oh-my-pi/packages/coding-agent/test/acp-agent.test.ts

import { afterEach, describe, expect, it } from "bun:test";
import * as fs from "node:fs";
import * as os from "node:os";
import * as path from "node:path";
import type { AgentSideConnection, PromptRequest, SessionNotification } from "@agentclientprotocol/sdk";
import type { Model } from "@oh-my-pi/pi-ai";
import { getConfigRootDir, setAgentDir } from "@oh-my-pi/pi-utils";
import { AcpAgent } from "../src/modes/acp/acp-agent";
import type { AgentSession, AgentSessionEvent } from "../src/session/agent-session";
import { SessionManager } from "../src/session/session-manager";

const TEST_MODELS: Model[] = [
	{
		id: "claude-sonnet-4-20250514",
		name: "Claude Sonnet",
		api: "anthropic-messages",
		provider: "anthropic",
		baseUrl: "https://example.invalid",
		reasoning: true,
		input: ["text", "image"],
		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
		contextWindow: 200_000,
		maxTokens: 8_192,
	},
	{
		id: "gpt-5.4",
		name: "GPT-5.4",
		api: "openai-responses",
		provider: "openai",
		baseUrl: "https://example.invalid",
		reasoning: true,
		input: ["text", "image"],
		cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
		contextWindow: 200_000,
		maxTokens: 8_192,
	},
];

function makeAssistantMessage(text: string, thinking?: string) {
	const content: Array<{ type: "text"; text: string } | { type: "thinking"; thinking: string }> = [
		{ type: "text", text },
	];
	if (thinking) {
		content.push({ type: "thinking" as const, thinking });
	}
	return {
		role: "assistant" as const,
		content,
		api: "anthropic-messages" as const,
		provider: "anthropic" as const,
		model: TEST_MODELS[0].id,
		usage: {
			input: 10,
			output: 5,
			cacheRead: 2,
			cacheWrite: 1,
			totalTokens: 18,
			cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 },
		},
		stopReason: "stop" as const,
		timestamp: Date.now(),
	};
}

class FakeAgentSession {
	sessionManager: SessionManager;
	sessionId: string;
	agent: { sessionId: string; waitForIdle: () => Promise<void> };
	model: Model | undefined;
	thinkingLevel: string | undefined;
	customCommands: [] = [];
	extensionRunner = undefined;
	searchDb = undefined;
	isStreaming = false;
	queuedMessageCount = 0;
	systemPrompt = "system";
	disposed = false;
	#listeners = new Set<(event: AgentSessionEvent) => void>();

	constructor(
		cwd: string,
		private readonly models: Model[] = TEST_MODELS,
	) {
		this.sessionManager = SessionManager.create(cwd);
		this.sessionId = this.sessionManager.getSessionId();
		this.agent = {
			sessionId: this.sessionId,
			waitForIdle: async () => {},
		};
		this.model = models[0];
	}

	get sessionName(): string {
		return this.sessionManager.getHeader()?.title ?? `Session ${this.sessionId}`;
	}

	get modelRegistry(): { getApiKey: (model: Model) => Promise<string> } {
		return {
			getApiKey: async (_model: Model) => "test-key",
		};
	}

	getAvailableModels(): Model[] {
		return this.models;
	}

	getAvailableThinkingLevels(): ReadonlyArray<string> {
		return ["low", "medium", "high"];
	}

	setThinkingLevel(level: string | undefined): void {
		this.thinkingLevel = level;
	}

	async setModel(model: Model): Promise<void> {
		this.model = model;
	}

	subscribe(listener: (event: AgentSessionEvent) => void): () => void {
		this.#listeners.add(listener);
		return () => {
			this.#listeners.delete(listener);
		};
	}

	async prompt(text: string): Promise<void> {
		this.isStreaming = true;
		this.sessionManager.appendMessage({ role: "user", content: text, timestamp: Date.now() });
		const assistantMessage = makeAssistantMessage("pong");
		for (const listener of this.#listeners) {
			listener({
				type: "message_update",
				message: assistantMessage,
				assistantMessageEvent: { type: "text_delta", delta: "pong" },
			} as AgentSessionEvent);
		}
		this.sessionManager.appendMessage(assistantMessage);
		for (const listener of this.#listeners) {
			listener({
				type: "agent_end",
				messages: [assistantMessage],
			} as AgentSessionEvent);
		}
		this.isStreaming = false;
	}

