[codex] feat(slash): add guide mode

core/config/profile/doLoadConfig.ts

@@ -28,6 +28,7 @@ import { getControlPlaneEnv } from "../../control-plane/env.js";
 import { PolicySingleton } from "../../control-plane/PolicySingleton";
 import { TeamAnalytics } from "../../control-plane/TeamAnalytics.js";
 import ContinueProxy from "../../llm/llms/stubs/ContinueProxy";
+import { guideSlashCommand } from "../../promptFiles/guidePrompt";
 import { initSlashCommand } from "../../promptFiles/initPrompt";
 import { getConfigDependentToolDefinitions } from "../../tools";
 import { encodeMCPToolUri } from "../../tools/callTool";
@@ -189,6 +190,7 @@ export default async function doLoadConfig(options: {
   }
 
   newConfig.slashCommands.push(initSlashCommand);
+  newConfig.slashCommands.push(guideSlashCommand);
 
   const proxyContextProvider = newConfig.contextProviders?.find(
     (cp) => cp.description.title === "continue-proxy",

core/config/profile/doLoadConfig.vitest.ts

@@ -95,6 +95,9 @@ vi.mock("../../control-plane/TeamAnalytics", () => ({
 vi.mock("../../promptFiles/initPrompt", () => ({
   initSlashCommand: { name: "init", description: "init" },
 }));
+vi.mock("../../promptFiles/guidePrompt", () => ({
+  guideSlashCommand: { name: "guide", description: "guide" },
+}));
 
 // Mock fs.existsSync to simulate missing file on disk
 vi.mock("fs", async (importOriginal) => {
@@ -133,26 +136,29 @@ const mockControlPlaneClient = {
 
 const mockLlmLogger = {} as any;
 
+function createPackageIdentifier(withContent = true): PackageIdentifier {
+  return {
+    uriType: "file",
+    fileUri: "vscode-remote://wsl+Ubuntu/home/user/.continue/agents/test.yaml",
+    ...(withContent
+      ? { content: "name: Test\nversion: 1.0.0\nschema: v1\n" }
+      : {}),
+  };
+}
+
 describe("doLoadConfig pre-read content bypass", () => {
   it("should use YAML loading when packageIdentifier has pre-read content, even if file does not exist on disk", async () => {
     mockLoadYaml.mockClear();
     mockLoadJson.mockClear();
 
-    const packageIdentifier: PackageIdentifier = {
-      uriType: "file",
-      fileUri:
-        "vscode-remote://wsl+Ubuntu/home/user/.continue/agents/test.yaml",
-      content: "name: Test\nversion: 1.0.0\nschema: v1\n",
-    };
-
     await doLoadConfig({
       ide: mockIde,
       controlPlaneClient: mockControlPlaneClient,
       llmLogger: mockLlmLogger,
       profileId: "test-profile",
-      overrideConfigYamlByPath: packageIdentifier.fileUri,
+      overrideConfigYamlByPath: createPackageIdentifier(true).fileUri,
       orgScopeId: null,
-      packageIdentifier,
+      packageIdentifier: createPackageIdentifier(true),
     });
 
     expect(mockLoadYaml).toHaveBeenCalled();
@@ -163,23 +169,33 @@ describe("doLoadConfig pre-read content bypass", () => {
     mockLoadYaml.mockClear();
     mockLoadJson.mockClear();
 
-    const packageIdentifier: PackageIdentifier = {
-      uriType: "file",
-      fileUri:
-        "vscode-remote://wsl+Ubuntu/home/user/.continue/agents/test.yaml",
-    };
-
     await doLoadConfig({
       ide: mockIde,
       controlPlaneClient: mockControlPlaneClient,
       llmLogger: mockLlmLogger,
       profileId: "test-profile",
-      overrideConfigYamlByPath: packageIdentifier.fileUri,
+      overrideConfigYamlByPath: createPackageIdentifier(false).fileUri,
       orgScopeId: null,
-      packageIdentifier,
+      packageIdentifier: createPackageIdentifier(false),
     });
 
     expect(mockLoadYaml).not.toHaveBeenCalled();
     expect(mockLoadJson).toHaveBeenCalled();
   });
+
+  it("should always include built-in init and guide slash commands", async () => {
+    const result = await doLoadConfig({
+      ide: mockIde,
+      controlPlaneClient: mockControlPlaneClient,
+      llmLogger: mockLlmLogger,
+      profileId: "test-profile",
+      overrideConfigYamlByPath: createPackageIdentifier(true).fileUri,
+      orgScopeId: null,
+      packageIdentifier: createPackageIdentifier(true),
+    });
+
+    const commandNames = result.config?.slashCommands.map((cmd) => cmd.name);
+    expect(commandNames).toContain("init");
+    expect(commandNames).toContain("guide");
+  });
 });

