disclaw/tests/unit/runner-model.test.ts
Nick Tabeling bc4d5e73ee
Some checks failed
CI / build-and-test (ubuntu-latest) (pull_request) Has been cancelled
CI / build-and-test (windows-latest) (pull_request) Has been cancelled
CI / lint (pull_request) Has been cancelled
feat(DIS-153): model selection per agent
Adds per-agent model configuration: ALLOWED_MODELS allowlist in
src/config/models.ts, model field in AgentYamlSchema (Zod default:
claude-sonnet-4-6), --model flag forwarded to the Claude CLI in
runner.ts, and a StringSelectMenuBuilder step in the /new-agent
command flow. Three unit tests cover model passthrough, default
fallback, and invalid-model rejection.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
2026-04-13 09:35:27 +02:00

165 lines
5.4 KiB
TypeScript

import { describe, it, expect, vi, beforeEach, afterEach, beforeAll, afterAll } from "vitest";
import * as fs from "fs";
import * as os from "os";
import * as path from "path";
/**
* Unit tests for DIS-153: model selection per agent.
*
* cross-spawn is mocked so no real process is launched.
* Each test inspects the args passed to spawn to verify --model is set correctly.
*/
// ---------------------------------------------------------------------------
// Spy on the args cross-spawn receives
// ---------------------------------------------------------------------------
let capturedArgs: string[] = [];
vi.mock("cross-spawn", () => {
function makeMockChild(args: string[]) {
capturedArgs = args;
const stdoutListeners: Record<string, Array<(...a: unknown[]) => void>> = {};
const stdoutStream = {
setEncoding: () => {},
on: (event: string, cb: (...a: unknown[]) => void) => {
if (!stdoutListeners[event]) stdoutListeners[event] = [];
stdoutListeners[event].push(cb);
if (event === "data") {
Promise.resolve().then(() => {
for (const handler of stdoutListeners["data"] ?? []) {
handler(JSON.stringify({ result: "ok" }));
}
});
}
},
};
const listeners: Record<string, Array<(...a: unknown[]) => void>> = {};
const emit = (event: string, ...a: unknown[]) => {
for (const cb of listeners[event] ?? []) cb(...a);
};
const child = {
stdout: stdoutStream,
stderr: { setEncoding: () => {}, on: () => {} },
stdin: { end: () => {} },
on: (event: string, cb: (...a: unknown[]) => void) => {
if (!listeners[event]) listeners[event] = [];
listeners[event].push(cb);
if (event === "close") {
Promise.resolve()
.then(() => Promise.resolve())
.then(() => emit("close", 0));
}
},
kill: () => {},
};
return child;
}
return {
default: (_cmd: string, args: string[]) => makeMockChild(args),
};
});
import { runAgent, _resetSemaphore } from "../../src/agent/runner";
import { _resetResolveClaudeCache } from "../../src/runtime/resolve-claude";
import { AgentYamlSchema } from "../../src/agent/schema";
// ---------------------------------------------------------------------------
// Shared workspace fixture
// ---------------------------------------------------------------------------
describe("runner model selection (DIS-153)", () => {
let workspaceDir: string;
const originalEnv = { ...process.env };
beforeAll(() => {
workspaceDir = fs.mkdtempSync(path.join(os.tmpdir(), "disclaw-model-test-"));
fs.writeFileSync(
path.join(workspaceDir, "CLAUDE.md"),
"# Model Test Agent\nYou are a test agent.\n",
"utf-8"
);
});
beforeEach(() => {
capturedArgs = [];
_resetResolveClaudeCache();
_resetSemaphore(4);
process.env.CLAUDE_PATH = "/fake/claude";
});
afterEach(() => {
process.env = { ...originalEnv };
_resetResolveClaudeCache();
});
afterAll(() => {
try {
fs.rmSync(workspaceDir, { recursive: true, force: true });
} catch {
// best-effort cleanup
}
});
// -------------------------------------------------------------------------
// Test 1: --model is passed when model is set in agent.yaml
// -------------------------------------------------------------------------
it("passes --model claude-sonnet-4-6 when model is set in agent.yaml", async () => {
fs.writeFileSync(
path.join(workspaceDir, "agent.yaml"),
`name: model-agent\ndisplay_name: Model Agent\nrole: Testing\nchannel_id: "111"\nmodel: claude-sonnet-4-6\n`,
"utf-8"
);
await runAgent({
workspacePath: workspaceDir,
channelName: "test",
userMessage: "hello",
conversationHistory: [],
});
const modelIdx = capturedArgs.indexOf("--model");
expect(modelIdx).toBeGreaterThan(-1);
expect(capturedArgs[modelIdx + 1]).toBe("claude-sonnet-4-6");
});
// -------------------------------------------------------------------------
// Test 2: Default model is used when model field is absent from agent.yaml
// -------------------------------------------------------------------------
it("passes default model when model field is absent from agent.yaml", async () => {
fs.writeFileSync(
path.join(workspaceDir, "agent.yaml"),
`name: model-agent\ndisplay_name: Model Agent\nrole: Testing\nchannel_id: "222"\n`,
"utf-8"
);
await runAgent({
workspacePath: workspaceDir,
channelName: "test",
userMessage: "hello",
conversationHistory: [],
});
const modelIdx = capturedArgs.indexOf("--model");
expect(modelIdx).toBeGreaterThan(-1);
// DEFAULT_MODEL is claude-sonnet-4-6
expect(capturedArgs[modelIdx + 1]).toBe("claude-sonnet-4-6");
});
// -------------------------------------------------------------------------
// Test 3: Invalid model in agent.yaml → Zod validation throws
// -------------------------------------------------------------------------
it("throws when agent.yaml contains an invalid model value", () => {
const rawYaml = `name: bad-model-agent\ndisplay_name: Bad\nrole: Testing\nchannel_id: "333"\nmodel: gpt-4-turbo\n`;
const parsed = { name: "bad-model-agent", display_name: "Bad", role: "Testing", channel_id: "333", model: "gpt-4-turbo" };
expect(() => AgentYamlSchema.parse(parsed)).toThrow();
});
});