feat: openai-compatible-strict-reasoning (2/2)

AGENTS.md

@@ -4,6 +4,7 @@ This file provides guidance to agents when working with code in this repository.
 
 - Settings View Pattern: When working on `SettingsView`, inputs must bind to the local `cachedState`, NOT the live `useExtensionState()`. The `cachedState` acts as a buffer for user edits, isolating them from the `ContextProxy` source-of-truth until the user explicitly clicks "Save". Wiring inputs directly to the live state causes race conditions.
 - Changesets: Do NOT create `.changeset` files for each commit or code change. Changesets are managed separately by maintainers and should not be generated by agents during normal development.
+- CHANGELOG: Do NOT update `CHANGELOG.md` or `src/CHANGELOG.md` in individual PRs. CHANGELOG entries are added in bulk during release preparation PRs.
 
 ## ESLint Suppressions
 

apps/cli/src/commands/cli/__tests__/list.test.ts

@@ -1,7 +1,45 @@
+import fs from "fs"
+import os from "os"
+import path from "path"
+import { EventEmitter } from "events"
+
+import { openRouterDefaultModelId, providerIdentifiers } from "@roo-code/types"
+
 import { readWorkspaceTaskSessions } from "@/lib/task-history/index.js"
-import { isRecord } from "@/lib/utils/guards.js"
 
-import { listSessions, parseFormat } from "../list.js"
+import { listModels, listSessions, parseFormat } from "../list.js"
+
+const extensionHostMock = vi.hoisted(() => ({
+	activate: vi.fn(async () => undefined),
+	dispose: vi.fn(async () => undefined),
+	options: [] as unknown[],
+	responses: [] as unknown[],
+	sendToExtension: vi.fn(),
+}))
+
+vi.mock("@/agent/index.js", () => ({
+	ExtensionHost: class extends EventEmitter {
+		client = {
+			isInitialized: () => true,
+			on: vi.fn(() => () => undefined),
+		}
+
+		constructor(options: unknown) {
+			super()
+			extensionHostMock.options.push(options)
+		}
+
+		activate = extensionHostMock.activate
+		dispose = extensionHostMock.dispose
+
+		sendToExtension(message: unknown): void {
+			extensionHostMock.sendToExtension(message)
+			for (const response of extensionHostMock.responses) {
+				this.emit("extensionWebviewMessage", response)
+			}
+		}
+	},
+}))
 
 vi.mock("@/lib/task-history/index.js", async (importOriginal) => {
 	const actual = await importOriginal<typeof import("@/lib/task-history/index.js")>()
@@ -39,30 +77,88 @@ describe("parseFormat", () => {
 	})
 })
 
-describe("router model extraction", () => {
-	// This mirrors the extraction logic in requestOpenRouterModels (list.ts:226-228)
-	const extractOpenRouterModels = (routerModelsRaw: unknown) => {
-		const routerModels = isRecord(routerModelsRaw) ? routerModelsRaw : {}
-		const openRouterModels = routerModels.openrouter
-		return isRecord(openRouterModels) ? openRouterModels : {}
-	}
+describe("listModels", () => {
+	let tempDir: string
+	let workspacePath: string
+	let extensionPath: string
 
-	it("extracts openrouter models from valid routerModels", () => {
-		const models = { "openai/gpt-4.1": { contextWindow: 128000, supportsPromptCache: false } }
-		const result = extractOpenRouterModels({ openrouter: models })
-		expect(result).toEqual(models)
+	beforeEach(() => {
+		vi.clearAllMocks()
+		extensionHostMock.options.length = 0
+		extensionHostMock.responses.length = 0
+
+		tempDir = fs.mkdtempSync(path.join(os.tmpdir(), "roo-list-test-"))
+		workspacePath = path.join(tempDir, "workspace")
+		extensionPath = path.join(tempDir, "extension")
+		fs.mkdirSync(workspacePath)
+		fs.mkdirSync(extensionPath)
+		fs.writeFileSync(path.join(extensionPath, "extension.js"), "")
 	})
 
-	it("returns empty object when routerModels is null", () => {
-		expect(extractOpenRouterModels(null)).toEqual({})
+	afterEach(() => {
+		fs.rmSync(tempDir, { recursive: true, force: true })
+		vi.restoreAllMocks()
 	})
 
