) + } + + return instructions.join("\n") +} + export function getRulesSection( cwd: string, supportsComputerUse: boolean, diffStrategy?: DiffStrategy, - context?: vscode.ExtensionContext, + experiments?: Record | undefined, ): string { - const settingsDir = context ? path.join(context.globalStorageUri.fsPath, "settings") : "" - const customModesPath = path.join(settingsDir, "cline_custom_modes.json") return `==== RULES @@ -21,15 +65,10 @@ RULES - Before using the execute_command tool, you must first think about the SYSTEM INFORMATION context provided to understand the user's environment and tailor your commands to ensure they are compatible with their system. You must also consider if the command you need to run should be executed in a specific directory outside of the current working directory '${cwd.toPosix()}', and if so prepend with \`cd\`'ing into that directory && then executing the command (as one command since you are stuck operating from '${cwd.toPosix()}'). For example, if you needed to run \`npm install\` in a project outside of '${cwd.toPosix()}', you would need to prepend with a \`cd\` i.e. pseudocode for this would be \`cd (path to project) && (command, in this case npm install)\`. - When using the search_files tool, craft your regex patterns carefully to balance specificity and flexibility. Based on the user's task you may use it to find code patterns, TODO comments, function definitions, or any text-based information across the project. The results include context, so analyze the surrounding code to better understand the matches. Leverage the search_files tool in combination with other tools for more comprehensive analysis. For example, use it to find specific code patterns, then use read_file to examine the full context of interesting matches before using write_to_file to make informed changes. - When creating a new project (such as an app, website, or any software project), organize all new files within a dedicated project directory unless the user specifies otherwise. Use appropriate file paths when writing files, as the write_to_file tool will automatically create any necessary directories. Structure the project logically, adhering to best practices for the specific type of project being created. Unless otherwise specified, new projects should be easily run without additional setup, for example most projects can be built in HTML, CSS, and JavaScript - which you can open in a browser. -${ - diffStrategy - ? "- You should use apply_diff instead of write_to_file when making changes to existing files since it is much faster and easier to apply a diff than to write the entire file again. Only use write_to_file to edit files when apply_diff has failed repeatedly to apply the diff." - : "- When you want to modify a file, use the write_to_file tool directly with the desired content. You do not need to display the content before using the tool." -} +${getEditingInstructions(diffStrategy, experiments)} - Some modes have restrictions on which files they can edit. If you attempt to edit a restricted file, the operation will be rejected with a FileRestrictionError that will specify which file patterns are allowed for the current mode. - Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. * For example, in architect mode trying to edit app.js would be rejected because architect mode can only edit files matching "\\.md$" -- Be sure to consider the type of project (e.g. Python, JavaScript, web application) when determining the appropriate structure and files to include. Also consider what files may be most relevant to accomplishing the task, for example looking at a project's manifest file would help you understand the project's dependencies, which you could incorporate into any code you write. - When making changes to code, always consider the context in which the code is being used. Ensure that your changes are compatible with the existing codebase and that they follow the project's coding standards and best practices. - 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. @@ -45,7 +84,6 @@ ${ - When presented with images, utilize your vision capabilities to thoroughly examine them and extract meaningful information. Incorporate these insights into your thought process as you accomplish the user's task. - At the end of each user message, you will automatically receive environment_details. This information is not written by the user themselves, but is auto-generated to provide