|
| 1 | +import { describe, it, expect, beforeEach, afterEach, vi } from "vitest" |
| 2 | +import type { Anthropic } from "@anthropic-ai/sdk" |
| 3 | + |
| 4 | +import type { ApiStreamChunk } from "../../transform/stream" |
| 5 | +import type { ApiHandlerCreateMessageMetadata } from "../../index" |
| 6 | +import type { ApiHandlerOptions } from "../../../shared/api" |
| 7 | +import { OpenAiNativeHandler } from "../openai-native" |
| 8 | + |
| 9 | +const createSessionMock = vi.hoisted(() => vi.fn()) |
| 10 | + |
| 11 | +vi.mock("openai", () => { |
| 12 | + return { |
| 13 | + __esModule: true, |
| 14 | + default: vi.fn().mockImplementation(() => ({ |
| 15 | + responses: { |
| 16 | + create: vi.fn(), |
| 17 | + }, |
| 18 | + })), |
| 19 | + } |
| 20 | +}) |
| 21 | + |
| 22 | +vi.mock("../../../integrations/codex/run", () => ({ |
| 23 | + CodexCliSession: { |
| 24 | + create: createSessionMock, |
| 25 | + }, |
| 26 | +})) |
| 27 | + |
| 28 | +import { CodexHandler } from "../codex" |
| 29 | + |
| 30 | +const systemPrompt = "You are Codex." |
| 31 | + |
| 32 | +const defaultOptions: ApiHandlerOptions = { |
| 33 | + apiModelId: "gpt-5-codex", |
| 34 | + openAiNativeApiKey: "test-api-key", |
| 35 | +} |
| 36 | + |
| 37 | +const metadata: ApiHandlerCreateMessageMetadata = { |
| 38 | + taskId: "task-123", |
| 39 | +} |
| 40 | + |
| 41 | +const makeSession = (chunks: ApiStreamChunk[]) => ({ |
| 42 | + runTurn: vi.fn().mockResolvedValue(asyncGeneratorFromChunks(chunks)), |
| 43 | + shutdown: vi.fn().mockResolvedValue(undefined), |
| 44 | +}) |
| 45 | + |
| 46 | +function asyncGeneratorFromChunks(chunks: ApiStreamChunk[]): AsyncGenerator<ApiStreamChunk> { |
| 47 | + return (async function* () { |
| 48 | + for (const chunk of chunks) { |
| 49 | + yield chunk |
| 50 | + } |
| 51 | + })() |
| 52 | +} |
| 53 | + |
| 54 | +describe("CodexHandler", () => { |
| 55 | + let handler: CodexHandler |
| 56 | + let fallbackSpy: ReturnType<typeof vi.spyOn> |
| 57 | + |
| 58 | + beforeEach(() => { |
| 59 | + createSessionMock.mockReset() |
| 60 | + fallbackSpy = vi.spyOn(OpenAiNativeHandler.prototype, "createMessage").mockImplementation(async function* () { |
| 61 | + yield { type: "text", text: "[fallback]" } |
| 62 | + }) |
| 63 | + handler = new CodexHandler(defaultOptions) |
| 64 | + }) |
| 65 | + |
| 66 | + afterEach(() => { |
| 67 | + fallbackSpy.mockRestore() |
| 68 | + }) |
| 69 | + |
| 70 | + it("omits duplicate prefix chunks before streaming new Codex output", async () => { |
| 71 | + const messages: Anthropic.Messages.MessageParam[] = [ |
| 72 | + { role: "user", content: "Initial request" }, |
| 73 | + { role: "assistant", content: "Task completed successfully." }, |
| 74 | + { role: "user", content: "Please continue" }, |
| 75 | + ] |
| 76 | + |
| 77 | + createSessionMock.mockResolvedValueOnce( |
| 78 | + makeSession([ |
| 79 | + { type: "text", text: "Task completed successfully." }, |
| 80 | + { type: "text", text: "Here are the next steps." }, |
| 81 | + { type: "usage", inputTokens: 10, outputTokens: 5 }, |
| 82 | + ]), |
| 83 | + ) |
| 84 | + |
| 85 | + const stream = handler.createMessage(systemPrompt, messages, metadata) |
| 86 | + const chunks: ApiStreamChunk[] = [] |
| 87 | + for await (const chunk of stream) { |
| 88 | + chunks.push(chunk) |
| 89 | + } |
| 90 | + |
| 91 | + const textChunks = chunks.filter((chunk) => chunk.type === "text") |
| 92 | + expect(textChunks).toHaveLength(1) |
| 93 | + expect(textChunks[0].text).toBe("Here are the next steps.") |
| 94 | + const usageChunk = chunks.find((chunk) => chunk.type === "usage") |
| 95 | + expect(usageChunk).toBeTruthy() |
| 96 | + expect(fallbackSpy).not.toHaveBeenCalled() |
| 97 | + }) |
| 98 | + |
| 99 | + it("replays the turn without the prior assistant message when Codex repeats itself", async () => { |
| 100 | + const messages: Anthropic.Messages.MessageParam[] = [ |
| 101 | + { role: "user", content: "Initial request" }, |
| 102 | + { role: "assistant", content: "Task completed successfully." }, |
| 103 | + { role: "user", content: "Please continue" }, |
| 104 | + ] |
| 105 | + |
| 106 | + createSessionMock |
| 107 | + .mockResolvedValueOnce( |
| 108 | + makeSession([ |
| 109 | + { type: "text", text: "Task completed successfully." }, |
| 110 | + { type: "usage", inputTokens: 10, outputTokens: 2 }, |
| 111 | + ]), |
| 112 | + ) |
| 113 | + .mockResolvedValueOnce( |
| 114 | + makeSession([ |
| 115 | + { type: "text", text: "Continuing with the follow-up work." }, |
| 116 | + { type: "usage", inputTokens: 12, outputTokens: 6 }, |
| 117 | + ]), |
| 118 | + ) |
| 119 | + |
| 120 | + const stream = handler.createMessage(systemPrompt, messages, metadata) |
| 121 | + const chunks: ApiStreamChunk[] = [] |
| 122 | + for await (const chunk of stream) { |
| 123 | + chunks.push(chunk) |
| 124 | + } |
| 125 | + |
| 126 | + const textChunks = chunks.filter((chunk) => chunk.type === "text") |
| 127 | + expect(textChunks).toHaveLength(1) |
| 128 | + expect(textChunks[0].text).toBe("Continuing with the follow-up work.") |
| 129 | + expect(createSessionMock).toHaveBeenCalledTimes(2) |
| 130 | + expect(fallbackSpy).not.toHaveBeenCalled() |
| 131 | + }) |
| 132 | +}) |
0 commit comments