	async abort(): Promise<void> {
		this.isStreaming = false;
	}

	async refreshMCPTools(_tools: unknown[]): Promise<void> {}

	getContextUsage(): undefined {
		return undefined;
	}

	async switchSession(sessionPath: string): Promise<boolean> {
		await this.sessionManager.setSessionFile(sessionPath);
		this.sessionId = this.sessionManager.getSessionId();
		this.agent.sessionId = this.sessionId;
		return true;
	}

	async dispose(): Promise<void> {
		this.disposed = true;
		await this.sessionManager.close();
	}

	async reload(): Promise<void> {}

	async newSession(): Promise<boolean> {
		await this.sessionManager.newSession();
		this.sessionId = this.sessionManager.getSessionId();
		this.agent.sessionId = this.sessionId;
		return true;
	}

	async branch(_entryId: string): Promise<{ cancelled: boolean }> {
		return { cancelled: false };
	}

	async navigateTree(_targetId: string): Promise<{ cancelled: boolean }> {
		return { cancelled: false };
	}

	getActiveToolNames(): string[] {
		return [];
	}

	getAllToolNames(): string[] {
		return [];
	}

	setActiveToolsByName(_toolNames: string[]): void {}

	async sendCustomMessage(_message: string, _options?: unknown): Promise<void> {}

	async sendUserMessage(_content: string, _options?: unknown): Promise<void> {}

	async compact(_instructions?: string, _options?: unknown): Promise<void> {}
}

interface AgentHarness {
	agent: AcpAgent;
	updates: SessionNotification[];
	abortController: AbortController;
	sessions: FakeAgentSession[];
	cwdA: string;
	cwdB: string;
	findSession(sessionId: string): FakeAgentSession | undefined;
}

function getChunkMessageId(notification: SessionNotification): string | undefined {
	const update = notification.update as { messageId?: string | null };
	return typeof update.messageId === "string" ? update.messageId : undefined;
}

const cleanupRoots: string[] = [];
const originalAgentDir = process.env.PI_CODING_AGENT_DIR;
const fallbackAgentDir = path.join(getConfigRootDir(), "agent");

afterEach(async () => {
	if (originalAgentDir) {
		setAgentDir(originalAgentDir);
	} else {
		setAgentDir(fallbackAgentDir);
		delete process.env.PI_CODING_AGENT_DIR;
	}

	for (const root of cleanupRoots.splice(0)) {
		await fs.promises.rm(root, { recursive: true, force: true });
	}
});

async function createHarness(): Promise<AgentHarness> {
	const root = await fs.promises.mkdtemp(path.join(os.tmpdir(), "omp-acp-test-"));
	cleanupRoots.push(root);
	const agentDir = path.join(root, "agent");
	const cwdA = path.join(root, "cwd-a");
	const cwdB = path.join(root, "cwd-b");
	await fs.promises.mkdir(agentDir, { recursive: true });
	await fs.promises.mkdir(cwdA, { recursive: true });
	await fs.promises.mkdir(cwdB, { recursive: true });
	setAgentDir(agentDir);

	const updates: SessionNotification[] = [];
	const abortController = new AbortController();
	const sessions: FakeAgentSession[] = [];
	const connection = {
		sessionUpdate: async (notification: SessionNotification) => {
			updates.push(notification);
		},
		signal: abortController.signal,
		closed: Promise.withResolvers<void>().promise,
	} as unknown as AgentSideConnection;

	const initialSession = new FakeAgentSession(cwdA);
	sessions.push(initialSession);
	const factory = async (cwd: string): Promise<AgentSession> => {
		const session = new FakeAgentSession(cwd);
		sessions.push(session);
		return session as unknown as AgentSession;
	};

	return {
		agent: new AcpAgent(connection, initialSession as unknown as AgentSession, factory),
		updates,
		abortController,
		sessions,
		cwdA,
		cwdB,
		findSession: (sessionId: string) => sessions.find(session => session.sessionId === sessionId),
	};
}

describe("ACP agent", () => {
	it("supports multiple live ACP sessions with model and lifecycle handlers", async () => {
		const harness = await createHarness();
		const first = await harness.agent.newSession({ cwd: harness.cwdA, mcpServers: [] });
		const second = await harness.agent.newSession({ cwd: harness.cwdB, mcpServers: [] });