potentially relevant context about the project structure and environment. While this information can be valuable for understanding the project context, do not treat it as a direct part of the user's request or response. Use it to inform your actions and decisions, but don't assume the user is explicitly asking about or referring to this information unless they clearly do so in their message. When using environment_details, explain your actions clearly to ensure the user understands, as they may not be aware of these details. - Before executing commands, check the "Actively Running Terminals" section in environment_details. If present, consider how these active processes might impact your task. For example, if a local development server is already running, you wouldn't need to start it again. If no active terminals are listed, proceed with command execution as normal. -- When using the write_to_file tool, ALWAYS provide the COMPLETE file content in your response. This is NON-NEGOTIABLE. Partial updates or placeholders like '// rest of code unchanged' are STRICTLY FORBIDDEN. You MUST include ALL parts of the file, even if they haven't been modified. Failure to do so will result in incomplete or broken code, severely impacting the user's project. - MCP operations should be used one at a time, similar to other tool usage. Wait for confirmation of success before proceeding with additional operations. - It is critical you wait for the user's response after each tool use, in order to confirm the success of the tool use. For example, if asked to make a todo app, you would create a file, wait for the user's response it was created successfully, then create another file if needed, wait for the user's response it was created successfully, etc.${ supportsComputerUse diff --git a/src/core/prompts/system.ts b/src/core/prompts/system.ts index 51719c3..f579771 100644 --- a/src/core/prompts/system.ts +++ b/src/core/prompts/system.ts @@ -39,6 +39,7 @@ async function generatePrompt( globalCustomInstructions?: string, preferredLanguage?: string, diffEnabled?: boolean, + experiments?: Record, ): Promise { if (!context) { throw new Error("Extension context is required for generating system prompt") @@ -68,6 +69,7 @@ ${getToolDescriptionsForMode( browserViewportSize, mcpHub, customModeConfigs, + experiments, )} ${getToolUseGuidelinesSection()} @@ -78,7 +80,7 @@ ${getCapabilitiesSection(cwd, supportsComputerUse, mcpHub, effectiveDiffStrategy ${modesSection} -${getRulesSection(cwd, supportsComputerUse, effectiveDiffStrategy, context)} +${getRulesSection(cwd, supportsComputerUse, effectiveDiffStrategy, experiments)} ${getSystemInfoSection(cwd, mode, customModeConfigs)} @@ -102,6 +104,7 @@ export const SYSTEM_PROMPT = async ( globalCustomInstructions?: string, preferredLanguage?: string, diffEnabled?: boolean, + experiments?: Record, ): Promise => { if (!context) { throw new Error("Extension context is required for generating system prompt") @@ -135,5 +138,6 @@ export const SYSTEM_PROMPT = async ( globalCustomInstructions, preferredLanguage, diffEnabled, + experiments, ) } diff --git a/src/core/prompts/tools/index.ts b/src/core/prompts/tools/index.ts index b704f4f..23e77fb 100644 --- a/src/core/prompts/tools/index.ts +++ b/src/core/prompts/tools/index.ts @@ -3,6 +3,8 @@ import { getReadFileDescription } from "./read-file" import { getWriteToFileDescription } from "./write-to-file" import { getSearchFilesDescription } from "./search-files" import { getListFilesDescription } from "./list-files" +import { getInsertCodeBlockDescription } from "./insert-code-block" +import { getSearchAndReplaceDescription } from "./search-and-replace" import { getListCodeDefinitionNamesDescription } from "./list-code-definition-names" import { getBrowserActionDescription } from "./browser-action" import { getAskFollowupQuestionDescription } from "./ask-followup-question" @@ -30,6 +32,8 @@ const toolDescriptionMap: Record string | undefined> use_mcp_tool: (args) => getUseMcpToolDescription(args), access_mcp_resource: (args) => getAccessMcpResourceDescription(args), switch_mode: () => getSwitchModeDescription(), + insert_code_block: (args) => getInsertCodeBlockDescription(args), + search_and_replace: (args) => getSearchAndReplaceDescription(args), apply_diff: (args) => args.diffStrategy ? args.diffStrategy.getToolDescription({ cwd: args.cwd, toolOptions: args.toolOptions }) : "", } @@ -42,6 +46,7 @@ export function getToolDescriptionsForMode( browserViewportSize?: string, mcpHub?: McpHub, customModes?: ModeConfig[], + experiments?: Record, ): string { const config = getModeConfig(mode, customModes) const args: ToolArgs = { @@ -60,7 +65,7 @@ export function getToolDescriptionsForMode( const toolGroup = TOOL_GROUPS[groupName] if (toolGroup) { toolGroup.forEach((tool) => { - if (isToolAllowedForMode(tool as ToolName, mode, customModes ?? [])) { + if (isToolAllowedForMode(tool as ToolName, mode, customModes ?? [], experiments ?? {})) { tools.add(tool) } }) @@ -100,4 +105,6 @@ export { getUseMcpToolDescription, getAccessMcpResourceDescription, getSwitchModeDescription, + getInsertCodeBlockDescription, + getSearchAndReplaceDescription, } diff --git a/src/core/prompts/tools/insert-code-block.ts b/src/core/prompts/tools/insert-code-block.ts new file mode 100644 index 0000000..7dce56e --- /dev/null +++ b/src/core/prompts/tools/insert-code-block.ts @@ -0,0 +1,35 @@ +import { ToolArgs } from "./types" + +export function getInsertCodeBlockDescription(args: ToolArgs): string { + return `## insert_code_block +Description: Inserts code blocks at specific line positions in a file. This is the primary tool for adding new code (functions/methods/classes, imports, attributes etc.) as it allows for precise insertions without overwriting existing content. The tool uses an efficient line-based insertion system that maintains file integrity and proper ordering of multiple insertions. Beware to use the proper indentation. This tool is the preferred way to add new code to files. +Parameters: +- path: (required) The path of the file to insert code into (relative to the current working directory ${args.cwd.toPosix()}) +- operations: (required) A JSON array of insertion operations. Each operation is an object with: + * start_line: (required) The line number where the code block should be inserted. The content currently at that line will end up below the inserted code block. + * content: (required) The code block to insert at the specified position. IMPORTANT NOTE: If the content is a single line, it can be a string. If it's a multi-line content, it should be a string with newline characters (\n) for line breaks. +Usage: + +File path here +[ + { + "start_line": 10, + "content": "Your code block here" + } +] + +Example: Insert a new function and its import statement + +src/app.ts +[ + { + "start_line": 1, + "content": "import { sum } from './utils';" + }, + { + "start_line": 10, + "content": "function calculateTotal(items: number[]): number {\n return items.reduce((sum, item) => sum + item, 0);\n}" + } +] +` +} diff --git a/src/core/prompts/tools/search-and-replace.ts b/src/core/prompts/tools/search-and-replace.ts new file mode 100644 index 0000000..eb48d85 --- /dev/null +++ b/src/core/prompts/tools/search-and-replace.ts @@ -0,0 +1,52 @@ +import { ToolArgs } from "./types" + +export function getSearchAndReplaceDescription(args: ToolArgs): string { + return `## search_and_replace +Description: Request to perform search and replace operations on a file. Each operation can specify a search pattern (string or regex) and replacement text, with optional line range restrictions and regex flags. Shows a diff preview before applying changes. +Parameters: +- path: (required) The path of the file to modify (relative to the current working directory ${args.cwd.toPosix()}) +- operations: (required) A JSON array of search/replace operations. Each operation is an object with: + * search: (required) The text or pattern to search for + * replace: (required) The text to replace matches with. If multiple lines need to be replaced, use "\n" for newlines + * start_line: (optional) Starting line number for restricted replacement + * end_line: (optional) Ending line number for restricted replacement + * use_regex: (optional) Whether to treat search as a regex pattern + * ignore_case: (optional) Whether to ignore case when matching + * regex_flags: (optional) Additional regex flags when use_regex is true +Usage: + +File path here +[ + { + "search": "text to find", + "replace": "replacement text", + "start_line": 1, + "end_line": 10 + } +] + +Example: Replace "foo" with "bar" in lines 1-10 of example.ts + +example.ts +[ + { + "search": "foo", + "replace": "bar", + "start_line": 1, + "end_line": 10 + } +] + +Example: Replace all occurrences of "old" with "new" using regex + +example.ts +[ + { + "search": "old\\w+", + "replace": "new$&", + "use_regex": true, + "ignore_case": true + } +] +` +} diff --git a/src/core/webview/ClineProvider.ts b/src/core/webview/ClineProvider.ts index 0f96df2..d2db67f 100644 --- a/src/core/webview/ClineProvider.ts +++ b/src/core/webview/ClineProvider.ts @@ -40,6 +40,7 @@ import { singleCompletionHandler } from "../../utils/single-completion-handler" import { getCommitInfo, searchCommits, getWorkingState } from "../../utils/git" import { ConfigManager } from "../config/ConfigManager" import { CustomModesManager } from "../config/CustomModesManager" +import { EXPERIMENT_IDS, experiments as Experiments, experimentDefault, ExperimentId } from "../../shared/experiments" import { CustomSupportPrompts, supportPrompt } from "../../shared/support-prompt" import { ACTION_NAMES } from "../CodeActionProvider" @@ -118,7 +119,7 @@ type GlobalStateKey = | "customModePrompts" | "customSupportPrompts" | "enhancementApiConfigId" - | "experimentalDiffStrategy" + | "experiments" // Map of experiment IDs to their enabled state | "autoApprovalEnabled" | "customModes" // Array of custom modes @@ -339,7 +340,7 @@ export class ClineProvider implements vscode.WebviewViewProvider { fuzzyMatchThreshold, mode, customInstructions: globalInstructions, - experimentalDiffStrategy, + experiments, } = await this.getState() const modePrompt = customModePrompts?.[mode] as PromptComponent @@ -354,7 +355,7 @@ export class ClineProvider implements vscode.WebviewViewProvider { task, images, undefined, - experimentalDiffStrategy, + experiments, ) } @@ -367,7 +368,7 @@ export class ClineProvider implements vscode.WebviewViewProvider { fuzzyMatchThreshold, mode, customInstructions: globalInstructions, - experimentalDiffStrategy, + experiments, } = await this.getState() const modePrompt = customModePrompts?.[mode] as PromptComponent @@ -382,7 +383,7 @@ export class ClineProvider implements vscode.WebviewViewProvider { undefined, undefined, historyItem, - experimentalDiffStrategy, + experiments, ) } @@ -1044,14 +1045,14 @@ export class ClineProvider implements vscode.WebviewViewProvider { diffEnabled, mcpEnabled, fuzzyMatchThreshold, - experimentalDiffStrategy, + experiments, } = await this.getState() // Create diffStrategy based on current model and settings const diffStrategy = getDiffStrategy( apiConfiguration.apiModelId || apiConfiguration.openRouterModelId || "", fuzzyMatchThreshold, - experimentalDiffStrategy, + Experiments.isEnabled(experiments, EXPERIMENT_IDS.DIFF_STRATEGY), ) const cwd = vscode.workspace.workspaceFolders?.map((folder) => folder.uri.fsPath).at(0) || "" @@ -1072,6 +1073,7 @@ export class ClineProvider implements vscode.WebviewViewProvider { customInstructions, preferredLanguage, diffEnabled, + experiments, ) await this.postMessageToWebview({ @@ -1207,14 +1209,28 @@ export class ClineProvider implements vscode.WebviewViewProvider { vscode.window.showErrorMessage("Failed to get list api configuration") } break - case "experimentalDiffStrategy": - await this.updateGlobalState("experimentalDiffStrategy", message.bool ?? false) - // Update diffStrategy in current Cline instance if it exists - if (this.cline) { - await this.cline.updateDiffStrategy(message.bool ?? false) + case "updateExperimental": { + if (!message.values) { + break } + + const updatedExperiments = { + ...((await this.getGlobalState("experiments")) ?? experimentDefault), + ...message.values, + } as Record + + await this.updateGlobalState("experiments", updatedExperiments) + + // Update diffStrategy in current Cline instance if it exists + if (message.values[EXPERIMENT_IDS.DIFF_STRATEGY] !== undefined && this.cline) { + await this.cline.updateDiffStrategy( + Experiments.isEnabled(updatedExperiments, EXPERIMENT_IDS.DIFF_STRATEGY), + ) + } + await this.postStateToWebview() break + } case "updateMcpTimeout": if (message.serverName && typeof message.timeout === "number") { try { @@ -1873,8 +1889,8 @@ export class ClineProvider implements vscode.WebviewViewProvider { customModePrompts, customSupportPrompts, enhancementApiConfigId, - experimentalDiffStrategy, autoApprovalEnabled, + experiments, } = await this.getState() const allowedCommands = vscode.workspace.getConfiguration("roo-cline").get("allowedCommands") || [] @@ -1914,9 +1930,9 @@ export class ClineProvider implements vscode.WebviewViewProvider { customModePrompts: customModePrompts ?? {}, customSupportPrompts: customSupportPrompts ?? {}, enhancementApiConfigId, - experimentalDiffStrategy: experimentalDiffStrategy ?? false, autoApprovalEnabled: autoApprovalEnabled ?? false, customModes: await this.customModesManager.getCustomModes(), + experiments: experiments ?? experimentDefault, } } @@ -2039,9 +2055,9 @@ export class ClineProvider implements vscode.WebviewViewProvider { customModePrompts, customSupportPrompts, enhancementApiConfigId, - experimentalDiffStrategy, autoApprovalEnabled, customModes, + experiments, ] = await Promise.all([ this.getGlobalState("apiProvider") as Promise, this.getGlobalState("apiModelId") as Promise, @@ -2109,9 +2125,9 @@ export class ClineProvider implements vscode.WebviewViewProvider { this.getGlobalState("customModePrompts") as Promise, this.getGlobalState("customSupportPrompts") as Promise, this.getGlobalState("enhancementApiConfigId") as Promise, - this.getGlobalState("experimentalDiffStrategy") as Promise, this.getGlobalState("autoApprovalEnabled") as Promise, this.customModesManager.getCustomModes(), + this.getGlobalState("experiments") as Promise | undefined>, ]) let apiProvider: ApiProvider @@ -2225,7 +2241,7 @@ export class ClineProvider implements vscode.WebviewViewProvider { customModePrompts: customModePrompts ?? {}, customSupportPrompts: customSupportPrompts ?? {}, enhancementApiConfigId, - experimentalDiffStrategy: experimentalDiffStrategy ?? false, + experiments: experiments ?? experimentDefault, autoApprovalEnabled: autoApprovalEnabled ?? false, customModes, } diff --git a/src/core/webview/__tests__/ClineProvider.test.ts b/src/core/webview/__tests__/ClineProvider.test.ts index 09b3db3..0fb7012 100644 --- a/src/core/webview/__tests__/ClineProvider.test.ts +++ b/src/core/webview/__tests__/ClineProvider.test.ts @@ -4,6 +4,7 @@ import { ExtensionMessage, ExtensionState } from "../../../shared/ExtensionMessa import { setSoundEnabled } from "../../../utils/sound" import { defaultModeSlug, modes } from "../../../shared/modes" import { addCustomInstructions } from "../../prompts/sections/custom-instructions" +import { experimentDefault, experiments } from "../../../shared/experiments" // Mock custom-instructions module const mockAddCustomInstructions = jest.fn() @@ -320,6 +321,7 @@ describe("ClineProvider", () => { requestDelaySeconds: 5, mode: defaultModeSlug, customModes: [], + experiments: experimentDefault, } const message: ExtensionMessage = { @@ -617,6 +619,7 @@ describe("ClineProvider", () => { mode: "code", diffEnabled: true, fuzzyMatchThreshold: 1.0, + experiments: experimentDefault, } as any) // Reset Cline mock @@ -636,7 +639,7 @@ describe("ClineProvider", () => { "Test task", undefined, undefined, - undefined, + experimentDefault, ) }) test("handles mode-specific custom instructions updates", async () => { @@ -887,6 +890,7 @@ describe("ClineProvider", () => { }, mcpEnabled: true, mode: "code" as const, + experiments: experimentDefault, } as any) const handler1 = getMessageHandler() @@ -918,6 +922,7 @@ describe("ClineProvider", () => { }, mcpEnabled: false, mode: "code" as const, + experiments: experimentDefault, } as any) const handler2 = getMessageHandler() @@ -985,6 +990,7 @@ describe("ClineProvider", () => { experimentalDiffStrategy: true, diffEnabled: true, fuzzyMatchThreshold: 0.8, + experiments: experimentDefault, } as any) // Mock SYSTEM_PROMPT to verify diffStrategy and diffEnabled are passed @@ -1012,6 +1018,7 @@ describe("ClineProvider", () => { undefined, // effectiveInstructions undefined, // preferredLanguage true, // diffEnabled + experimentDefault, ) // Run the test again to verify it's consistent @@ -1034,6 +1041,7 @@ describe("ClineProvider", () => { experimentalDiffStrategy: true, diffEnabled: false, fuzzyMatchThreshold: 0.8, + experiments: experimentDefault, } as any) // Mock SYSTEM_PROMPT to verify diffEnabled is passed as false @@ -1061,6 +1069,7 @@ describe("ClineProvider", () => { undefined, // effectiveInstructions undefined, // preferredLanguage false, // diffEnabled + experimentDefault, ) }) @@ -1077,6 +1086,7 @@ describe("ClineProvider", () => { mode: "architect", mcpEnabled: false, browserViewportSize: "900x600", + experiments: experimentDefault, } as any) // Mock SYSTEM_PROMPT to call addCustomInstructions diff --git a/src/shared/ExtensionMessage.ts b/src/shared/ExtensionMessage.ts index 56075b8..4cd6370 100644 --- a/src/shared/ExtensionMessage.ts +++ b/src/shared/ExtensionMessage.ts @@ -6,6 +6,7 @@ import { McpServer } from "./mcp" import { GitCommit } from "../utils/git" import { Mode, CustomModePrompts, ModeConfig } from "./modes" import { CustomSupportPrompts } from "./support-prompt" +import { ExperimentId } from "./experiments" export interface LanguageModelChatSelector { vendor?: string @@ -108,7 +109,7 @@ export interface ExtensionState { mode: Mode modeApiConfigs?: Record enhancementApiConfigId?: string - experimentalDiffStrategy?: boolean + experiments: Record // Map of experiment IDs to their enabled state autoApprovalEnabled?: boolean customModes: ModeConfig[] toolRequirements?: Record // Map of tool names to their requirements (e.g. {"apply_diff": true} if diffEnabled) diff --git a/src/shared/WebviewMessage.ts b/src/shared/WebviewMessage.ts index a027a93..357a237 100644 --- a/src/shared/WebviewMessage.ts +++ b/src/shared/WebviewMessage.ts @@ -75,7 +75,7 @@ export interface WebviewMessage { | "getSystemPrompt" | "systemPrompt" | "enhancementApiConfigId" - | "experimentalDiffStrategy" + | "updateExperimental" | "autoApprovalEnabled" | "updateCustomMode" | "deleteCustomMode" diff --git a/src/shared/__tests__/modes.test.ts b/src/shared/__tests__/modes.test.ts index 252ed6e..b9e46cb 100644 --- a/src/shared/__tests__/modes.test.ts +++ b/src/shared/__tests__/modes.test.ts @@ -14,6 +14,12 @@ describe("isToolAllowedForMode", () => { roleDefinition: "You are a CSS editor", groups: ["read", ["edit", { fileRegex: "\\.css$" }], "browser"], }, + { + slug: "test-exp-mode", + name: "Test Exp Mode", + roleDefinition: "You are an experimental tester", + groups: ["read", "edit", "browser"], + }, ] it("allows always available tools", () => { @@ -240,6 +246,87 @@ describe("isToolAllowedForMode", () => { expect(isToolAllowedForMode("write_to_file", "markdown-editor", customModes, toolRequirements)).toBe(false) }) + + describe("experimental tools", () => { + it("disables tools when experiment is disabled", () => { + const experiments = { + search_and_replace: false, + insert_code_block: false, + } + + expect( + isToolAllowedForMode( + "search_and_replace", + "test-exp-mode", + customModes, + undefined, + undefined, + experiments, + ), + ).toBe(false) + + expect( + isToolAllowedForMode( + "insert_code_block", + "test-exp-mode", + customModes, + undefined, + undefined, + experiments, + ), + ).toBe(false) + }) + + it("allows tools when experiment is enabled", () => { + const experiments = { + search_and_replace: true, + insert_code_block: true, + } + + expect( + isToolAllowedForMode( + "search_and_replace", + "test-exp-mode", + customModes, + undefined, + undefined, + experiments, + ), + ).toBe(true) + + expect( + isToolAllowedForMode( + "insert_code_block", + "test-exp-mode", + customModes, + undefined, + undefined, + experiments, + ), + ).toBe(true) + }) + + it("allows non-experimental tools when experiments are disabled", () => { + const experiments = { + search_and_replace: false, + insert_code_block: false, + } + + expect( + isToolAllowedForMode("read_file", "markdown-editor", customModes, undefined, undefined, experiments), + ).toBe(true) + expect( + isToolAllowedForMode( + "write_to_file", + "markdown-editor", + customModes, + undefined, + { path: "test.md" }, + experiments, + ), + ).toBe(true) + }) + }) }) describe("FileRestrictionError", () => { diff --git a/src/shared/experiments.ts b/src/shared/experiments.ts new file mode 100644 index 0000000..71eda74 --- /dev/null +++ b/src/shared/experiments.ts @@ -0,0 +1,69 @@ +export const EXPERIMENT_IDS = { + DIFF_STRATEGY: "experimentalDiffStrategy", + SEARCH_AND_REPLACE: "search_and_replace", + INSERT_BLOCK: "insert_code_block", +} as const + +export type ExperimentKey = keyof typeof EXPERIMENT_IDS +export type ExperimentId = valueof + +export interface ExperimentConfig { + name: string + description: string + enabled: boolean +} + +type valueof = X[keyof X] + +export const experimentConfigsMap: Record = { + DIFF_STRATEGY: { + name: "Use experimental unified diff strategy", + description: + "Enable the experimental unified diff strategy. This strategy might reduce the number of retries caused by model errors but may cause unexpected behavior or incorrect edits. Only enable if you understand the risks and are willing to carefully review all changes.", + enabled: false, + }, + SEARCH_AND_REPLACE: { + name: "Use experimental search and replace tool", + description: + "Enable the experimental search and replace tool, allowing Roo to replace multiple instances of a search term in one request.", + enabled: false, + }, + INSERT_BLOCK: { + name: "Use experimental insert block tool", + + description: + "Enable the experimental