-	it("returns empty object when openrouter key is missing", () => {
-		expect(extractOpenRouterModels({ requesty: {} })).toEqual({})
+	const captureStdout = async (fn: () => Promise<void>): Promise<string> => {
+		const stdoutSpy = vi.spyOn(process.stdout, "write").mockImplementation(() => true)
+		await fn()
+		return stdoutSpy.mock.calls.map(([chunk]) => String(chunk)).join("")
+	}
+
+	it("creates a host with resolved paths and returns OpenRouter models", async () => {
+		const models = { "openai/gpt-4.1": { contextWindow: 128000, supportsPromptCache: false } }
+		extensionHostMock.responses.push(
+			{ type: "unrelatedMessage" },
+			{ type: "routerModels", routerModels: { [providerIdentifiers.openrouter]: models } },
+		)
+
+		const output = await captureStdout(() =>
+			listModels({
+				format: "json",
+				workspace: path.relative(process.cwd(), workspacePath),
+				extension: path.relative(process.cwd(), extensionPath),
+				apiKey: "test-api-key",
+				debug: true,
+			}),
+		)
+
+		expect(extensionHostMock.options).toEqual([
+			expect.objectContaining({
+				mode: "code",
+				provider: providerIdentifiers.openrouter,
+				model: openRouterDefaultModelId,
+				apiKey: "test-api-key",
+				workspacePath,
+				extensionPath,
+				nonInteractive: true,
+				ephemeral: true,
+				debug: true,
+				exitOnComplete: true,
+				exitOnError: false,
+				disableOutput: true,
+			}),
+		])
+		expect(extensionHostMock.activate).toHaveBeenCalledOnce()
+		expect(extensionHostMock.sendToExtension).toHaveBeenCalledWith({
+			type: "requestRouterModels",
+			values: { provider: providerIdentifiers.openrouter },
+		})
+		expect(extensionHostMock.dispose).toHaveBeenCalledOnce()
+		expect(JSON.parse(output)).toEqual({ models })
 	})
 
-	it("returns empty object when openrouter value is not a record", () => {
-		expect(extractOpenRouterModels({ openrouter: "invalid" })).toEqual({})
+	it.each([
+		["a malformed routerModels value", null],
+		["a malformed OpenRouter value", { [providerIdentifiers.openrouter]: "invalid" }],
+	])("returns an empty model record for %s", async (_description, routerModels) => {
+		extensionHostMock.responses.push({ type: "routerModels", routerModels })
+
+		const output = await captureStdout(() =>
+			listModels({ format: "json", workspace: workspacePath, extension: extensionPath }),
+		)
+
+		expect(JSON.parse(output)).toEqual({ models: {} })
 	})
 })
 

apps/cli/src/commands/cli/__tests__/run.test.ts

@@ -2,6 +2,152 @@ import fs from "fs"
 import path from "path"
 import os from "os"
 
+import { providerIdentifiers } from "@roo-code/types"
+import { DEFAULT_FLAGS, FlagOptions } from "@/types/index.js"
+import {
+	resolveLegacyRequireApproval,
+	resolveModel,
+	resolveProvider,
+	resolveReasoningEffort,
+	resolveWorkspacePath,
+	run,
+} from "../run.js"
+
+const runCommandMocks = vi.hoisted(() => ({
+	activate: vi.fn(async () => undefined),
+	dispose: vi.fn(async () => undefined),
+	loadSettings: vi.fn(),
+	options: [] as unknown[],
+	runTask: vi.fn(async () => undefined),
+}))
+
+vi.mock("@/lib/storage/index.js", () => ({
+	loadSettings: runCommandMocks.loadSettings,
+}))
+
+vi.mock("@/agent/index.js", () => ({
+	ExtensionHost: class {
+		client = {}
+
+		constructor(options: unknown) {
+			runCommandMocks.options.push(options)
+		}
+
+		activate = runCommandMocks.activate
+		dispose = runCommandMocks.dispose
+		runTask = runCommandMocks.runTask
+	},
+}))
+
+describe("resolveModel", () => {
+	it("uses the CLI flag before the settings model", () => {
+		expect(resolveModel("flag-model", "settings-model")).toBe("flag-model")
+	})
+
+	it("uses the settings model when the CLI flag is absent", () => {
+		expect(resolveModel(undefined, "settings-model")).toBe("settings-model")
+	})
+
+	it("uses the default model when neither the CLI flag nor settings provide one", () => {
+		expect(resolveModel()).toBe(DEFAULT_FLAGS.model)
+	})
+})
+
+describe("resolveReasoningEffort", () => {
+	it("uses CLI, settings, and default values in priority order", () => {
+		expect(resolveReasoningEffort("high", "low")).toBe("high")
+		expect(resolveReasoningEffort(undefined, "low")).toBe("low")
+		expect(resolveReasoningEffort()).toBe(DEFAULT_FLAGS.reasoningEffort)
+	})
+})
+
+describe("resolveProvider", () => {
+	it("uses CLI, settings, and openrouter values in priority order", () => {
+		expect(resolveProvider(providerIdentifiers.anthropic, providerIdentifiers.gemini)).toBe(
+			providerIdentifiers.anthropic,
+		)
+		expect(resolveProvider(undefined, providerIdentifiers.gemini)).toBe(providerIdentifiers.gemini)
+		expect(resolveProvider()).toBe(providerIdentifiers.openrouter)
+	})
+})
+
+describe("resolveWorkspacePath", () => {
+	it("resolves the provided workspace path", () => {
+		expect(resolveWorkspacePath("relative/workspace")).toBe(path.resolve("relative/workspace"))
+	})
+
+	it("uses the current working directory when workspace is absent", () => {
+		expect(resolveWorkspacePath()).toBe(process.cwd())
+	})
+})
+
+describe("resolveLegacyRequireApproval", () => {
+	it.each([
+		{ requireApproval: true, dangerouslySkipPermissions: true, expected: true },
+		{ requireApproval: false, dangerouslySkipPermissions: false, expected: false },
+		{ requireApproval: undefined, dangerouslySkipPermissions: false, expected: true },
+		{ requireApproval: undefined, dangerouslySkipPermissions: true, expected: false },
+		{ requireApproval: undefined, dangerouslySkipPermissions: undefined, expected: undefined },
+	])(
+		"resolves requireApproval=$requireApproval and dangerouslySkipPermissions=$dangerouslySkipPermissions",
+		({ requireApproval, dangerouslySkipPermissions, expected }) => {
+			expect(resolveLegacyRequireApproval(requireApproval, dangerouslySkipPermissions)).toBe(expected)
+		},
+	)
+})
+
+describe("run command option resolution", () => {
+	let workspacePath: string
+
+	beforeEach(() => {
+		vi.clearAllMocks()
+		runCommandMocks.options.length = 0
+		workspacePath = fs.mkdtempSync(path.join(os.tmpdir(), "roo-run-test-"))
+	})
+
+	afterEach(() => {
+		fs.rmSync(workspacePath, { recursive: true, force: true })
+		vi.restoreAllMocks()
+	})
+
+	it("passes resolved settings and workspace values to the extension host", async () => {
+		runCommandMocks.loadSettings.mockResolvedValue({
+			model: "settings-model",
+			reasoningEffort: "high",
+			provider: providerIdentifiers.anthropic,
+			dangerouslySkipPermissions: false,
+		})
+		const exitSpy = vi.spyOn(process, "exit").mockImplementation(() => undefined as never)
+		const flags: FlagOptions = {
+			continue: false,
+			workspace: path.relative(process.cwd(), workspacePath),
+			print: true,
+			stdinPromptStream: false,
+			signalOnlyExit: false,
+			debug: false,
+			requireApproval: false,
+			exitOnError: false,
+			apiKey: "test-api-key",
+			ephemeral: true,
+			oneshot: false,
+		}
+
+		await run("test prompt", flags)
+
+		expect(runCommandMocks.options).toEqual([
+			expect.objectContaining({
+				model: "settings-model",
+				reasoningEffort: "high",
+				provider: providerIdentifiers.anthropic,
+				workspacePath,
+				nonInteractive: false,
+			}),
+		])
+		expect(runCommandMocks.runTask).toHaveBeenCalledWith("test prompt", undefined)
+		expect(exitSpy).toHaveBeenCalledWith(0)
+	})
+})
+
 describe("run command --prompt-file option", () => {
 	let tempDir: string
 	let promptFilePath: string