		expect(first.models?.availableModels.map(model => model.modelId)).toEqual(
			TEST_MODELS.map(model => `${model.provider}/${model.id}`),
		);

		await harness.agent.unstable_setSessionModel({
			sessionId: first.sessionId,
			modelId: `${TEST_MODELS[1]!.provider}/${TEST_MODELS[1]!.id}`,
		});
		await harness.agent.setSessionConfigOption({
			sessionId: first.sessionId,
			configId: "thinking",
			value: "high",
		});

		const firstSession = harness.findSession(first.sessionId);
		const secondSession = harness.findSession(second.sessionId);
		expect(firstSession?.model?.id).toBe(TEST_MODELS[1]!.id);
		expect(firstSession?.thinkingLevel).toBe("high");
		expect(secondSession?.model?.id).toBe(TEST_MODELS[0]!.id);
		expect(secondSession?.thinkingLevel).toBeUndefined();

		firstSession?.sessionManager.appendMessage({ role: "user", content: "fork me", timestamp: Date.now() });
		await firstSession?.sessionManager.flush();

		const forked = await harness.agent.unstable_forkSession({
			sessionId: first.sessionId,
			cwd: harness.cwdA,
			mcpServers: [],
		});
		const forkedSession = harness.findSession(forked.sessionId);
		const forkedMessages = forkedSession?.sessionManager.buildSessionContext().messages ?? [];
		expect(forked.sessionId).not.toBe(first.sessionId);
		expect(forkedMessages.some(message => message.role === "user" && message.content === "fork me")).toBe(true);

		await harness.agent.unstable_closeSession({ sessionId: forked.sessionId });
		await expect(harness.agent.setSessionMode({ sessionId: forked.sessionId, modeId: "default" })).rejects.toThrow(
			"Unsupported ACP session",
		);

		harness.abortController.abort();
		await Bun.sleep(0);
	});

	it("replays messageIds and returns turn usage for prompts", async () => {
		const harness = await createHarness();
		const stored = new FakeAgentSession(harness.cwdA);
		harness.sessions.push(stored);
		stored.sessionManager.appendMessage({ role: "user", content: "hello", timestamp: Date.now() });
		stored.sessionManager.appendMessage(makeAssistantMessage("reply", "reasoning"));
		await stored.sessionManager.ensureOnDisk();
		await stored.sessionManager.flush();

		await harness.agent.loadSession({ sessionId: stored.sessionId, cwd: harness.cwdA, mcpServers: [] });
		const replayChunks = harness.updates.filter(
			update =>
				update.sessionId === stored.sessionId &&
				(update.update.sessionUpdate === "user_message_chunk" ||
					update.update.sessionUpdate === "agent_message_chunk" ||
					update.update.sessionUpdate === "agent_thought_chunk"),
		);
		const replayAssistantChunks = replayChunks.filter(
			update =>
				update.update.sessionUpdate === "agent_message_chunk" ||
				update.update.sessionUpdate === "agent_thought_chunk",
		);

		expect(
			replayChunks.every(
				update => typeof getChunkMessageId(update) === "string" && getChunkMessageId(update)!.length > 0,
			),
		).toBe(true);
		expect(new Set(replayAssistantChunks.map(update => getChunkMessageId(update))).size).toBe(1);

		const live = await harness.agent.newSession({ cwd: harness.cwdB, mcpServers: [] });
		const response = await harness.agent.prompt({
			sessionId: live.sessionId,
			messageId: "05b17a6f-b310-4be7-b767-6b4f3a84eb63",
			prompt: [{ type: "text", text: "ping" }],
		} as PromptRequest);

		const liveChunks = harness.updates.filter(
			update => update.sessionId === live.sessionId && update.update.sessionUpdate === "agent_message_chunk",
		);
		expect(response.userMessageId).toBe("05b17a6f-b310-4be7-b767-6b4f3a84eb63");
		expect(response.usage).toEqual({
			inputTokens: 10,
			outputTokens: 5,
			cachedReadTokens: 2,
			cachedWriteTokens: 1,
			totalTokens: 18,
		});
		expect(
			liveChunks.some(
				update => typeof getChunkMessageId(update) === "string" && getChunkMessageId(update)!.length > 0,
			),
		).toBe(true);

		harness.abortController.abort();
		await Bun.sleep(0);
	});
});