413 lines
12 KiB
JavaScript
413 lines
12 KiB
JavaScript
import test from "node:test";
|
|
import assert from "node:assert/strict";
|
|
import os from "node:os";
|
|
import path from "node:path";
|
|
import { mkdtemp, mkdir, rm, writeFile } from "node:fs/promises";
|
|
import { DatabaseSync } from "node:sqlite";
|
|
|
|
import {
|
|
buildCodexTaskExecution,
|
|
prepareCodexTaskExecution,
|
|
} from "../local-agent/codex-task-runner.mjs";
|
|
|
|
let runtimeRoot = "";
|
|
|
|
async function ensureRuntimeRoot() {
|
|
if (!runtimeRoot) {
|
|
runtimeRoot = await mkdtemp(path.join(os.tmpdir(), "boss-local-agent-task-runner-"));
|
|
}
|
|
return runtimeRoot;
|
|
}
|
|
|
|
test.after(async () => {
|
|
if (runtimeRoot) {
|
|
await rm(runtimeRoot, { recursive: true, force: true });
|
|
}
|
|
});
|
|
|
|
async function createCodexStateDb(threads) {
|
|
const root = await ensureRuntimeRoot();
|
|
const dbPath = path.join(root, `state-${Math.random().toString(16).slice(2)}.sqlite`);
|
|
const db = new DatabaseSync(dbPath);
|
|
db.exec(`
|
|
CREATE TABLE threads (
|
|
id TEXT PRIMARY KEY,
|
|
rollout_path TEXT NOT NULL,
|
|
created_at INTEGER NOT NULL,
|
|
updated_at INTEGER NOT NULL,
|
|
source TEXT NOT NULL,
|
|
model_provider TEXT NOT NULL,
|
|
cwd TEXT NOT NULL,
|
|
title TEXT NOT NULL,
|
|
sandbox_policy TEXT NOT NULL,
|
|
approval_mode TEXT NOT NULL,
|
|
tokens_used INTEGER NOT NULL DEFAULT 0,
|
|
has_user_event INTEGER NOT NULL DEFAULT 0,
|
|
archived INTEGER NOT NULL DEFAULT 0,
|
|
archived_at INTEGER,
|
|
git_sha TEXT,
|
|
git_branch TEXT,
|
|
git_origin_url TEXT,
|
|
cli_version TEXT NOT NULL DEFAULT '',
|
|
first_user_message TEXT NOT NULL DEFAULT '',
|
|
agent_nickname TEXT,
|
|
agent_role TEXT,
|
|
memory_mode TEXT NOT NULL DEFAULT 'enabled',
|
|
model TEXT,
|
|
reasoning_effort TEXT,
|
|
agent_path TEXT
|
|
);
|
|
`);
|
|
const insertThread = db.prepare(`
|
|
INSERT INTO threads (
|
|
id, rollout_path, created_at, updated_at, source, model_provider, cwd, title,
|
|
sandbox_policy, approval_mode, tokens_used, has_user_event, archived,
|
|
cli_version, first_user_message, agent_nickname, agent_role, memory_mode, model, reasoning_effort
|
|
) VALUES (?, ?, ?, ?, 'desktop', 'openai', ?, ?, ?, 'never', 0, 1, 0, '0.118.0', '', '', '', 'enabled', 'gpt-5.4', 'medium')
|
|
`);
|
|
for (const thread of threads) {
|
|
insertThread.run(
|
|
thread.id,
|
|
thread.rolloutPath ?? path.join(root, `${thread.id}.jsonl`),
|
|
1774845600,
|
|
1774845618,
|
|
thread.cwd,
|
|
thread.title ?? thread.id,
|
|
thread.sandboxPolicy ?? '{"type":"workspace-write"}',
|
|
);
|
|
}
|
|
db.close();
|
|
return dbPath;
|
|
}
|
|
|
|
test("conversation reply resumes the real Codex thread when thread ref is available", () => {
|
|
const execution = buildCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
masterAgentModel: "gpt-5.4",
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "请回复用户",
|
|
sourceMessageId: "msg-1",
|
|
sourceMessageBody: "请回复用户",
|
|
sourceMessageSentAt: "2026-04-21T09:00:00.000Z",
|
|
mirrorBossUserMessageToCodexDesktop: true,
|
|
targetCodexThreadRef: "019d-thread-real",
|
|
targetCodexFolderRef: "/Users/kris/code/meiyesaas",
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(execution.mode, "resume");
|
|
assert.equal(execution.cwd, "/Users/kris/code/meiyesaas");
|
|
assert.deepEqual(execution.args, [
|
|
"exec",
|
|
"resume",
|
|
"--skip-git-repo-check",
|
|
"-o",
|
|
"/tmp/reply.txt",
|
|
"-m",
|
|
"gpt-5.4",
|
|
"019d-thread-real",
|
|
"请回复用户",
|
|
]);
|
|
assert.deepEqual(execution.desktopMirror, {
|
|
enabled: true,
|
|
targetThreadRef: "019d-thread-real",
|
|
sourceMessageId: "msg-1",
|
|
sourceMessageBody: "请回复用户",
|
|
sourceMessageSentAt: "2026-04-21T09:00:00.000Z",
|
|
});
|
|
});
|
|
|
|
test("dispatch execution falls back to targetThreadId when codex thread ref is missing", () => {
|
|
const execution = buildCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
},
|
|
{
|
|
taskType: "dispatch_execution",
|
|
executionPrompt: "请执行群聊任务",
|
|