core/promptFiles/guidePrompt.ts

@@ -0,0 +1,52 @@
+import { SlashCommandWithSource } from "..";
+
+export const GUIDE_PROMPT_CONTENT = `
+You are Continue Guide Mode, an interactive mentor for developers who need help turning a rough idea into a concrete implementation plan.
+
+Your job is not to jump straight into coding. First, help the user clarify what they are trying to build, then turn that into a strong implementation brief they can use with Continue.
+
+## Goals
+- Help beginners describe their project clearly.
+- Ask discovery questions that improve the next coding step.
+- Teach the user what information helps AI produce better results.
+- End with a concrete, structured specification and a suggested next prompt.
+
+## How to behave
+- Be encouraging, practical, and concise.
+- Ask only for information that materially improves the plan.
+- Prefer plain language over jargon.
+- If the user already answered some questions, do not ask them again.
+- If the user gives a vague idea, ask targeted follow-up questions.
+- Ask at most 2 questions per response so the interaction stays lightweight.
+
+## Discovery Areas
+Collect enough detail to cover these five areas:
+1. What they want to build.
+2. Who it is for.
+3. What problem it solves.
+4. Their experience level.
+5. Requirements, constraints, or preferred technologies.
+
+## Response strategy
+- If key information is missing, ask the next most important question or two.
+- Once you have enough information, stop asking questions and produce a structured specification.
+- If the user asks for direct help before the discovery is complete, give a short answer and then continue the discovery flow.
+
+## When you have enough information
+Produce these sections in order:
+1. **Project Summary**: one short paragraph.
+2. **Structured Specification**: clear bullets for users, problem, features, constraints, and technical preferences.
+3. **Implementation Plan**: 3-6 concrete build steps.
+4. **Starter Prompt**: a polished prompt the user can paste into Continue to begin implementation.
+5. **What To Clarify Next**: optional, only if meaningful gaps remain.
+
+## If the user has not provided any project idea yet
+Start by asking: "What are you trying to build? Describe your idea in your own words."
+`.trim();
+
+export const guideSlashCommand: SlashCommandWithSource = {
+  name: "guide",
+  description: "Turn a rough idea into a structured build plan",
+  source: "built-in",
+  prompt: GUIDE_PROMPT_CONTENT,
+};

docs/guides/cli.mdx

@@ -4,8 +4,8 @@ sidebarTitle: "Continue CLI (cn)"
 description: "Learn how to use Continue's command-line interface for context engineering, automated coding tasks, and headless development workflows with customizable models, rules, and tools"
 ---
 
-import { OSAutoDetect } from '/snippets/OSAutoDetect.jsx'
-import CLIInstall from '/snippets/cli-install.mdx'
+import { OSAutoDetect } from "/snippets/OSAutoDetect.jsx";
+import CLIInstall from "/snippets/cli-install.mdx";
 
 <OSAutoDetect />
 
@@ -39,7 +39,7 @@ Out of the box, `cn` comes with tools that let it understand your codebase, edit
 - Write a new feature
 - And a lot more
 
-Use '@' to give it file context, or '/' to run slash commands.
+Use '@' to give it file context, or '/' to run slash commands. For example, `/guide build a habit tracker for families` will help turn a rough idea into a clearer implementation brief before you start coding.
 
 If you want to resume a previous conversation, run `cn --resume`.
 

