Quickstart Guide

1import { createAgent } from 'smoke-monkey-harness';
2
3const agent = createAgent({
4  provider: 'nvidia',
5  model: 'nvidia/nemotron-3-super-120b-a12b',
6  apiKey: process.env.NVIDIA_API_KEY,
7  workspacePath: process.cwd(),
8});
9
10// Run an autonomous goal
11const result = await agent.run(
12  'Find all deprecated express endpoints, update them, and run tests.'
13);
14
15console.log('Result status:', result.status); // 'completed'

1import { createAgent } from 'smoke-monkey-harness';
2
3const agent = createAgent({
4  provider: 'gemini',
5  model: 'gemini-1.5-pro',
6  apiKey: process.env.GEMINI_API_KEY,
7  workspacePath: process.cwd(),
8  permissions: 'ask-default', // or custom policy resolver
9});
10
11// Prompt approval before running dangerous terminal commands
12agent.on('permission.required', async (event) => {
13  const { toolCallId, toolName, args } = event.data;
14  console.log(`Permission requested: ${toolName}`, args);
15
16  const approved = await myUiConfirmationModal(toolName, args);
17  agent.resolvePermission(toolCallId, approved ? 'allow' : 'deny');
18});
19
20// When the model needs clarification from the developer
21agent.on('ask_user.required', async (event) => {
22  const answer = await promptUserInChat(event.data.question);
23  agent.respond(event.data.toolCallId, answer);
24});

1import { SmokeMonkeyChat, FetchTransport } from '@smoke-monkey/ui';
2import '@smoke-monkey/ui/ui.css';
3
4// Point at your backend SSE route
5const transport = new FetchTransport({
6  url: '/api/agent-chat',
7  buildRequest: ({ messages, model }) => ({
8    headers: { 'Content-Type': 'application/json' },
9    body: JSON.stringify({ model, messages }),
10  }),
11});
12
13export function AgentWorkspace() {
14  return (
15    <div className="w-full h-screen p-4 bg-obsidian">
16      <SmokeMonkeyChat
17        transport={transport}
18        theme="dark"
19        layout="coding"
20        features={{ reasoning: true, toolCalls: true, charts: true }}
21      />
22    </div>
23  );
24}

1// Place inside ~/.claude.json or project .mcp.json:
2{
3  "mcpServers": {
4    "smoke-monkey-harness": {
5      "command": "npx",
6      "args": ["-y", "smoke-monkey-harness-mcp"]
7    }
8  }
9}
10
11// Inside Claude or Cursor prompt:
12// "Use harness_plan to design a coding assistant,
13// then scaffold it with harness_scaffold and verify with harness_verify."

1import { createAgent } from 'smoke-monkey-harness';
2
3const agent = createAgent({
4  provider: 'ollama',
5  model: 'qwen3:8b', // or llama3.1, deepseek-r1
6  baseUrl: 'http://localhost:11434/v1',
7  workspacePath: process.cwd(),
8});
9
10const result = await agent.run('Audit the dependencies in package.json.');
11console.log(result.messages.at(-1)?.content);