targetThreadId: "019d-thread-fallback",
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(execution.mode, "resume");
|
|
assert.deepEqual(execution.args, [
|
|
"exec",
|
|
"resume",
|
|
"--skip-git-repo-check",
|
|
"-o",
|
|
"/tmp/reply.txt",
|
|
"019d-thread-fallback",
|
|
"请执行群聊任务",
|
|
]);
|
|
});
|
|
|
|
test("master agent reply without target thread stays on ephemeral exec", () => {
|
|
const execution = buildCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
masterAgentModel: "gpt-5.4",
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "你是主 Agent",
|
|
},
|
|
"/tmp/master.txt",
|
|
);
|
|
|
|
assert.equal(execution.mode, "ephemeral");
|
|
assert.equal(execution.cwd, "/Users/kris/code/boss");
|
|
assert.deepEqual(execution.args, [
|
|
"exec",
|
|
"--ephemeral",
|
|
"--skip-git-repo-check",
|
|
"-C",
|
|
"/Users/kris/code/boss",
|
|
"-s",
|
|
"workspace-write",
|
|
"-o",
|
|
"/tmp/master.txt",
|
|
"-m",
|
|
"gpt-5.4",
|
|
"你是主 Agent",
|
|
]);
|
|
assert.deepEqual(execution.desktopMirror, { enabled: false });
|
|
});
|
|
|
|
test("relay conversation reply mirrors the clean Boss user message into the desktop child thread", () => {
|
|
const execution = buildCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
masterAgentModel: "gpt-5.4",
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "你是主 Agent",
|
|
relayViaMasterAgent: true,
|
|
sourceMessageId: "msg-relay",
|
|
sourceMessageBody: "帮我推进当前线程",
|
|
sourceMessageSentAt: "2026-04-21T09:10:00.000Z",
|
|
mirrorBossUserMessageToCodexDesktop: true,
|
|
targetCodexThreadRef: "019d-thread-real",
|
|
targetCodexFolderRef: "/Users/kris/code/meiyesaas",
|
|
},
|
|
"/tmp/master.txt",
|
|
);
|
|
|
|
assert.deepEqual(execution.desktopMirror, {
|
|
enabled: true,
|
|
targetThreadRef: "019d-thread-real",
|
|
sourceMessageId: "msg-relay",
|
|
sourceMessageBody: "帮我推进当前线程",
|
|
sourceMessageSentAt: "2026-04-21T09:10:00.000Z",
|
|
});
|
|
assert.notEqual(execution.desktopMirror.sourceMessageBody, execution.args.at(-1));
|
|
});
|
|
|
|
test("conversation reply preflight fails closed when target cwd is missing", async () => {
|
|
const missingFolder = "/tmp/boss-local-agent-missing-workdir";
|
|
const stateDbPath = await createCodexStateDb([
|
|
{
|
|
id: "019d-thread-real",
|
|
cwd: "/Users/kris/code/boss",
|
|
title: "Real thread",
|
|
},
|
|
]);
|
|
const result = await prepareCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
codexStateDbPath: stateDbPath,
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "请回复用户",
|
|
targetCodexThreadRef: "019d-thread-real",
|
|
targetCodexFolderRef: missingFolder,
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(result.ok, false);
|
|
assert.equal(result.error.code, "LOCAL_AGENT_CODEX_WORKDIR_INVALID");
|
|
assert.match(result.error.message, /LOCAL_AGENT_CODEX_WORKDIR_INVALID/);
|
|
assert.match(result.error.message, /missing-workdir/);
|
|
});
|
|
|
|
test("conversation reply preflight accepts session-only Codex threads when state db is stale", async () => {
|
|
const root = await ensureRuntimeRoot();
|
|
const validCwd = path.join(root, "session-only-project");
|
|
const sessionsDir = path.join(root, "sessions-only", "2026", "05", "02");
|
|
const threadId = "019d-session-only-task";
|
|
await mkdir(validCwd, { recursive: true });
|
|
await mkdir(sessionsDir, { recursive: true });
|
|
await writeFile(
|
|
path.join(sessionsDir, `rollout-2026-05-02T10-10-00-${threadId}.jsonl`),
|
|
`${JSON.stringify({
|
|
timestamp: "2026-05-02T02:10:00.000Z",
|
|
type: "session_meta",
|
|
payload: {
|
|
id: threadId,
|
|
cwd: validCwd,
|
|
},
|
|
})}\n`,
|
|
"utf8",
|
|
);
|
|
const stateDbPath = await createCodexStateDb([
|
|
{
|
|
id: "019d-thread-other",
|
|
cwd: validCwd,
|
|
title: "Other thread",
|
|
},
|
|
]);
|
|
|
|
const result = await prepareCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
codexStateDbPath: stateDbPath,
|
|
codexSessionsDir: path.join(root, "sessions-only"),
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "请回复用户",
|
|