docs/reference/json-reference.mdx

@@ -399,6 +399,25 @@ config.json
 }
 ```
 
+### `/guide`
+
+The guide slash command helps a user turn a rough idea into a more actionable build plan. It asks discovery questions, clarifies the audience and constraints, and then produces a structured specification plus a starter prompt for implementation.
+
+config.json
+
+```json
+{
+  "slashCommands": [
+    {
+      "name": "guide",
+      "description": "Turn a rough idea into a structured build plan"
+    }
+  ]
+}
+```
+
+Example: `/guide build a portfolio site for junior developers`
+
 Example:
 
 config.json
@@ -480,7 +499,6 @@ Several experimental config parameters are available, as described below:
 - `defaultContext`: Defines the default context for the LLM. Uses the same format as `contextProviders` but includes an additional `query` property to specify custom query parameters.=
 
 - `modelRoles`:
-
   - `inlineEdit`: Model title for inline edits.
   - `applyCodeBlock`: Model title for applying code blocks.
   - `repoMapFileSelection`: Model title for repo map selections.
@@ -521,18 +539,15 @@ Some deprecated `config.json` settings are no longer stored in config and have b
 - `disableSessionTitles`/`ui.getChatTitles`: This value will be migrated to the safest merged value (`true` if either are `true`). `getChatTitles` takes precedence if set to false
 
 - `tabAutocompleteOptions`
-
   - `useCache`: This value will override during migration.
   - `disableInFiles`: This value will be migrated to the safest merged value (arrays of file matches merged/deduplicated)
   - `multilineCompletions`: This value will override during migration.
 
 - `experimental`
-
   - `useChromiumForDocsCrawling`: This value will override during migration.
   - `readResponseTTS`: This value will override during migration.
 
 - `ui` - all will override during migration
-
   - `codeBlockToolbarPosition`
   - `fontSize`
   - `codeWrap`

extensions/cli/src/commands/commands.integration.test.ts

@@ -28,6 +28,7 @@ describe("Slash Commands Integration", () => {
       expect(commandNames).toContain("login");
       expect(commandNames).toContain("logout");
       expect(commandNames).toContain("whoami");
+      expect(commandNames).toContain("guide");
 
       expect(commandNames).toContain("model");
       expect(commandNames).toContain("config");
@@ -50,6 +51,7 @@ describe("Slash Commands Integration", () => {
           "login",
           "logout",
           "whoami",
+          "guide",
           "model",
           "config",
         ].includes(cmd.name),

extensions/cli/src/commands/commands.ts

@@ -81,6 +81,11 @@ export const SYSTEM_SLASH_COMMANDS: SystemCommand[] = [
     description: "Create an AGENTS.md file",
     category: "system",
   },
+  {
+    name: "guide",
+    description: "Turn a rough idea into a structured build plan",
+    category: "system",
+  },
   {
     name: "compact",
     description: "Summarize chat history into a compact form",

extensions/cli/src/slashCommands.test.ts

@@ -136,6 +136,20 @@ describe("slashCommands", () => {
       expect(result?.exit).toBeUndefined();
     });
 
+    it("should handle /guide command", async () => {
+      const result = await handleSlashCommands(
+        "/guide Build a portfolio site for junior developers",
+        mockAssistant,
+      );
+
+      expect(result).toBeDefined();
+      expect(result?.newInput).toContain("Continue Guide Mode");
+      expect(result?.newInput).toContain(
+        "Build a portfolio site for junior developers",
+      );
+      expect(result?.output).toBeUndefined();
+    });
+
     it("should handle /info command when not authenticated", async () => {
       const { isAuthenticated } = await import("./auth/workos.js");
       const { services } = await import("./services/index.js");

extensions/cli/src/slashCommands.ts

@@ -25,6 +25,50 @@ import {
   loadMarkdownSkills,
 } from "./util/loadMarkdownSkills.js";
 
+const GUIDE_PROMPT = `
+You are Continue Guide Mode, an interactive mentor for developers who need help turning a rough idea into a concrete implementation plan.
+
+Your job is not to jump straight into coding. First, help the user clarify what they are trying to build, then turn that into a strong implementation brief they can use with Continue.
+
+## Goals
+- Help beginners describe their project clearly.
+- Ask discovery questions that improve the next coding step.
+- Teach the user what information helps AI produce better results.
+- End with a concrete, structured specification and a suggested next prompt.
+
+## How to behave
+- Be encouraging, practical, and concise.
+- Ask only for information that materially improves the plan.
+- Prefer plain language over jargon.
+- If the user already answered some questions, do not ask them again.
+- If the user gives a vague idea, ask targeted follow-up questions.
+- Ask at most 2 questions per response so the interaction stays lightweight.
+
+## Discovery Areas
+Collect enough detail to cover these five areas:
+1. What they want to build.
+2. Who it is for.
+3. What problem it solves.
+4. Their experience level.
+5. Requirements, constraints, or preferred technologies.
+
+## Response strategy
+- If key information is missing, ask the next most important question or two.
+- Once you have enough information, stop asking questions and produce a structured specification.
+- If the user asks for direct help before the discovery is complete, give a short answer and then continue the discovery flow.
+
+## When you have enough information
+Produce these sections in order:
+1. **Project Summary**: one short paragraph.
+2. **Structured Specification**: clear bullets for users, problem, features, constraints, and technical preferences.
+3. **Implementation Plan**: 3-6 concrete build steps.
+4. **Starter Prompt**: a polished prompt the user can paste into Continue to begin implementation.
+5. **What To Clarify Next**: optional, only if meaningful gaps remain.
+
+## If the user has not provided any project idea yet
+Start by asking: "What are you trying to build? Describe your idea in your own words."
+`.trim();
+
 type CommandHandler = (
   args: string[],
   assistant: AssistantConfig,
@@ -63,6 +107,17 @@ async function handleHelp(_args: string[], _assistant: AssistantConfig) {
   return { output: helpMessage };
 }
 
+async function handleGuide(args: string[]): Promise<SlashCommandResult> {
+  const idea = args.join(" ").trim();
+  return {
+    newInput: idea
+      ? `${GUIDE_PROMPT}
+
+${idea}`
+      : GUIDE_PROMPT,
+  };
+}
+
 async function handleLogin() {
   try {
     const newAuthState = await services.auth.login();
@@ -361,6 +416,7 @@ const commandHandlers: Record<string, CommandHandler> = {
   init: (args, assistant) => {
     return handleInit(args, assistant);
   },
+  guide: (args) => handleGuide(args),
   update: () => {
     return { openUpdateSelector: true };
   },