boss/tests/hermes-runner.test.ts

import assert from "node:assert/strict";
import { mkdtemp, realpath, writeFile } from "node:fs/promises";
import { tmpdir } from "node:os";
import { join } from "node:path";
import test from "node:test";

import { runHermesCommandForTesting } from "../src/lib/execution/backends/hermes-runner.ts";

async function createTempScript(source: string) {
  const dir = await mkdtemp(join(tmpdir(), "hermes-runner-"));
  const scriptPath = join(dir, "hermes-script.mjs");
  await writeFile(scriptPath, source, "utf8");
  return { dir, scriptPath };
}

test("Hermes runner 会按固定 chat -q -Q 形态执行并提取正文", async () => {
  const workspace = await mkdtemp(join(tmpdir(), "hermes-runner-cwd-"));
  const expectedWorkspace = await realpath(workspace);
  const { scriptPath } = await createTempScript(`
    process.stdout.write(JSON.stringify({
      argv: process.argv.slice(2),
      cwd: process.cwd(),
      envSource: process.env.HERMES_SESSION_SOURCE || ""
    }) + "\\n");
    process.stdout.write("Hermes smoke completed\\n\\n");
    process.stdout.write("session_id: hermes-session-123\\n");
  `);

  const result = await runHermesCommandForTesting({
    config: {
      enabled: true,
      command: process.execPath,
      args: [scriptPath],
      cwd: workspace,
      timeoutMs: 1000,
      defaultModel: "gpt-5.4",
      toolsets: ["web", "terminal"],
      skills: ["boss-dev"],
      sourceTag: "tool",
    },
    payload: {
      executionPrompt: "请输出链路正常",
      model: "gpt-5.5",
    },
  });

  assert.equal(result.status, "completed");
  if (result.status !== "completed") {
    assert.fail("expected completed");
  }

  const lines = result.output.split("\n");
  const metadata = JSON.parse(lines[0] ?? "{}") as {
    argv: string[];
    cwd: string;
    envSource: string;
  };
  assert.deepEqual(metadata.argv, [
    "chat",
    "-q",
    "请输出链路正常",
    "-Q",
    "--source",
    "tool",
    "-m",
    "gpt-5.5",
    "-t",
    "web,terminal",
    "-s",
    "boss-dev",
  ]);
  assert.equal(metadata.cwd, expectedWorkspace);
  assert.equal(metadata.envSource, "");
  assert.equal(lines.at(-1), "Hermes smoke completed");
  assert.equal(result.sessionId, "hermes-session-123");
});

test("Hermes runner 会把非零退出码映射成 stderr 或退出码错误", async () => {
  const { scriptPath } = await createTempScript(`
    process.stderr.write("hermes crashed");
    process.exit(2);
  `);

  const result = await runHermesCommandForTesting({
    config: {
      enabled: true,
      command: process.execPath,
      args: [scriptPath],
      timeoutMs: 1000,
      sourceTag: "tool",
    },
    payload: {
      executionPrompt: "anything",
    },
  });

  assert.equal(result.status, "failed");
  if (result.status !== "failed") {
    assert.fail("expected failed");
  }

  assert.match(result.error, /hermes crashed/);
});

test("Hermes runner 在输出只有 session_id 时会视为失败", async () => {
  const { scriptPath } = await createTempScript(`
    process.stdout.write("session_id: hermes-session-123\\n");
  `);

  const result = await runHermesCommandForTesting({
    config: {
      enabled: true,
      command: process.execPath,
      args: [scriptPath],
      timeoutMs: 1000,
      sourceTag: "tool",
    },
    payload: {
      executionPrompt: "anything",
    },
  });

  assert.equal(result.status, "failed");
  if (result.status !== "failed") {
    assert.fail("expected failed");
  }

  assert.match(result.error, /EMPTY_HERMES_RESPONSE/);
});

test("Hermes runner 超时后返回 HERMES_TIMEOUT", async () => {
  const { scriptPath } = await createTempScript(`
    setTimeout(() => {
      process.stdout.write("late response\\n");
    }, 500);
  `);

  const result = await runHermesCommandForTesting({
    config: {
      enabled: true,
      command: process.execPath,
      args: [scriptPath],
      timeoutMs: 50,
      sourceTag: "tool",
    },
    payload: {
      executionPrompt: "slow",
    },
  });

  assert.equal(result.status, "failed");
  if (result.status !== "failed") {
    assert.fail("expected failed");
  }

  assert.match(result.error, /HERMES_TIMEOUT/);
});