apps/cli/src/commands/cli/list.ts

@@ -6,7 +6,7 @@ import pWaitFor from "p-wait-for"
 
 import type { TaskSessionEntry } from "@roo-code/core/cli"
 import type { Command, ModelRecord, WebviewMessage } from "@roo-code/types"
-import { openRouterDefaultModelId } from "@roo-code/types"
+import { openRouterDefaultModelId, providerIdentifiers } from "@roo-code/types"
 
 import { ExtensionHost, type ExtensionHostOptions } from "@/agent/index.js"
 import { readWorkspaceTaskSessions } from "@/lib/task-history/index.js"
@@ -105,13 +105,13 @@ function outputSessionsText(sessions: SessionLike[]): void {
 async function createListHost(options: BaseListOptions, hostOptions: ListHostOptions): Promise<ExtensionHost> {
 	const workspacePath = resolveWorkspacePath(options.workspace)
 	const extensionPath = resolveExtensionPath(options.extension)
-	const apiKey = options.apiKey || getApiKeyFromEnv("openrouter")
+	const apiKey = options.apiKey || getApiKeyFromEnv(providerIdentifiers.openrouter)
 
 	const extensionHostOptions: ExtensionHostOptions = {
 		mode: "code",
 		reasoningEffort: undefined,
 		user: null,
-		provider: "openrouter",
+		provider: providerIdentifiers.openrouter,
 		model: openRouterDefaultModelId,
 		apiKey,
 		workspacePath,
@@ -217,14 +217,14 @@ function requestModes(host: ExtensionHost): Promise<ModeLike[]> {
 function requestOpenRouterModels(host: ExtensionHost): Promise<ModelRecord> {
 	return requestFromExtension(
 		host,
-		{ type: "requestRouterModels", values: { provider: "openrouter" } },
+		{ type: "requestRouterModels", values: { provider: providerIdentifiers.openrouter } },
 		(message) => {
 			if (message.type !== "routerModels") {
 				return undefined
 			}
 
 			const routerModels = isRecord(message.routerModels) ? message.routerModels : {}
-			const openRouterModels = routerModels.openrouter
+			const openRouterModels = routerModels[providerIdentifiers.openrouter]
 			return isRecord(openRouterModels) ? (openRouterModels as ModelRecord) : {}
 		},
 	)