targetCodexThreadRef: threadId,
|
|
targetCodexFolderRef: validCwd,
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(result.ok, true);
|
|
assert.equal(result.execution.mode, "resume");
|
|
assert.equal(result.execution.cwd, validCwd);
|
|
assert.deepEqual(result.execution.args.slice(-2), [threadId, "请回复用户"]);
|
|
});
|
|
|
|
test("dispatch execution preflight fails closed when target thread ref is missing", async () => {
|
|
const result = await prepareCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
},
|
|
{
|
|
taskType: "dispatch_execution",
|
|
executionPrompt: "请执行群聊任务",
|
|
targetCodexFolderRef: "/Users/kris/code/boss",
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(result.ok, false);
|
|
assert.equal(result.error.code, "LOCAL_AGENT_CODEX_THREAD_BINDING_MISSING");
|
|
assert.match(result.error.message, /LOCAL_AGENT_CODEX_THREAD_BINDING_MISSING/);
|
|
});
|
|
|
|
test("conversation reply preflight fails closed when target thread ref is stale in local Codex state", async () => {
|
|
const root = await ensureRuntimeRoot();
|
|
const validCwd = path.join(root, "project-stale");
|
|
await mkdir(validCwd, { recursive: true });
|
|
const stateDbPath = await createCodexStateDb([
|
|
{
|
|
id: "019d-thread-other",
|
|
cwd: validCwd,
|
|
title: "Other thread",
|
|
},
|
|
]);
|
|
|
|
const result = await prepareCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
codexStateDbPath: stateDbPath,
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "请回复用户",
|
|
targetCodexThreadRef: "019d-thread-stale",
|
|
targetCodexFolderRef: validCwd,
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(result.ok, false);
|
|
assert.equal(result.error.code, "LOCAL_AGENT_CODEX_THREAD_BINDING_STALE");
|
|
assert.match(result.error.message, /LOCAL_AGENT_CODEX_THREAD_BINDING_STALE/);
|
|
});
|
|
|
|
test("conversation reply preflight fails closed when target cwd mismatches the live Codex thread binding", async () => {
|
|
const root = await ensureRuntimeRoot();
|
|
const liveCwd = path.join(root, "project-live");
|
|
const staleCwd = path.join(root, "project-stale-mismatch");
|
|
await mkdir(liveCwd, { recursive: true });
|
|
await mkdir(staleCwd, { recursive: true });
|
|
const stateDbPath = await createCodexStateDb([
|
|
{
|
|
id: "019d-thread-live",
|
|
cwd: liveCwd,
|
|
title: "Live thread",
|
|
},
|
|
]);
|
|
|
|
const result = await prepareCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
codexStateDbPath: stateDbPath,
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "请回复用户",
|
|
targetCodexThreadRef: "019d-thread-live",
|
|
targetCodexFolderRef: staleCwd,
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(result.ok, false);
|
|
assert.equal(result.error.code, "LOCAL_AGENT_CODEX_THREAD_BINDING_MISMATCH");
|
|
assert.match(result.error.message, /LOCAL_AGENT_CODEX_THREAD_BINDING_MISMATCH/);
|
|
assert.match(result.error.message, /project-live/);
|
|
});
|
|
|
|
test("conversation reply preflight fails closed when target Codex thread is read-only", async () => {
|
|
const root = await ensureRuntimeRoot();
|
|
const readonlyCwd = path.join(root, "project-readonly");
|
|
await mkdir(readonlyCwd, { recursive: true });
|
|
const stateDbPath = await createCodexStateDb([
|
|
{
|
|
id: "019d-thread-readonly",
|
|
cwd: readonlyCwd,
|
|
title: "Read-only thread",
|
|
sandboxPolicy: '{"type":"read-only"}',
|
|
},
|
|
]);
|
|
|
|
const result = await prepareCodexTaskExecution(
|
|
{
|
|
masterAgentWorkdir: "/Users/kris/code/boss",
|
|
masterAgentSandbox: "workspace-write",
|
|
codexStateDbPath: stateDbPath,
|
|
},
|
|
{
|
|
taskType: "conversation_reply",
|
|
executionPrompt: "请继续开发",
|
|
targetCodexThreadRef: "019d-thread-readonly",
|
|
targetCodexFolderRef: readonlyCwd,
|
|
},
|
|
"/tmp/reply.txt",
|
|
);
|
|
|
|
assert.equal(result.ok, false);
|
|
assert.equal(result.error.code, "LOCAL_AGENT_CODEX_THREAD_READ_ONLY");
|
|
assert.match(result.error.message, /LOCAL_AGENT_CODEX_THREAD_READ_ONLY/);
|
|
assert.match(result.error.message, /read-only/);
|
|
});
|