insert block tool, allowing Roo to insert multiple code blocks at once at specific line numbers without needing to create a diff.", + enabled: false, + }, +} + +export const experimentDefault = Object.fromEntries( + Object.entries(experimentConfigsMap).map(([_, config]) => [ + EXPERIMENT_IDS[_ as keyof typeof EXPERIMENT_IDS] as ExperimentId, + config.enabled, + ]), +) as Record + +export const experiments = { + get: (id: ExperimentKey): ExperimentConfig | undefined => { + return experimentConfigsMap[id] + }, + isEnabled: (experimentsConfig: Record, id: ExperimentId): boolean => { + return experimentsConfig[id] ?? experimentDefault[id] + }, +} as const + +// Expose experiment details for UI - pre-compute from map for better performance +export const experimentLabels = Object.fromEntries( + Object.entries(experimentConfigsMap).map(([_, config]) => [ + EXPERIMENT_IDS[_ as keyof typeof EXPERIMENT_IDS] as ExperimentId, + config.name, + ]), +) as Record + +export const experimentDescriptions = Object.fromEntries( + Object.entries(experimentConfigsMap).map(([_, config]) => [ + EXPERIMENT_IDS[_ as keyof typeof EXPERIMENT_IDS] as ExperimentId, + config.description, + ]), +) as Record diff --git a/src/shared/modes.ts b/src/shared/modes.ts index 4fa9fb0..abbeea7 100644 --- a/src/shared/modes.ts +++ b/src/shared/modes.ts @@ -160,12 +160,19 @@ export function isToolAllowedForMode( customModes: ModeConfig[], toolRequirements?: Record, toolParams?: Record, // All tool parameters + experiments?: Record, ): boolean { // Always allow these tools if (ALWAYS_AVAILABLE_TOOLS.includes(tool as any)) { return true } + if (experiments && tool in experiments) { + if (!experiments[tool]) { + return false + } + } + // Check tool requirements if any exist if (toolRequirements && tool in toolRequirements) { if (!toolRequirements[tool]) { @@ -198,7 +205,7 @@ export function isToolAllowedForMode( const filePath = toolParams?.path if ( filePath && - (toolParams.diff || toolParams.content) && + (toolParams.diff || toolParams.content || toolParams.operations) && !doesFileMatchRegex(filePath, options.fileRegex) ) { throw new FileRestrictionError(mode.name, options.fileRegex, options.description, filePath) diff --git a/src/shared/tool-groups.ts b/src/shared/tool-groups.ts index 96c9f03..d95b16e 100644 --- a/src/shared/tool-groups.ts +++ b/src/shared/tool-groups.ts @@ -21,7 +21,7 @@ export const TOOL_DISPLAY_NAMES = { // Define available tool groups export const TOOL_GROUPS: Record = { read: ["read_file", "search_files", "list_files", "list_code_definition_names"], - edit: ["write_to_file", "apply_diff"], + edit: ["write_to_file", "apply_diff", "insert_code_block", "search_and_replace"], browser: ["browser_action"], command: ["execute_command"], mcp: ["use_mcp_tool", "access_mcp_resource"], diff --git a/webview-ui/src/components/settings/ExperimentalFeature.tsx b/webview-ui/src/components/settings/ExperimentalFeature.tsx new file mode 100644 index 0000000..b896e8e --- /dev/null +++ b/webview-ui/src/components/settings/ExperimentalFeature.tsx @@ -0,0 +1,36 @@ +import { VSCodeCheckbox } from "@vscode/webview-ui-toolkit/react" + +interface ExperimentalFeatureProps { + name: string + description: string + enabled: boolean + onChange: (value: boolean) => void +} + +const ExperimentalFeature = ({ name, description, enabled, onChange }: ExperimentalFeatureProps) => { + return ( +
+
+ ⚠️ + onChange(e.target.checked)}> + {name} + +
+

+ {description} +

+
+ ) +} + +export default ExperimentalFeature diff --git a/webview-ui/src/components/settings/SettingsView.tsx b/webview-ui/src/components/settings/SettingsView.tsx index 12b97f7..64b762a 100644 --- a/webview-ui/src/components/settings/SettingsView.tsx +++ b/webview-ui/src/components/settings/SettingsView.tsx @@ -4,6 +4,8 @@ import { useExtensionState } from "../../context/ExtensionStateContext" import { validateApiConfiguration, validateModelId } from "../../utils/validate" import { vscode } from "../../utils/vscode" import ApiOptions from "./ApiOptions" +import ExperimentalFeature from "./ExperimentalFeature" +import { EXPERIMENT_IDS, experimentConfigsMap } from "../../../../src/shared/experiments" import ApiConfigManager from "./ApiConfigManager" type SettingsViewProps = { @@ -51,8 +53,8 @@ const SettingsView = ({ onDone }: SettingsViewProps) => { setRequestDelaySeconds, currentApiConfigName, listApiConfigMeta, - experimentalDiffStrategy, - setExperimentalDiffStrategy, + experiments, + setExperimentEnabled, alwaysAllowModeSwitch, setAlwaysAllowModeSwitch, } = useExtensionState() @@ -94,7 +96,12 @@ const SettingsView = ({ onDone }: SettingsViewProps) => { text: currentApiConfigName, apiConfiguration, }) - vscode.postMessage({ type: "experimentalDiffStrategy", bool: experimentalDiffStrategy }) + + vscode.postMessage({ + type: "updateExperimental", + values: experiments, + }) + vscode.postMessage({ type: "alwaysAllowModeSwitch", bool: alwaysAllowModeSwitch }) onDone() } @@ -583,7 +590,7 @@ const SettingsView = ({ onDone }: SettingsViewProps) => { setDiffEnabled(e.target.checked) if (!e.target.checked) { // Reset experimental strategy when diffs are disabled - setExperimentalDiffStrategy(false) + setExperimentEnabled(EXPERIMENT_IDS.DIFF_STRATEGY, false) } }}> Enable editing through diffs @@ -599,35 +606,14 @@ const SettingsView = ({ onDone }: SettingsViewProps) => {

{diffEnabled && ( -
-
- ⚠️ - setExperimentalDiffStrategy(e.target.checked)}> - - Use experimental unified diff strategy - - -
-

- Enable the experimental unified diff strategy. This strategy might reduce the number - of retries caused by model errors but may cause unexpected behavior or incorrect - edits. Only enable if you understand the risks and are willing to carefully review - all changes. -

- -
+
+ setExperimentEnabled(EXPERIMENT_IDS.DIFF_STRATEGY, enabled)} + /> +
Match precision {

)} + {Object.entries(experimentConfigsMap) + .filter((config) => config[0] !== "DIFF_STRATEGY") + .map((config) => ( + + setExperimentEnabled( + EXPERIMENT_IDS[config[0] as keyof typeof EXPERIMENT_IDS], + enabled, + ) + } + /> + ))}
diff --git a/webview-ui/src/context/ExtensionStateContext.tsx b/webview-ui/src/context/ExtensionStateContext.tsx index 7d0159d..9738edf 100644 --- a/webview-ui/src/context/ExtensionStateContext.tsx +++ b/webview-ui/src/context/ExtensionStateContext.tsx @@ -16,6 +16,7 @@ import { McpServer } from "../../../src/shared/mcp" import { checkExistKey } from "../../../src/shared/checkExistApiConfig" import { Mode, CustomModePrompts, defaultModeSlug, defaultPrompts, ModeConfig } from "../../../src/shared/modes" import { CustomSupportPrompts } from "../../../src/shared/support-prompt" +import { experimentDefault, ExperimentId } from "../../../src/shared/experiments" export interface ExtensionStateContextType extends ExtensionState { didHydrateState: boolean @@ -63,9 +64,7 @@ export interface ExtensionStateContextType extends ExtensionState { setCustomSupportPrompts: (value: CustomSupportPrompts) => void enhancementApiConfigId?: string setEnhancementApiConfigId: (value: string) => void - experimentalDiffStrategy: boolean - setExperimentalDiffStrategy: (value: boolean) => void - autoApprovalEnabled?: boolean + setExperimentEnabled: (id: ExperimentId, enabled: boolean) => void setAutoApprovalEnabled: (value: boolean) => void handleInputChange: (field: keyof ApiConfiguration) => (event: any) => void customModes: ModeConfig[] @@ -98,8 +97,8 @@ export const ExtensionStateContextProvider: React.FC<{ children: React.ReactNode mode: defaultModeSlug, customModePrompts: defaultPrompts, customSupportPrompts: {}, + experiments: experimentDefault, enhancementApiConfigId: "", - experimentalDiffStrategy: false, autoApprovalEnabled: false, customModes: [], }) @@ -242,7 +241,8 @@ export const ExtensionStateContextProvider: React.FC<{ children: React.ReactNode fuzzyMatchThreshold: state.fuzzyMatchThreshold, writeDelayMs: state.writeDelayMs, screenshotQuality: state.screenshotQuality, - experimentalDiffStrategy: state.experimentalDiffStrategy ?? false, + setExperimentEnabled: (id, enabled) => + setState((prevState) => ({ ...prevState, experiments: { ...prevState.experiments, [id]: enabled } })), setApiConfiguration: (value) => setState((prevState) => ({ ...prevState, @@ -279,8 +279,6 @@ export const ExtensionStateContextProvider: React.FC<{ children: React.ReactNode setCustomSupportPrompts: (value) => setState((prevState) => ({ ...prevState, customSupportPrompts: value })), setEnhancementApiConfigId: (value) => setState((prevState) => ({ ...prevState, enhancementApiConfigId: value })), - setExperimentalDiffStrategy: (value) => - setState((prevState) => ({ ...prevState, experimentalDiffStrategy: value })), setAutoApprovalEnabled: (value) => setState((prevState) => ({ ...prevState, autoApprovalEnabled: value })), handleInputChange, setCustomModes: (value) => setState((prevState) => ({ ...prevState, customModes: value })),